TLA Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * heapam.c
4 : * heap access method code
5 : *
6 : * Portions Copyright (c) 1996-2023, 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/bufmask.h"
35 : #include "access/genam.h"
36 : #include "access/heapam.h"
37 : #include "access/heapam_xlog.h"
38 : #include "access/heaptoast.h"
39 : #include "access/hio.h"
40 : #include "access/multixact.h"
41 : #include "access/parallel.h"
42 : #include "access/relscan.h"
43 : #include "access/subtrans.h"
44 : #include "access/syncscan.h"
45 : #include "access/sysattr.h"
46 : #include "access/tableam.h"
47 : #include "access/transam.h"
48 : #include "access/valid.h"
49 : #include "access/visibilitymap.h"
50 : #include "access/xact.h"
51 : #include "access/xlog.h"
52 : #include "access/xloginsert.h"
53 : #include "access/xlogutils.h"
54 : #include "catalog/catalog.h"
55 : #include "commands/vacuum.h"
56 : #include "miscadmin.h"
57 : #include "pgstat.h"
58 : #include "port/atomics.h"
59 : #include "port/pg_bitutils.h"
60 : #include "storage/bufmgr.h"
61 : #include "storage/freespace.h"
62 : #include "storage/lmgr.h"
63 : #include "storage/predicate.h"
64 : #include "storage/procarray.h"
65 : #include "storage/smgr.h"
66 : #include "storage/spin.h"
67 : #include "storage/standby.h"
68 : #include "utils/datum.h"
69 : #include "utils/inval.h"
70 : #include "utils/lsyscache.h"
71 : #include "utils/relcache.h"
72 : #include "utils/snapmgr.h"
73 : #include "utils/spccache.h"
74 :
75 :
76 : static HeapTuple heap_prepare_insert(Relation relation, HeapTuple tup,
77 : TransactionId xid, CommandId cid, int options);
78 : static XLogRecPtr log_heap_update(Relation reln, Buffer oldbuf,
79 : Buffer newbuf, HeapTuple oldtup,
80 : HeapTuple newtup, HeapTuple old_key_tuple,
81 : bool all_visible_cleared, bool new_all_visible_cleared);
82 : static Bitmapset *HeapDetermineColumnsInfo(Relation relation,
83 : Bitmapset *interesting_cols,
84 : Bitmapset *external_cols,
85 : HeapTuple oldtup, HeapTuple newtup,
86 : bool *has_external);
87 : static bool heap_acquire_tuplock(Relation relation, ItemPointer tid,
88 : LockTupleMode mode, LockWaitPolicy wait_policy,
89 : bool *have_tuple_lock);
90 : static void compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
91 : uint16 old_infomask2, TransactionId add_to_xmax,
92 : LockTupleMode mode, bool is_update,
93 : TransactionId *result_xmax, uint16 *result_infomask,
94 : uint16 *result_infomask2);
95 : static TM_Result heap_lock_updated_tuple(Relation rel, HeapTuple tuple,
96 : ItemPointer ctid, TransactionId xid,
97 : LockTupleMode mode);
98 : static int heap_log_freeze_plan(HeapTupleFreeze *tuples, int ntuples,
99 : xl_heap_freeze_plan *plans_out,
100 : OffsetNumber *offsets_out);
101 : static void GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
102 : uint16 *new_infomask2);
103 : static TransactionId MultiXactIdGetUpdateXid(TransactionId xmax,
104 : uint16 t_infomask);
105 : static bool DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
106 : LockTupleMode lockmode, bool *current_is_member);
107 : static void MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
108 : Relation rel, ItemPointer ctid, XLTW_Oper oper,
109 : int *remaining);
110 : static bool ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
111 : uint16 infomask, Relation rel, int *remaining);
112 : static void index_delete_sort(TM_IndexDeleteOp *delstate);
113 : static int bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate);
114 : static XLogRecPtr log_heap_new_cid(Relation relation, HeapTuple tup);
115 : static HeapTuple ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
116 : bool *copy);
117 :
118 :
119 : /*
120 : * Each tuple lock mode has a corresponding heavyweight lock, and one or two
121 : * corresponding MultiXactStatuses (one to merely lock tuples, another one to
122 : * update them). This table (and the macros below) helps us determine the
123 : * heavyweight lock mode and MultiXactStatus values to use for any particular
124 : * tuple lock strength.
125 : *
126 : * Don't look at lockstatus/updstatus directly! Use get_mxact_status_for_lock
127 : * instead.
128 : */
129 : static const struct
130 : {
131 : LOCKMODE hwlock;
132 : int lockstatus;
133 : int updstatus;
134 : }
135 :
136 : tupleLockExtraInfo[MaxLockTupleMode + 1] =
137 : {
138 : { /* LockTupleKeyShare */
139 : AccessShareLock,
140 : MultiXactStatusForKeyShare,
141 : -1 /* KeyShare does not allow updating tuples */
142 : },
143 : { /* LockTupleShare */
144 : RowShareLock,
145 : MultiXactStatusForShare,
146 : -1 /* Share does not allow updating tuples */
147 : },
148 : { /* LockTupleNoKeyExclusive */
149 : ExclusiveLock,
150 : MultiXactStatusForNoKeyUpdate,
151 : MultiXactStatusNoKeyUpdate
152 : },
153 : { /* LockTupleExclusive */
154 : AccessExclusiveLock,
155 : MultiXactStatusForUpdate,
156 : MultiXactStatusUpdate
157 : }
158 : };
159 :
160 : /* Get the LOCKMODE for a given MultiXactStatus */
161 : #define LOCKMODE_from_mxstatus(status) \
162 : (tupleLockExtraInfo[TUPLOCK_from_mxstatus((status))].hwlock)
163 :
164 : /*
165 : * Acquire heavyweight locks on tuples, using a LockTupleMode strength value.
166 : * This is more readable than having every caller translate it to lock.h's
167 : * LOCKMODE.
168 : */
169 : #define LockTupleTuplock(rel, tup, mode) \
170 : LockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
171 : #define UnlockTupleTuplock(rel, tup, mode) \
172 : UnlockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
173 : #define ConditionalLockTupleTuplock(rel, tup, mode) \
174 : ConditionalLockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
175 :
176 : #ifdef USE_PREFETCH
177 : /*
178 : * heap_index_delete_tuples and index_delete_prefetch_buffer use this
179 : * structure to coordinate prefetching activity
180 : */
181 : typedef struct
182 : {
183 : BlockNumber cur_hblkno;
184 : int next_item;
185 : int ndeltids;
186 : TM_IndexDelete *deltids;
187 : } IndexDeletePrefetchState;
188 : #endif
189 :
190 : /* heap_index_delete_tuples bottom-up index deletion costing constants */
191 : #define BOTTOMUP_MAX_NBLOCKS 6
192 : #define BOTTOMUP_TOLERANCE_NBLOCKS 3
193 :
194 : /*
195 : * heap_index_delete_tuples uses this when determining which heap blocks it
196 : * must visit to help its bottom-up index deletion caller
197 : */
198 : typedef struct IndexDeleteCounts
199 : {
200 : int16 npromisingtids; /* Number of "promising" TIDs in group */
201 : int16 ntids; /* Number of TIDs in group */
202 : int16 ifirsttid; /* Offset to group's first deltid */
203 : } IndexDeleteCounts;
204 :
205 : /*
206 : * This table maps tuple lock strength values for each particular
207 : * MultiXactStatus value.
208 : */
209 : static const int MultiXactStatusLock[MaxMultiXactStatus + 1] =
210 : {
211 : LockTupleKeyShare, /* ForKeyShare */
212 : LockTupleShare, /* ForShare */
213 : LockTupleNoKeyExclusive, /* ForNoKeyUpdate */
214 : LockTupleExclusive, /* ForUpdate */
215 : LockTupleNoKeyExclusive, /* NoKeyUpdate */
216 : LockTupleExclusive /* Update */
217 : };
218 :
219 : /* Get the LockTupleMode for a given MultiXactStatus */
220 : #define TUPLOCK_from_mxstatus(status) \
221 : (MultiXactStatusLock[(status)])
222 :
223 : /* ----------------------------------------------------------------
224 : * heap support routines
225 : * ----------------------------------------------------------------
226 : */
227 :
228 : /* ----------------
229 : * initscan - scan code common to heap_beginscan and heap_rescan
230 : * ----------------
231 : */
232 : static void
233 GIC 1218368 : initscan(HeapScanDesc scan, ScanKey key, bool keep_startblock)
234 : {
235 1218368 : ParallelBlockTableScanDesc bpscan = NULL;
236 : bool allow_strat;
237 ECB : bool allow_sync;
238 :
239 : /*
240 : * Determine the number of blocks we have to scan.
241 : *
242 : * It is sufficient to do this once at scan start, since any tuples added
243 : * while the scan is in progress will be invisible to my snapshot anyway.
244 : * (That is not true when using a non-MVCC snapshot. However, we couldn't
245 : * guarantee to return tuples added after scan start anyway, since they
246 : * might go into pages we already scanned. To guarantee consistent
247 : * results for a non-MVCC snapshot, the caller must hold some higher-level
248 : * lock that ensures the interesting tuple(s) won't change.)
249 : */
250 GIC 1218368 : if (scan->rs_base.rs_parallel != NULL)
251 : {
252 1949 : bpscan = (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel;
253 1949 : scan->rs_nblocks = bpscan->phs_nblocks;
254 ECB : }
255 : else
256 CBC 1216419 : scan->rs_nblocks = RelationGetNumberOfBlocks(scan->rs_base.rs_rd);
257 ECB :
258 : /*
259 : * If the table is large relative to NBuffers, use a bulk-read access
260 : * strategy and enable synchronized scanning (see syncscan.c). Although
261 : * the thresholds for these features could be different, we make them the
262 : * same so that there are only two behaviors to tune rather than four.
263 : * (However, some callers need to be able to disable one or both of these
264 : * behaviors, independently of the size of the table; also there is a GUC
265 : * variable that can disable synchronized scanning.)
266 : *
267 : * Note that table_block_parallelscan_initialize has a very similar test;
268 : * if you change this, consider changing that one, too.
269 : */
270 GIC 1218367 : if (!RelationUsesLocalBuffers(scan->rs_base.rs_rd) &&
271 1212618 : scan->rs_nblocks > NBuffers / 4)
272 : {
273 10597 : allow_strat = (scan->rs_base.rs_flags & SO_ALLOW_STRAT) != 0;
274 CBC 10597 : allow_sync = (scan->rs_base.rs_flags & SO_ALLOW_SYNC) != 0;
275 ECB : }
276 : else
277 CBC 1207770 : allow_strat = allow_sync = false;
278 ECB :
279 GIC 1218367 : if (allow_strat)
280 : {
281 ECB : /* During a rescan, keep the previous strategy object. */
282 GIC 8779 : if (scan->rs_strategy == NULL)
283 CBC 8703 : scan->rs_strategy = GetAccessStrategy(BAS_BULKREAD);
284 : }
285 : else
286 ECB : {
287 CBC 1209588 : if (scan->rs_strategy != NULL)
288 UIC 0 : FreeAccessStrategy(scan->rs_strategy);
289 GIC 1209588 : scan->rs_strategy = NULL;
290 : }
291 ECB :
292 GBC 1218367 : if (scan->rs_base.rs_parallel != NULL)
293 ECB : {
294 : /* For parallel scan, believe whatever ParallelTableScanDesc says. */
295 GIC 1949 : if (scan->rs_base.rs_parallel->phs_syncscan)
296 CBC 2 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
297 : else
298 GIC 1947 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
299 ECB : }
300 CBC 1216418 : else if (keep_startblock)
301 : {
302 ECB : /*
303 : * When rescanning, we want to keep the previous startblock setting,
304 : * so that rewinding a cursor doesn't generate surprising results.
305 : * Reset the active syncscan setting, though.
306 : */
307 GIC 380886 : if (allow_sync && synchronize_seqscans)
308 UIC 0 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
309 : else
310 GIC 380886 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
311 ECB : }
312 GBC 835532 : else if (allow_sync && synchronize_seqscans)
313 : {
314 CBC 63 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
315 GIC 63 : scan->rs_startblock = ss_get_location(scan->rs_base.rs_rd, scan->rs_nblocks);
316 ECB : }
317 : else
318 : {
319 CBC 835469 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
320 GIC 835469 : scan->rs_startblock = 0;
321 : }
322 :
323 CBC 1218367 : scan->rs_numblocks = InvalidBlockNumber;
324 1218367 : scan->rs_inited = false;
325 GIC 1218367 : scan->rs_ctup.t_data = NULL;
326 1218367 : ItemPointerSetInvalid(&scan->rs_ctup.t_self);
327 CBC 1218367 : scan->rs_cbuf = InvalidBuffer;
328 1218367 : scan->rs_cblock = InvalidBlockNumber;
329 ECB :
330 : /* page-at-a-time fields are always invalid when not rs_inited */
331 :
332 : /*
333 : * copy the scan key, if appropriate
334 : */
335 GIC 1218367 : if (key != NULL && scan->rs_base.rs_nkeys > 0)
336 643366 : memcpy(scan->rs_base.rs_key, key, scan->rs_base.rs_nkeys * sizeof(ScanKeyData));
337 :
338 : /*
339 ECB : * Currently, we only have a stats counter for sequential heap scans (but
340 : * e.g for bitmap scans the underlying bitmap index scans will be counted,
341 : * and for sample scans we update stats for tuple fetches).
342 : */
343 GIC 1218367 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN)
344 1181294 : pgstat_count_heap_scan(scan->rs_base.rs_rd);
345 1218367 : }
346 :
347 ECB : /*
348 : * heap_setscanlimits - restrict range of a heapscan
349 : *
350 : * startBlk is the page to start at
351 : * numBlks is number of pages to scan (InvalidBlockNumber means "all")
352 : */
353 : void
354 GIC 1821 : heap_setscanlimits(TableScanDesc sscan, BlockNumber startBlk, BlockNumber numBlks)
355 : {
356 1821 : HeapScanDesc scan = (HeapScanDesc) sscan;
357 :
358 CBC 1821 : Assert(!scan->rs_inited); /* else too late to change */
359 : /* else rs_startblock is significant */
360 1821 : Assert(!(scan->rs_base.rs_flags & SO_ALLOW_SYNC));
361 :
362 ECB : /* Check startBlk is valid (but allow case of zero blocks...) */
363 GIC 1821 : Assert(startBlk == 0 || startBlk < scan->rs_nblocks);
364 ECB :
365 GIC 1821 : scan->rs_startblock = startBlk;
366 1821 : scan->rs_numblocks = numBlks;
367 CBC 1821 : }
368 :
369 ECB : /*
370 : * heapgetpage - subroutine for heapgettup()
371 : *
372 : * This routine reads and pins the specified page of the relation.
373 : * In page-at-a-time mode it performs additional work, namely determining
374 : * which tuples on the page are visible.
375 : */
376 : void
377 GNC 4549768 : heapgetpage(TableScanDesc sscan, BlockNumber block)
378 : {
379 GIC 4549768 : HeapScanDesc scan = (HeapScanDesc) sscan;
380 : Buffer buffer;
381 ECB : Snapshot snapshot;
382 : Page page;
383 : int lines;
384 : int ntup;
385 : OffsetNumber lineoff;
386 : bool all_visible;
387 :
388 GNC 4549768 : Assert(block < scan->rs_nblocks);
389 :
390 : /* release previous scan buffer, if any */
391 CBC 4549768 : if (BufferIsValid(scan->rs_cbuf))
392 : {
393 GIC 3660198 : ReleaseBuffer(scan->rs_cbuf);
394 CBC 3660198 : scan->rs_cbuf = InvalidBuffer;
395 : }
396 ECB :
397 : /*
398 : * Be sure to check for interrupts at least once per page. Checks at
399 : * higher code levels won't be able to stop a seqscan that encounters many
400 : * pages' worth of consecutive dead tuples.
401 : */
402 GIC 4549768 : CHECK_FOR_INTERRUPTS();
403 :
404 : /* read page using selected strategy */
405 GNC 4549767 : scan->rs_cbuf = ReadBufferExtended(scan->rs_base.rs_rd, MAIN_FORKNUM, block,
406 : RBM_NORMAL, scan->rs_strategy);
407 4549767 : scan->rs_cblock = block;
408 ECB :
409 GIC 4549767 : if (!(scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE))
410 CBC 87463 : return;
411 :
412 4462304 : buffer = scan->rs_cbuf;
413 4462304 : snapshot = scan->rs_base.rs_snapshot;
414 :
415 ECB : /*
416 : * Prune and repair fragmentation for the whole page, if possible.
417 : */
418 GIC 4462304 : heap_page_prune_opt(scan->rs_base.rs_rd, buffer);
419 :
420 : /*
421 ECB : * We must hold share lock on the buffer content while examining tuple
422 : * visibility. Afterwards, however, the tuples we have found to be
423 : * visible are guaranteed good as long as we hold the buffer pin.
424 : */
425 GIC 4462304 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
426 :
427 GNC 4462304 : page = BufferGetPage(buffer);
428 4462304 : TestForOldSnapshot(snapshot, scan->rs_base.rs_rd, page);
429 4462302 : lines = PageGetMaxOffsetNumber(page);
430 CBC 4462302 : ntup = 0;
431 ECB :
432 : /*
433 : * If the all-visible flag indicates that all tuples on the page are
434 : * visible to everyone, we can skip the per-tuple visibility tests.
435 : *
436 : * Note: In hot standby, a tuple that's already visible to all
437 : * transactions on the primary might still be invisible to a read-only
438 : * transaction in the standby. We partly handle this problem by tracking
439 : * the minimum xmin of visible tuples as the cut-off XID while marking a
440 : * page all-visible on the primary and WAL log that along with the
441 : * visibility map SET operation. In hot standby, we wait for (or abort)
442 : * all transactions that can potentially may not see one or more tuples on
443 : * the page. That's how index-only scans work fine in hot standby. A
444 : * crucial difference between index-only scans and heap scans is that the
445 : * index-only scan completely relies on the visibility map where as heap
446 : * scan looks at the page-level PD_ALL_VISIBLE flag. We are not sure if
447 : * the page-level flag can be trusted in the same way, because it might
448 : * get propagated somehow without being explicitly WAL-logged, e.g. via a
449 : * full page write. Until we can prove that beyond doubt, let's check each
450 : * tuple for visibility the hard way.
451 : */
452 GNC 4462302 : all_visible = PageIsAllVisible(page) && !snapshot->takenDuringRecovery;
453 :
454 225462979 : for (lineoff = FirstOffsetNumber; lineoff <= lines; lineoff++)
455 ECB : {
456 GNC 221000685 : ItemId lpp = PageGetItemId(page, lineoff);
457 : HeapTupleData loctup;
458 : bool valid;
459 :
460 221000685 : if (!ItemIdIsNormal(lpp))
461 13109416 : continue;
462 ECB :
463 GNC 207891269 : loctup.t_tableOid = RelationGetRelid(scan->rs_base.rs_rd);
464 207891269 : loctup.t_data = (HeapTupleHeader) PageGetItem(page, lpp);
465 207891269 : loctup.t_len = ItemIdGetLength(lpp);
466 207891269 : ItemPointerSet(&(loctup.t_self), block, lineoff);
467 ECB :
468 GNC 207891269 : if (all_visible)
469 23568478 : valid = true;
470 : else
471 184322791 : valid = HeapTupleSatisfiesVisibility(&loctup, snapshot, buffer);
472 :
473 207891269 : HeapCheckForSerializableConflictOut(valid, scan->rs_base.rs_rd,
474 : &loctup, buffer, snapshot);
475 :
476 207891261 : if (valid)
477 202263456 : scan->rs_vistuples[ntup++] = lineoff;
478 ECB : }
479 :
480 GIC 4462294 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
481 :
482 CBC 4462294 : Assert(ntup <= MaxHeapTuplesPerPage);
483 GIC 4462294 : scan->rs_ntuples = ntup;
484 ECB : }
485 :
486 : /*
487 : * heapgettup_initial_block - return the first BlockNumber to scan
488 : *
489 : * Returns InvalidBlockNumber when there are no blocks to scan. This can
490 : * occur with empty tables and in parallel scans when parallel workers get all
491 : * of the pages before we can get a chance to get our first page.
492 : */
493 : static BlockNumber
494 GNC 1180761 : heapgettup_initial_block(HeapScanDesc scan, ScanDirection dir)
495 : {
496 1180761 : Assert(!scan->rs_inited);
497 :
498 : /* When there are no pages to scan, return InvalidBlockNumber */
499 1180761 : if (scan->rs_nblocks == 0 || scan->rs_numblocks == 0)
500 290783 : return InvalidBlockNumber;
501 :
502 889978 : if (ScanDirectionIsForward(dir))
503 : {
504 : /* serial scan */
505 889947 : if (scan->rs_base.rs_parallel == NULL)
506 888651 : return scan->rs_startblock;
507 : else
508 : {
509 : /* parallel scan */
510 1296 : table_block_parallelscan_startblock_init(scan->rs_base.rs_rd,
511 1296 : scan->rs_parallelworkerdata,
512 1296 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
513 :
514 : /* may return InvalidBlockNumber if there are no more blocks */
515 1296 : return table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
516 1296 : scan->rs_parallelworkerdata,
517 1296 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
518 : }
519 : }
520 : else
521 : {
522 : /* backward parallel scan not supported */
523 31 : Assert(scan->rs_base.rs_parallel == NULL);
524 :
525 : /*
526 : * Disable reporting to syncscan logic in a backwards scan; it's not
527 : * very likely anyone else is doing the same thing at the same time,
528 : * and much more likely that we'll just bollix things for forward
529 : * scanners.
530 : */
531 31 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
532 :
533 : /*
534 : * Start from last page of the scan. Ensure we take into account
535 : * rs_numblocks if it's been adjusted by heap_setscanlimits().
536 : */
537 31 : if (scan->rs_numblocks != InvalidBlockNumber)
538 3 : return (scan->rs_startblock + scan->rs_numblocks - 1) % scan->rs_nblocks;
539 :
540 28 : if (scan->rs_startblock > 0)
541 UNC 0 : return scan->rs_startblock - 1;
542 :
543 GNC 28 : return scan->rs_nblocks - 1;
544 : }
545 : }
546 :
547 :
548 : /*
549 : * heapgettup_start_page - helper function for heapgettup()
550 : *
551 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
552 : * to the number of tuples on this page. Also set *lineoff to the first
553 : * offset to scan with forward scans getting the first offset and backward
554 : * getting the final offset on the page.
555 : */
556 : static Page
557 85369 : heapgettup_start_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
558 : OffsetNumber *lineoff)
559 : {
560 : Page page;
561 :
562 85369 : Assert(scan->rs_inited);
563 85369 : Assert(BufferIsValid(scan->rs_cbuf));
564 :
565 : /* Caller is responsible for ensuring buffer is locked if needed */
566 85369 : page = BufferGetPage(scan->rs_cbuf);
567 :
568 85369 : TestForOldSnapshot(scan->rs_base.rs_snapshot, scan->rs_base.rs_rd, page);
569 :
570 85369 : *linesleft = PageGetMaxOffsetNumber(page) - FirstOffsetNumber + 1;
571 :
572 85369 : if (ScanDirectionIsForward(dir))
573 85369 : *lineoff = FirstOffsetNumber;
574 : else
575 UNC 0 : *lineoff = (OffsetNumber) (*linesleft);
576 :
577 : /* lineoff now references the physically previous or next tid */
578 GNC 85369 : return page;
579 : }
580 :
581 :
582 : /*
583 : * heapgettup_continue_page - helper function for heapgettup()
584 : *
585 : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
586 : * to the number of tuples left to scan on this page. Also set *lineoff to
587 : * the next offset to scan according to the ScanDirection in 'dir'.
588 : */
589 : static inline Page
590 7503635 : heapgettup_continue_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
591 : OffsetNumber *lineoff)
592 : {
593 : Page page;
594 :
595 7503635 : Assert(scan->rs_inited);
596 7503635 : Assert(BufferIsValid(scan->rs_cbuf));
597 :
598 : /* Caller is responsible for ensuring buffer is locked if needed */
599 7503635 : page = BufferGetPage(scan->rs_cbuf);
600 :
601 7503635 : TestForOldSnapshot(scan->rs_base.rs_snapshot, scan->rs_base.rs_rd, page);
602 :
603 7503635 : if (ScanDirectionIsForward(dir))
604 : {
605 7503635 : *lineoff = OffsetNumberNext(scan->rs_coffset);
606 7503635 : *linesleft = PageGetMaxOffsetNumber(page) - (*lineoff) + 1;
607 : }
608 : else
609 : {
610 : /*
611 : * The previous returned tuple may have been vacuumed since the
612 : * previous scan when we use a non-MVCC snapshot, so we must
613 : * re-establish the lineoff <= PageGetMaxOffsetNumber(page) invariant
614 : */
615 UNC 0 : *lineoff = Min(PageGetMaxOffsetNumber(page), OffsetNumberPrev(scan->rs_coffset));
616 0 : *linesleft = *lineoff;
617 : }
618 :
619 : /* lineoff now references the physically previous or next tid */
620 GNC 7503635 : return page;
621 : }
622 :
623 : /*
624 : * heapgettup_advance_block - helper for heapgettup() and heapgettup_pagemode()
625 : *
626 : * Given the current block number, the scan direction, and various information
627 : * contained in the scan descriptor, calculate the BlockNumber to scan next
628 : * and return it. If there are no further blocks to scan, return
629 : * InvalidBlockNumber to indicate this fact to the caller.
630 : *
631 : * This should not be called to determine the initial block number -- only for
632 : * subsequent blocks.
633 : *
634 : * This also adjusts rs_numblocks when a limit has been imposed by
635 : * heap_setscanlimits().
636 : */
637 : static inline BlockNumber
638 4091735 : heapgettup_advance_block(HeapScanDesc scan, BlockNumber block, ScanDirection dir)
639 : {
640 4091735 : if (ScanDirectionIsForward(dir))
641 : {
642 4091690 : if (scan->rs_base.rs_parallel == NULL)
643 : {
644 3995847 : block++;
645 :
646 : /* wrap back to the start of the heap */
647 3995847 : if (block >= scan->rs_nblocks)
648 435528 : block = 0;
649 :
650 : /* we're done if we're back at where we started */
651 3995847 : if (block == scan->rs_startblock)
652 435487 : return InvalidBlockNumber;
653 :
654 : /* check if the limit imposed by heap_setscanlimits() is met */
655 3560360 : if (scan->rs_numblocks != InvalidBlockNumber)
656 : {
657 1551 : if (--scan->rs_numblocks == 0)
658 1487 : return InvalidBlockNumber;
659 : }
660 :
661 : /*
662 : * Report our new scan position for synchronization purposes. We
663 : * don't do that when moving backwards, however. That would just
664 : * mess up any other forward-moving scanners.
665 : *
666 : * Note: we do this before checking for end of scan so that the
667 : * final state of the position hint is back at the start of the
668 : * rel. That's not strictly necessary, but otherwise when you run
669 : * the same query multiple times the starting position would shift
670 : * a little bit backwards on every invocation, which is confusing.
671 : * We don't guarantee any specific ordering in general, though.
672 : */
673 3558873 : if (scan->rs_base.rs_flags & SO_ALLOW_SYNC)
674 13162 : ss_report_location(scan->rs_base.rs_rd, block);
675 :
676 3558873 : return block;
677 : }
678 : else
679 : {
680 95843 : return table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
681 95843 : scan->rs_parallelworkerdata, (ParallelBlockTableScanDesc)
682 95843 : scan->rs_base.rs_parallel);
683 : }
684 : }
685 : else
686 : {
687 : /* we're done if the last block is the start position */
688 45 : if (block == scan->rs_startblock)
689 45 : return InvalidBlockNumber;
690 :
691 : /* check if the limit imposed by heap_setscanlimits() is met */
692 UNC 0 : if (scan->rs_numblocks != InvalidBlockNumber)
693 : {
694 0 : if (--scan->rs_numblocks == 0)
695 0 : return InvalidBlockNumber;
696 : }
697 :
698 : /* wrap to the end of the heap when the last page was page 0 */
699 0 : if (block == 0)
700 0 : block = scan->rs_nblocks;
701 :
702 0 : block--;
703 :
704 0 : return block;
705 : }
706 : }
707 :
708 : /* ----------------
709 : * heapgettup - fetch next heap tuple
710 : *
711 : * Initialize the scan if not already done; then advance to the next
712 : * tuple as indicated by "dir"; return the next tuple in scan->rs_ctup,
713 : * or set scan->rs_ctup.t_data = NULL if no more tuples.
714 : *
715 ECB : * Note: the reason nkeys/key are passed separately, even though they are
716 : * kept in the scan descriptor, is that the caller may not want us to check
717 : * the scankeys.
718 : *
719 : * Note: when we fall off the end of the scan in either direction, we
720 : * reset rs_inited. This means that a further request with the same
721 : * scan direction will restart the scan, which is a bit odd, but a
722 : * request with the opposite scan direction will start a fresh scan
723 : * in the proper direction. The latter is required behavior for cursors,
724 : * while the former case is generally undefined behavior in Postgres
725 : * so we don't care too much.
726 : * ----------------
727 : */
728 : static void
729 GIC 7521441 : heapgettup(HeapScanDesc scan,
730 : ScanDirection dir,
731 ECB : int nkeys,
732 : ScanKey key)
733 : {
734 GIC 7521441 : HeapTuple tuple = &(scan->rs_ctup);
735 : BlockNumber block;
736 : Page page;
737 : OffsetNumber lineoff;
738 : int linesleft;
739 ECB :
740 GNC 7521441 : if (unlikely(!scan->rs_inited))
741 : {
742 17806 : block = heapgettup_initial_block(scan, dir);
743 : /* ensure rs_cbuf is invalid when we get InvalidBlockNumber */
744 17806 : Assert(block != InvalidBlockNumber || !BufferIsValid(scan->rs_cbuf));
745 17806 : scan->rs_inited = true;
746 : }
747 : else
748 ECB : {
749 : /* continue from previously returned page/tuple */
750 GNC 7503635 : block = scan->rs_cblock;
751 :
752 7503635 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
753 7503635 : page = heapgettup_continue_page(scan, dir, &linesleft, &lineoff);
754 7503635 : goto continue_page;
755 ECB : }
756 :
757 : /*
758 : * advance the scan until we find a qualifying tuple or run out of stuff
759 : * to scan
760 : */
761 GNC 103028 : while (block != InvalidBlockNumber)
762 ECB : {
763 GNC 85369 : heapgetpage((TableScanDesc) scan, block);
764 85369 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
765 85369 : page = heapgettup_start_page(scan, dir, &linesleft, &lineoff);
766 7589004 : continue_page:
767 :
768 : /*
769 : * Only continue scanning the page while we have lines left.
770 EUB : *
771 : * Note that this protects us from accessing line pointers past
772 : * PageGetMaxOffsetNumber(); both for forward scans when we resume the
773 : * table scan, and for when we start scanning a new page.
774 : */
775 GNC 7623853 : for (; linesleft > 0; linesleft--, lineoff += dir)
776 : {
777 : bool visible;
778 7538631 : ItemId lpp = PageGetItemId(page, lineoff);
779 EUB :
780 GNC 7538631 : if (!ItemIdIsNormal(lpp))
781 29667 : continue;
782 :
783 7508964 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
784 7508964 : tuple->t_len = ItemIdGetLength(lpp);
785 7508964 : ItemPointerSet(&(tuple->t_self), block, lineoff);
786 :
787 7508964 : visible = HeapTupleSatisfiesVisibility(tuple,
788 : scan->rs_base.rs_snapshot,
789 : scan->rs_cbuf);
790 :
791 7508964 : HeapCheckForSerializableConflictOut(visible, scan->rs_base.rs_rd,
792 : tuple, scan->rs_cbuf,
793 : scan->rs_base.rs_snapshot);
794 :
795 : /* skip tuples not visible to this snapshot */
796 7508964 : if (!visible)
797 5182 : continue;
798 :
799 : /* skip any tuples that don't match the scan key */
800 7503782 : if (key != NULL &&
801 UNC 0 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
802 : nkeys, key))
803 0 : continue;
804 :
805 GNC 7503782 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
806 7503782 : scan->rs_coffset = lineoff;
807 7503782 : return;
808 : }
809 ECB :
810 : /*
811 : * if we get here, it means we've exhausted the items on this page and
812 : * it's time to move to the next.
813 : */
814 GIC 85222 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
815 ECB :
816 : /* get the BlockNumber to scan next */
817 GNC 85222 : block = heapgettup_advance_block(scan, block, dir);
818 : }
819 :
820 : /* end of scan */
821 17659 : if (BufferIsValid(scan->rs_cbuf))
822 3811 : ReleaseBuffer(scan->rs_cbuf);
823 :
824 17659 : scan->rs_cbuf = InvalidBuffer;
825 17659 : scan->rs_cblock = InvalidBlockNumber;
826 17659 : tuple->t_data = NULL;
827 17659 : scan->rs_inited = false;
828 : }
829 :
830 : /* ----------------
831 : * heapgettup_pagemode - fetch next heap tuple in page-at-a-time mode
832 : *
833 : * Same API as heapgettup, but used in page-at-a-time mode
834 : *
835 : * The internal logic is much the same as heapgettup's too, but there are some
836 : * differences: we do not take the buffer content lock (that only needs to
837 : * happen inside heapgetpage), and we iterate through just the tuples listed
838 : * in rs_vistuples[] rather than all tuples on the page. Notice that
839 : * lineindex is 0-based, where the corresponding loop variable lineoff in
840 : * heapgettup is 1-based.
841 : * ----------------
842 ECB : */
843 : static void
844 GIC 49060348 : heapgettup_pagemode(HeapScanDesc scan,
845 : ScanDirection dir,
846 : int nkeys,
847 ECB : ScanKey key)
848 : {
849 GIC 49060348 : HeapTuple tuple = &(scan->rs_ctup);
850 : BlockNumber block;
851 : Page page;
852 ECB : int lineindex;
853 : int linesleft;
854 :
855 GNC 49060348 : if (unlikely(!scan->rs_inited))
856 ECB : {
857 GNC 1162955 : block = heapgettup_initial_block(scan, dir);
858 : /* ensure rs_cbuf is invalid when we get InvalidBlockNumber */
859 1162955 : Assert(block != InvalidBlockNumber || !BufferIsValid(scan->rs_cbuf));
860 1162955 : scan->rs_inited = true;
861 : }
862 : else
863 : {
864 : /* continue from previously returned page/tuple */
865 47897393 : block = scan->rs_cblock; /* current page */
866 47897393 : page = BufferGetPage(scan->rs_cbuf);
867 47897393 : TestForOldSnapshot(scan->rs_base.rs_snapshot, scan->rs_base.rs_rd, page);
868 :
869 47897393 : lineindex = scan->rs_cindex + dir;
870 47897393 : if (ScanDirectionIsForward(dir))
871 47897065 : linesleft = scan->rs_ntuples - lineindex;
872 : else
873 328 : linesleft = scan->rs_cindex;
874 : /* lineindex now references the next or previous visible tid */
875 :
876 47897393 : goto continue_page;
877 : }
878 :
879 ECB : /*
880 : * advance the scan until we find a qualifying tuple or run out of stuff
881 : * to scan
882 : */
883 GNC 5169468 : while (block != InvalidBlockNumber)
884 ECB : {
885 GNC 4458029 : heapgetpage((TableScanDesc) scan, block);
886 4458018 : page = BufferGetPage(scan->rs_cbuf);
887 4458018 : TestForOldSnapshot(scan->rs_base.rs_snapshot, scan->rs_base.rs_rd, page);
888 4458018 : linesleft = scan->rs_ntuples;
889 4458018 : lineindex = ScanDirectionIsForward(dir) ? 0 : linesleft - 1;
890 :
891 : /* lineindex now references the next or previous visible tid */
892 52355411 : continue_page:
893 :
894 196997813 : for (; linesleft > 0; linesleft--, lineindex += dir)
895 ECB : {
896 : ItemId lpp;
897 : OffsetNumber lineoff;
898 :
899 GIC 192991300 : lineoff = scan->rs_vistuples[lineindex];
900 GNC 192991300 : lpp = PageGetItemId(page, lineoff);
901 GIC 192991300 : Assert(ItemIdIsNormal(lpp));
902 ECB :
903 GNC 192991300 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
904 GIC 192991300 : tuple->t_len = ItemIdGetLength(lpp);
905 GNC 192991300 : ItemPointerSet(&(tuple->t_self), block, lineoff);
906 :
907 : /* skip any tuples that don't match the scan key */
908 192991300 : if (key != NULL &&
909 145561268 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
910 : nkeys, key))
911 144642402 : continue;
912 :
913 48348898 : scan->rs_cindex = lineindex;
914 GBC 48348898 : return;
915 : }
916 :
917 : /* get the BlockNumber to scan next */
918 GNC 4006513 : block = heapgettup_advance_block(scan, block, dir);
919 : }
920 :
921 : /* end of scan */
922 711439 : if (BufferIsValid(scan->rs_cbuf))
923 434021 : ReleaseBuffer(scan->rs_cbuf);
924 711439 : scan->rs_cbuf = InvalidBuffer;
925 711439 : scan->rs_cblock = InvalidBlockNumber;
926 711439 : tuple->t_data = NULL;
927 711439 : scan->rs_inited = false;
928 : }
929 :
930 :
931 : /* ----------------------------------------------------------------
932 : * heap access method interface
933 ECB : * ----------------------------------------------------------------
934 : */
935 :
936 :
937 : TableScanDesc
938 GIC 837428 : heap_beginscan(Relation relation, Snapshot snapshot,
939 ECB : int nkeys, ScanKey key,
940 : ParallelTableScanDesc parallel_scan,
941 : uint32 flags)
942 : {
943 : HeapScanDesc scan;
944 :
945 : /*
946 : * increment relation ref count while scanning relation
947 : *
948 : * This is just to make really sure the relcache entry won't go away while
949 : * the scan has a pointer to it. Caller should be holding the rel open
950 : * anyway, so this is redundant in all normal scenarios...
951 : */
952 CBC 837428 : RelationIncrementReferenceCount(relation);
953 ECB :
954 : /*
955 : * allocate and initialize scan descriptor
956 : */
957 GIC 837428 : scan = (HeapScanDesc) palloc(sizeof(HeapScanDescData));
958 :
959 837428 : scan->rs_base.rs_rd = relation;
960 837428 : scan->rs_base.rs_snapshot = snapshot;
961 CBC 837428 : scan->rs_base.rs_nkeys = nkeys;
962 GIC 837428 : scan->rs_base.rs_flags = flags;
963 CBC 837428 : scan->rs_base.rs_parallel = parallel_scan;
964 GIC 837428 : scan->rs_strategy = NULL; /* set in initscan */
965 ECB :
966 : /*
967 : * Disable page-at-a-time mode if it's not a MVCC-safe snapshot.
968 : */
969 CBC 837428 : if (!(snapshot && IsMVCCSnapshot(snapshot)))
970 GIC 42063 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
971 :
972 ECB : /*
973 : * For seqscan and sample scans in a serializable transaction, acquire a
974 : * predicate lock on the entire relation. This is required not only to
975 : * lock all the matching tuples, but also to conflict with new insertions
976 : * into the table. In an indexscan, we take page locks on the index pages
977 : * covering the range specified in the scan qual, but in a heap scan there
978 : * is nothing more fine-grained to lock. A bitmap scan is a different
979 : * story, there we have already scanned the index and locked the index
980 : * pages covering the predicate. But in that case we still have to lock
981 : * any matching heap tuples. For sample scan we could optimize the locking
982 : * to be at least page-level granularity, but we'd need to add per-tuple
983 : * locking for that.
984 : */
985 GIC 837428 : if (scan->rs_base.rs_flags & (SO_TYPE_SEQSCAN | SO_TYPE_SAMPLESCAN))
986 : {
987 : /*
988 : * Ensure a missing snapshot is noticed reliably, even if the
989 : * isolation mode means predicate locking isn't performed (and
990 ECB : * therefore the snapshot isn't used here).
991 : */
992 GIC 802194 : Assert(snapshot);
993 802194 : PredicateLockRelation(relation, snapshot);
994 : }
995 :
996 : /* we only need to set this up once */
997 837428 : scan->rs_ctup.t_tableOid = RelationGetRelid(relation);
998 ECB :
999 : /*
1000 : * Allocate memory to keep track of page allocation for parallel workers
1001 : * when doing a parallel scan.
1002 : */
1003 GIC 837428 : if (parallel_scan != NULL)
1004 1895 : scan->rs_parallelworkerdata = palloc(sizeof(ParallelBlockTableScanWorkerData));
1005 ECB : else
1006 CBC 835533 : scan->rs_parallelworkerdata = NULL;
1007 :
1008 : /*
1009 : * we do this here instead of in initscan() because heap_rescan also calls
1010 : * initscan() and we don't want to allocate memory again
1011 : */
1012 837428 : if (nkeys > 0)
1013 GIC 643366 : scan->rs_base.rs_key = (ScanKey) palloc(sizeof(ScanKeyData) * nkeys);
1014 : else
1015 CBC 194062 : scan->rs_base.rs_key = NULL;
1016 ECB :
1017 GIC 837428 : initscan(scan, key, false);
1018 :
1019 837427 : return (TableScanDesc) scan;
1020 : }
1021 :
1022 : void
1023 380940 : heap_rescan(TableScanDesc sscan, ScanKey key, bool set_params,
1024 : bool allow_strat, bool allow_sync, bool allow_pagemode)
1025 : {
1026 380940 : HeapScanDesc scan = (HeapScanDesc) sscan;
1027 :
1028 CBC 380940 : if (set_params)
1029 : {
1030 15 : if (allow_strat)
1031 15 : scan->rs_base.rs_flags |= SO_ALLOW_STRAT;
1032 : else
1033 UIC 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_STRAT;
1034 ECB :
1035 GIC 15 : if (allow_sync)
1036 6 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
1037 ECB : else
1038 CBC 9 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
1039 :
1040 GIC 15 : if (allow_pagemode && scan->rs_base.rs_snapshot &&
1041 15 : IsMVCCSnapshot(scan->rs_base.rs_snapshot))
1042 CBC 15 : scan->rs_base.rs_flags |= SO_ALLOW_PAGEMODE;
1043 : else
1044 UIC 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1045 ECB : }
1046 :
1047 : /*
1048 : * unpin scan buffers
1049 : */
1050 GIC 380940 : if (BufferIsValid(scan->rs_cbuf))
1051 2374 : ReleaseBuffer(scan->rs_cbuf);
1052 ECB :
1053 : /*
1054 : * reinitialize scan descriptor
1055 EUB : */
1056 GIC 380940 : initscan(scan, key, true);
1057 CBC 380940 : }
1058 :
1059 ECB : void
1060 CBC 836289 : heap_endscan(TableScanDesc sscan)
1061 : {
1062 GIC 836289 : HeapScanDesc scan = (HeapScanDesc) sscan;
1063 :
1064 : /* Note: no locking manipulations needed */
1065 :
1066 : /*
1067 : * unpin scan buffers
1068 : */
1069 CBC 836289 : if (BufferIsValid(scan->rs_cbuf))
1070 GIC 455999 : ReleaseBuffer(scan->rs_cbuf);
1071 ECB :
1072 : /*
1073 : * decrement relation reference count and free scan descriptor storage
1074 : */
1075 GIC 836289 : RelationDecrementReferenceCount(scan->rs_base.rs_rd);
1076 :
1077 836289 : if (scan->rs_base.rs_key)
1078 CBC 643338 : pfree(scan->rs_base.rs_key);
1079 EUB :
1080 GIC 836289 : if (scan->rs_strategy != NULL)
1081 CBC 8694 : FreeAccessStrategy(scan->rs_strategy);
1082 :
1083 GIC 836289 : if (scan->rs_parallelworkerdata != NULL)
1084 1895 : pfree(scan->rs_parallelworkerdata);
1085 :
1086 836289 : if (scan->rs_base.rs_flags & SO_TEMP_SNAPSHOT)
1087 126389 : UnregisterSnapshot(scan->rs_base.rs_snapshot);
1088 ECB :
1089 GIC 836289 : pfree(scan);
1090 CBC 836289 : }
1091 :
1092 : HeapTuple
1093 GIC 14659596 : heap_getnext(TableScanDesc sscan, ScanDirection direction)
1094 : {
1095 14659596 : HeapScanDesc scan = (HeapScanDesc) sscan;
1096 :
1097 ECB : /*
1098 : * This is still widely used directly, without going through table AM, so
1099 EUB : * add a safety check. It's possible we should, at a later point,
1100 : * downgrade this to an assert. The reason for checking the AM routine,
1101 : * rather than the AM oid, is that this allows to write regression tests
1102 ECB : * that create another AM reusing the heap handler.
1103 : */
1104 GIC 14659596 : if (unlikely(sscan->rs_rd->rd_tableam != GetHeapamTableAmRoutine()))
1105 UIC 0 : ereport(ERROR,
1106 : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1107 : errmsg_internal("only heap AM is supported")));
1108 :
1109 ECB : /*
1110 : * We don't expect direct calls to heap_getnext with valid CheckXidAlive
1111 : * for catalog or regular tables. See detailed comments in xact.c where
1112 : * these variables are declared. Normally we have such a check at tableam
1113 : * level API but this is called from many places so we need to ensure it
1114 : * here.
1115 : */
1116 GIC 14659596 : if (unlikely(TransactionIdIsValid(CheckXidAlive) && !bsysscan))
1117 UIC 0 : elog(ERROR, "unexpected heap_getnext call during logical decoding");
1118 :
1119 : /* Note: no locking manipulations needed */
1120 :
1121 GIC 14659596 : if (scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
1122 7639618 : heapgettup_pagemode(scan, direction,
1123 7639618 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1124 : else
1125 7019978 : heapgettup(scan, direction,
1126 7019978 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1127 :
1128 14659596 : if (scan->rs_ctup.t_data == NULL)
1129 92788 : return NULL;
1130 :
1131 : /*
1132 : * if we get here it means we have a new current scan tuple, so point to
1133 : * the proper return buffer and return the tuple.
1134 : */
1135 :
1136 14566808 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1137 :
1138 14566808 : return &scan->rs_ctup;
1139 : }
1140 :
1141 : bool
1142 41919130 : heap_getnextslot(TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
1143 : {
1144 41919130 : HeapScanDesc scan = (HeapScanDesc) sscan;
1145 :
1146 : /* Note: no locking manipulations needed */
1147 :
1148 41919130 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1149 CBC 41417667 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1150 : else
1151 GIC 501463 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1152 :
1153 41919119 : if (scan->rs_ctup.t_data == NULL)
1154 : {
1155 CBC 636263 : ExecClearTuple(slot);
1156 GIC 636263 : return false;
1157 : }
1158 :
1159 : /*
1160 : * if we get here it means we have a new current scan tuple, so point to
1161 : * the proper return buffer and return the tuple.
1162 : */
1163 :
1164 41282856 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1165 ECB :
1166 GIC 41282856 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
1167 : scan->rs_cbuf);
1168 41282856 : return true;
1169 : }
1170 ECB :
1171 : void
1172 CBC 89 : heap_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
1173 : ItemPointer maxtid)
1174 : {
1175 GIC 89 : HeapScanDesc scan = (HeapScanDesc) sscan;
1176 : BlockNumber startBlk;
1177 : BlockNumber numBlks;
1178 ECB : ItemPointerData highestItem;
1179 : ItemPointerData lowestItem;
1180 :
1181 EUB : /*
1182 : * For relations without any pages, we can simply leave the TID range
1183 : * unset. There will be no tuples to scan, therefore no tuples outside
1184 : * the given TID range.
1185 : */
1186 GIC 89 : if (scan->rs_nblocks == 0)
1187 24 : return;
1188 :
1189 : /*
1190 : * Set up some ItemPointers which point to the first and last possible
1191 ECB : * tuples in the heap.
1192 : */
1193 GIC 83 : ItemPointerSet(&highestItem, scan->rs_nblocks - 1, MaxOffsetNumber);
1194 83 : ItemPointerSet(&lowestItem, 0, FirstOffsetNumber);
1195 :
1196 ECB : /*
1197 : * If the given maximum TID is below the highest possible TID in the
1198 : * relation, then restrict the range to that, otherwise we scan to the end
1199 : * of the relation.
1200 : */
1201 CBC 83 : if (ItemPointerCompare(maxtid, &highestItem) < 0)
1202 66 : ItemPointerCopy(maxtid, &highestItem);
1203 :
1204 : /*
1205 : * If the given minimum TID is above the lowest possible TID in the
1206 : * relation, then restrict the range to only scan for TIDs above that.
1207 : */
1208 83 : if (ItemPointerCompare(mintid, &lowestItem) > 0)
1209 26 : ItemPointerCopy(mintid, &lowestItem);
1210 ECB :
1211 : /*
1212 : * Check for an empty range and protect from would be negative results
1213 : * from the numBlks calculation below.
1214 : */
1215 CBC 83 : if (ItemPointerCompare(&highestItem, &lowestItem) < 0)
1216 : {
1217 ECB : /* Set an empty range of blocks to scan */
1218 CBC 18 : heap_setscanlimits(sscan, 0, 0);
1219 18 : return;
1220 : }
1221 ECB :
1222 : /*
1223 : * Calculate the first block and the number of blocks we must scan. We
1224 : * could be more aggressive here and perform some more validation to try
1225 : * and further narrow the scope of blocks to scan by checking if the
1226 : * lowerItem has an offset above MaxOffsetNumber. In this case, we could
1227 : * advance startBlk by one. Likewise, if highestItem has an offset of 0
1228 : * we could scan one fewer blocks. However, such an optimization does not
1229 : * seem worth troubling over, currently.
1230 : */
1231 CBC 65 : startBlk = ItemPointerGetBlockNumberNoCheck(&lowestItem);
1232 :
1233 65 : numBlks = ItemPointerGetBlockNumberNoCheck(&highestItem) -
1234 GIC 65 : ItemPointerGetBlockNumberNoCheck(&lowestItem) + 1;
1235 :
1236 : /* Set the start block and number of blocks to scan */
1237 CBC 65 : heap_setscanlimits(sscan, startBlk, numBlks);
1238 ECB :
1239 : /* Finally, set the TID range in sscan */
1240 GIC 65 : ItemPointerCopy(&lowestItem, &sscan->rs_mintid);
1241 CBC 65 : ItemPointerCopy(&highestItem, &sscan->rs_maxtid);
1242 ECB : }
1243 :
1244 : bool
1245 GIC 2970 : heap_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
1246 ECB : TupleTableSlot *slot)
1247 : {
1248 GIC 2970 : HeapScanDesc scan = (HeapScanDesc) sscan;
1249 2970 : ItemPointer mintid = &sscan->rs_mintid;
1250 2970 : ItemPointer maxtid = &sscan->rs_maxtid;
1251 :
1252 : /* Note: no locking manipulations needed */
1253 : for (;;)
1254 : {
1255 3063 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1256 3063 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1257 : else
1258 UIC 0 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1259 :
1260 GIC 3063 : if (scan->rs_ctup.t_data == NULL)
1261 : {
1262 47 : ExecClearTuple(slot);
1263 47 : return false;
1264 : }
1265 :
1266 : /*
1267 : * heap_set_tidrange will have used heap_setscanlimits to limit the
1268 : * range of pages we scan to only ones that can contain the TID range
1269 : * we're scanning for. Here we must filter out any tuples from these
1270 ECB : * pages that are outside of that range.
1271 : */
1272 GIC 3016 : if (ItemPointerCompare(&scan->rs_ctup.t_self, mintid) < 0)
1273 : {
1274 CBC 93 : ExecClearTuple(slot);
1275 ECB :
1276 : /*
1277 : * When scanning backwards, the TIDs will be in descending order.
1278 : * Future tuples in this direction will be lower still, so we can
1279 : * just return false to indicate there will be no more tuples.
1280 : */
1281 CBC 93 : if (ScanDirectionIsBackward(direction))
1282 UIC 0 : return false;
1283 :
1284 CBC 93 : continue;
1285 ECB : }
1286 :
1287 : /*
1288 : * Likewise for the final page, we must filter out TIDs greater than
1289 : * maxtid.
1290 : */
1291 GIC 2923 : if (ItemPointerCompare(&scan->rs_ctup.t_self, maxtid) > 0)
1292 : {
1293 CBC 36 : ExecClearTuple(slot);
1294 ECB :
1295 : /*
1296 : * When scanning forward, the TIDs will be in ascending order.
1297 : * Future tuples in this direction will be higher still, so we can
1298 : * just return false to indicate there will be no more tuples.
1299 : */
1300 GIC 36 : if (ScanDirectionIsForward(direction))
1301 36 : return false;
1302 LBC 0 : continue;
1303 : }
1304 :
1305 CBC 2887 : break;
1306 : }
1307 :
1308 ECB : /*
1309 : * if we get here it means we have a new current scan tuple, so point to
1310 : * the proper return buffer and return the tuple.
1311 : */
1312 GIC 2887 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
1313 :
1314 CBC 2887 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
1315 2887 : return true;
1316 ECB : }
1317 :
1318 : /*
1319 : * heap_fetch - retrieve tuple with given tid
1320 : *
1321 : * On entry, tuple->t_self is the TID to fetch. We pin the buffer holding
1322 : * the tuple, fill in the remaining fields of *tuple, and check the tuple
1323 : * against the specified snapshot.
1324 : *
1325 : * If successful (tuple found and passes snapshot time qual), then *userbuf
1326 : * is set to the buffer holding the tuple and true is returned. The caller
1327 : * must unpin the buffer when done with the tuple.
1328 : *
1329 : * If the tuple is not found (ie, item number references a deleted slot),
1330 : * then tuple->t_data is set to NULL, *userbuf is set to InvalidBuffer,
1331 : * and false is returned.
1332 : *
1333 : * If the tuple is found but fails the time qual check, then the behavior
1334 : * depends on the keep_buf parameter. If keep_buf is false, the results
1335 : * are the same as for the tuple-not-found case. If keep_buf is true,
1336 : * then tuple->t_data and *userbuf are returned as for the success case,
1337 EUB : * and again the caller must unpin the buffer; but false is returned.
1338 : *
1339 : * heap_fetch does not follow HOT chains: only the exact TID requested will
1340 : * be fetched.
1341 : *
1342 : * It is somewhat inconsistent that we ereport() on invalid block number but
1343 ECB : * return false on invalid item number. There are a couple of reasons though.
1344 : * One is that the caller can relatively easily check the block number for
1345 : * validity, but cannot check the item number without reading the page
1346 EUB : * himself. Another is that when we are following a t_ctid link, we can be
1347 : * reasonably confident that the page number is valid (since VACUUM shouldn't
1348 : * truncate off the destination page without having killed the referencing
1349 : * tuple first), but the item number might well not be good.
1350 : */
1351 : bool
1352 GIC 196204 : heap_fetch(Relation relation,
1353 : Snapshot snapshot,
1354 : HeapTuple tuple,
1355 ECB : Buffer *userbuf,
1356 : bool keep_buf)
1357 : {
1358 CBC 196204 : ItemPointer tid = &(tuple->t_self);
1359 ECB : ItemId lp;
1360 : Buffer buffer;
1361 : Page page;
1362 : OffsetNumber offnum;
1363 : bool valid;
1364 :
1365 : /*
1366 : * Fetch and pin the appropriate page of the relation.
1367 : */
1368 CBC 196204 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
1369 ECB :
1370 : /*
1371 : * Need share lock on buffer to examine tuple commit status.
1372 : */
1373 GIC 196204 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1374 196204 : page = BufferGetPage(buffer);
1375 196204 : TestForOldSnapshot(snapshot, relation, page);
1376 :
1377 : /*
1378 : * We'd better check for out-of-range offnum in case of VACUUM since the
1379 : * TID was obtained.
1380 : */
1381 196204 : offnum = ItemPointerGetOffsetNumber(tid);
1382 CBC 196204 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1383 : {
1384 LBC 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1385 0 : ReleaseBuffer(buffer);
1386 UIC 0 : *userbuf = InvalidBuffer;
1387 LBC 0 : tuple->t_data = NULL;
1388 0 : return false;
1389 : }
1390 :
1391 : /*
1392 : * get the item line pointer corresponding to the requested tid
1393 : */
1394 GIC 196204 : lp = PageGetItemId(page, offnum);
1395 ECB :
1396 : /*
1397 : * Must check for deleted tuple.
1398 : */
1399 CBC 196204 : if (!ItemIdIsNormal(lp))
1400 ECB : {
1401 CBC 2 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1402 GIC 2 : ReleaseBuffer(buffer);
1403 2 : *userbuf = InvalidBuffer;
1404 2 : tuple->t_data = NULL;
1405 2 : return false;
1406 : }
1407 ECB :
1408 : /*
1409 : * fill in *tuple fields
1410 : */
1411 GIC 196202 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1412 196202 : tuple->t_len = ItemIdGetLength(lp);
1413 196202 : tuple->t_tableOid = RelationGetRelid(relation);
1414 :
1415 : /*
1416 : * check tuple visibility, then release lock
1417 : */
1418 196202 : valid = HeapTupleSatisfiesVisibility(tuple, snapshot, buffer);
1419 :
1420 196202 : if (valid)
1421 196162 : PredicateLockTID(relation, &(tuple->t_self), snapshot,
1422 CBC 196162 : HeapTupleHeaderGetXmin(tuple->t_data));
1423 :
1424 GIC 196202 : HeapCheckForSerializableConflictOut(valid, relation, tuple, buffer, snapshot);
1425 ECB :
1426 CBC 196202 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1427 :
1428 GIC 196202 : if (valid)
1429 : {
1430 : /*
1431 : * All checks passed, so return the tuple as valid. Caller is now
1432 : * responsible for releasing the buffer.
1433 : */
1434 196162 : *userbuf = buffer;
1435 ECB :
1436 GIC 196162 : return true;
1437 : }
1438 :
1439 : /* Tuple failed time qual, but maybe caller wants to see it anyway. */
1440 40 : if (keep_buf)
1441 27 : *userbuf = buffer;
1442 : else
1443 : {
1444 13 : ReleaseBuffer(buffer);
1445 13 : *userbuf = InvalidBuffer;
1446 CBC 13 : tuple->t_data = NULL;
1447 ECB : }
1448 :
1449 CBC 40 : return false;
1450 : }
1451 :
1452 : /*
1453 : * heap_hot_search_buffer - search HOT chain for tuple satisfying snapshot
1454 : *
1455 : * On entry, *tid is the TID of a tuple (either a simple tuple, or the root
1456 : * of a HOT chain), and buffer is the buffer holding this tuple. We search
1457 : * for the first chain member satisfying the given snapshot. If one is
1458 : * found, we update *tid to reference that tuple's offset number, and
1459 : * return true. If no match, return false without modifying *tid.
1460 ECB : *
1461 : * heapTuple is a caller-supplied buffer. When a match is found, we return
1462 : * the tuple here, in addition to updating *tid. If no match is found, the
1463 : * contents of this buffer on return are undefined.
1464 : *
1465 : * If all_dead is not NULL, we check non-visible tuples to see if they are
1466 : * globally dead; *all_dead is set true if all members of the HOT chain
1467 : * are vacuumable, false if not.
1468 : *
1469 : * Unlike heap_fetch, the caller must already have pin and (at least) share
1470 : * lock on the buffer; it is still pinned/locked at exit.
1471 : */
1472 : bool
1473 GBC 25144495 : heap_hot_search_buffer(ItemPointer tid, Relation relation, Buffer buffer,
1474 EUB : Snapshot snapshot, HeapTuple heapTuple,
1475 : bool *all_dead, bool first_call)
1476 ECB : {
1477 GNC 25144495 : Page page = BufferGetPage(buffer);
1478 GIC 25144495 : TransactionId prev_xmax = InvalidTransactionId;
1479 EUB : BlockNumber blkno;
1480 : OffsetNumber offnum;
1481 : bool at_chain_start;
1482 : bool valid;
1483 : bool skip;
1484 CBC 25144495 : GlobalVisState *vistest = NULL;
1485 ECB :
1486 : /* If this is not the first call, previous call returned a (live!) tuple */
1487 CBC 25144495 : if (all_dead)
1488 GIC 21146953 : *all_dead = first_call;
1489 :
1490 25144495 : blkno = ItemPointerGetBlockNumber(tid);
1491 25144495 : offnum = ItemPointerGetOffsetNumber(tid);
1492 25144495 : at_chain_start = first_call;
1493 CBC 25144495 : skip = !first_call;
1494 ECB :
1495 : /* XXX: we should assert that a snapshot is pushed or registered */
1496 GBC 25144495 : Assert(TransactionIdIsValid(RecentXmin));
1497 25144495 : Assert(BufferGetBlockNumber(buffer) == blkno);
1498 :
1499 : /* Scan through possible multiple members of HOT-chain */
1500 : for (;;)
1501 GIC 925026 : {
1502 : ItemId lp;
1503 :
1504 ECB : /* check for bogus TID */
1505 GNC 26069521 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1506 ECB : break;
1507 :
1508 GNC 26069521 : lp = PageGetItemId(page, offnum);
1509 :
1510 : /* check for unused, dead, or redirected items */
1511 GIC 26069521 : if (!ItemIdIsNormal(lp))
1512 ECB : {
1513 : /* We should only see a redirect at start of chain */
1514 CBC 806940 : if (ItemIdIsRedirected(lp) && at_chain_start)
1515 ECB : {
1516 : /* Follow the redirect */
1517 CBC 395211 : offnum = ItemIdGetRedirect(lp);
1518 395211 : at_chain_start = false;
1519 GIC 395211 : continue;
1520 : }
1521 ECB : /* else must be end of chain */
1522 CBC 411729 : break;
1523 ECB : }
1524 :
1525 : /*
1526 : * Update heapTuple to point to the element of the HOT chain we're
1527 : * currently investigating. Having t_self set correctly is important
1528 : * because the SSI checks and the *Satisfies routine for historical
1529 : * MVCC snapshots need the correct tid to decide about the visibility.
1530 : */
1531 GNC 25262581 : heapTuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1532 GIC 25262581 : heapTuple->t_len = ItemIdGetLength(lp);
1533 25262581 : heapTuple->t_tableOid = RelationGetRelid(relation);
1534 25262581 : ItemPointerSet(&heapTuple->t_self, blkno, offnum);
1535 :
1536 : /*
1537 : * Shouldn't see a HEAP_ONLY tuple at chain start.
1538 : */
1539 25262581 : if (at_chain_start && HeapTupleIsHeapOnly(heapTuple))
1540 UIC 0 : break;
1541 :
1542 : /*
1543 : * The xmin should match the previous xmax value, else chain is
1544 : * broken.
1545 ECB : */
1546 GIC 25792396 : if (TransactionIdIsValid(prev_xmax) &&
1547 CBC 529815 : !TransactionIdEquals(prev_xmax,
1548 ECB : HeapTupleHeaderGetXmin(heapTuple->t_data)))
1549 UIC 0 : break;
1550 ECB :
1551 : /*
1552 : * When first_call is true (and thus, skip is initially false) we'll
1553 : * return the first tuple we find. But on later passes, heapTuple
1554 : * will initially be pointing to the tuple we returned last time.
1555 : * Returning it again would be incorrect (and would loop forever), so
1556 : * we skip it and return the next match we find.
1557 : */
1558 GIC 25262581 : if (!skip)
1559 : {
1560 ECB : /* If it's visible per the snapshot, we must return it */
1561 GIC 25088789 : valid = HeapTupleSatisfiesVisibility(heapTuple, snapshot, buffer);
1562 25088789 : HeapCheckForSerializableConflictOut(valid, relation, heapTuple,
1563 : buffer, snapshot);
1564 :
1565 25088784 : if (valid)
1566 : {
1567 CBC 18390562 : ItemPointerSetOffsetNumber(tid, offnum);
1568 GIC 18390562 : PredicateLockTID(relation, &heapTuple->t_self, snapshot,
1569 18390562 : HeapTupleHeaderGetXmin(heapTuple->t_data));
1570 18390562 : if (all_dead)
1571 CBC 14678740 : *all_dead = false;
1572 18390562 : return true;
1573 ECB : }
1574 : }
1575 CBC 6872014 : skip = false;
1576 ECB :
1577 : /*
1578 : * If we can't see it, maybe no one else can either. At caller
1579 : * request, check whether all chain members are dead to all
1580 : * transactions.
1581 : *
1582 : * Note: if you change the criterion here for what is "dead", fix the
1583 : * planner's get_actual_variable_range() function to match.
1584 : */
1585 CBC 6872014 : if (all_dead && *all_dead)
1586 ECB : {
1587 CBC 6466218 : if (!vistest)
1588 6381533 : vistest = GlobalVisTestFor(relation);
1589 ECB :
1590 GIC 6466218 : if (!HeapTupleIsSurelyDead(heapTuple, vistest))
1591 6143314 : *all_dead = false;
1592 : }
1593 :
1594 : /*
1595 ECB : * Check to see if HOT chain continues past this tuple; if so fetch
1596 : * the next offnum and loop around.
1597 : */
1598 CBC 6872014 : if (HeapTupleIsHotUpdated(heapTuple))
1599 ECB : {
1600 CBC 529815 : Assert(ItemPointerGetBlockNumber(&heapTuple->t_data->t_ctid) ==
1601 : blkno);
1602 GIC 529815 : offnum = ItemPointerGetOffsetNumber(&heapTuple->t_data->t_ctid);
1603 529815 : at_chain_start = false;
1604 529815 : prev_xmax = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
1605 : }
1606 : else
1607 : break; /* end of chain */
1608 : }
1609 :
1610 6753928 : return false;
1611 : }
1612 :
1613 : /*
1614 : * heap_get_latest_tid - get the latest tid of a specified tuple
1615 : *
1616 : * Actually, this gets the latest version that is visible according to the
1617 : * scan's snapshot. Create a scan using SnapshotDirty to get the very latest,
1618 : * possibly uncommitted version.
1619 : *
1620 : * *tid is both an input and an output parameter: it is updated to
1621 : * show the latest version of the row. Note that it will not be changed
1622 ECB : * if no version of the row passes the snapshot test.
1623 : */
1624 : void
1625 CBC 147 : heap_get_latest_tid(TableScanDesc sscan,
1626 : ItemPointer tid)
1627 : {
1628 147 : Relation relation = sscan->rs_rd;
1629 147 : Snapshot snapshot = sscan->rs_snapshot;
1630 : ItemPointerData ctid;
1631 : TransactionId priorXmax;
1632 ECB :
1633 : /*
1634 : * table_tuple_get_latest_tid() verified that the passed in tid is valid.
1635 : * Assume that t_ctid links are valid however - there shouldn't be invalid
1636 : * ones in the table.
1637 : */
1638 GIC 147 : Assert(ItemPointerIsValid(tid));
1639 :
1640 : /*
1641 ECB : * Loop to chase down t_ctid links. At top of loop, ctid is the tuple we
1642 : * need to examine, and *tid is the TID we will return if ctid turns out
1643 : * to be bogus.
1644 : *
1645 : * Note that we will loop until we reach the end of the t_ctid chain.
1646 : * Depending on the snapshot passed, there might be at most one visible
1647 : * version of the row, but we don't try to optimize for that.
1648 : */
1649 GIC 147 : ctid = *tid;
1650 147 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
1651 : for (;;)
1652 45 : {
1653 : Buffer buffer;
1654 : Page page;
1655 : OffsetNumber offnum;
1656 : ItemId lp;
1657 : HeapTupleData tp;
1658 : bool valid;
1659 :
1660 : /*
1661 : * Read, pin, and lock the page.
1662 : */
1663 192 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&ctid));
1664 192 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1665 192 : page = BufferGetPage(buffer);
1666 192 : TestForOldSnapshot(snapshot, relation, page);
1667 ECB :
1668 : /*
1669 : * Check for bogus item number. This is not treated as an error
1670 : * condition because it can happen while following a t_ctid link. We
1671 : * just assume that the prior tid is OK and return it unchanged.
1672 : */
1673 CBC 192 : offnum = ItemPointerGetOffsetNumber(&ctid);
1674 GIC 192 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1675 ECB : {
1676 UIC 0 : UnlockReleaseBuffer(buffer);
1677 LBC 0 : break;
1678 ECB : }
1679 CBC 192 : lp = PageGetItemId(page, offnum);
1680 192 : if (!ItemIdIsNormal(lp))
1681 : {
1682 UIC 0 : UnlockReleaseBuffer(buffer);
1683 0 : break;
1684 : }
1685 :
1686 : /* OK to access the tuple */
1687 GIC 192 : tp.t_self = ctid;
1688 192 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
1689 192 : tp.t_len = ItemIdGetLength(lp);
1690 192 : tp.t_tableOid = RelationGetRelid(relation);
1691 :
1692 : /*
1693 : * After following a t_ctid link, we might arrive at an unrelated
1694 : * tuple. Check for XMIN match.
1695 ECB : */
1696 GIC 237 : if (TransactionIdIsValid(priorXmax) &&
1697 45 : !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(tp.t_data)))
1698 ECB : {
1699 UIC 0 : UnlockReleaseBuffer(buffer);
1700 0 : break;
1701 : }
1702 :
1703 ECB : /*
1704 : * Check tuple visibility; if visible, set it as the new result
1705 : * candidate.
1706 : */
1707 GIC 192 : valid = HeapTupleSatisfiesVisibility(&tp, snapshot, buffer);
1708 192 : HeapCheckForSerializableConflictOut(valid, relation, &tp, buffer, snapshot);
1709 192 : if (valid)
1710 135 : *tid = ctid;
1711 ECB :
1712 : /*
1713 : * If there's a valid t_ctid link, follow it, else we're done.
1714 : */
1715 GIC 273 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
1716 138 : HeapTupleHeaderIsOnlyLocked(tp.t_data) ||
1717 114 : HeapTupleHeaderIndicatesMovedPartitions(tp.t_data) ||
1718 57 : ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
1719 ECB : {
1720 CBC 147 : UnlockReleaseBuffer(buffer);
1721 GIC 147 : break;
1722 ECB : }
1723 :
1724 GIC 45 : ctid = tp.t_data->t_ctid;
1725 45 : priorXmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
1726 CBC 45 : UnlockReleaseBuffer(buffer);
1727 ECB : } /* end of loop */
1728 CBC 147 : }
1729 ECB :
1730 :
1731 : /*
1732 : * UpdateXmaxHintBits - update tuple hint bits after xmax transaction ends
1733 : *
1734 : * This is called after we have waited for the XMAX transaction to terminate.
1735 : * If the transaction aborted, we guarantee the XMAX_INVALID hint bit will
1736 : * be set on exit. If the transaction committed, we set the XMAX_COMMITTED
1737 : * hint bit if possible --- but beware that that may not yet be possible,
1738 : * if the transaction committed asynchronously.
1739 : *
1740 : * Note that if the transaction was a locker only, we set HEAP_XMAX_INVALID
1741 : * even if it commits.
1742 : *
1743 : * Hence callers should look only at XMAX_INVALID.
1744 : *
1745 : * Note this is not allowed for tuples whose xmax is a multixact.
1746 : */
1747 : static void
1748 GIC 158 : UpdateXmaxHintBits(HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
1749 ECB : {
1750 CBC 158 : Assert(TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple), xid));
1751 GIC 158 : Assert(!(tuple->t_infomask & HEAP_XMAX_IS_MULTI));
1752 ECB :
1753 CBC 158 : if (!(tuple->t_infomask & (HEAP_XMAX_COMMITTED | HEAP_XMAX_INVALID)))
1754 ECB : {
1755 GIC 288 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask) &&
1756 130 : TransactionIdDidCommit(xid))
1757 104 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_COMMITTED,
1758 : xid);
1759 : else
1760 54 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_INVALID,
1761 ECB : InvalidTransactionId);
1762 : }
1763 GIC 158 : }
1764 ECB :
1765 :
1766 : /*
1767 : * GetBulkInsertState - prepare status object for a bulk insert
1768 : */
1769 : BulkInsertState
1770 GIC 2215 : GetBulkInsertState(void)
1771 ECB : {
1772 : BulkInsertState bistate;
1773 :
1774 GIC 2215 : bistate = (BulkInsertState) palloc(sizeof(BulkInsertStateData));
1775 2215 : bistate->strategy = GetAccessStrategy(BAS_BULKWRITE);
1776 CBC 2215 : bistate->current_buf = InvalidBuffer;
1777 GNC 2215 : bistate->next_free = InvalidBlockNumber;
1778 2215 : bistate->last_free = InvalidBlockNumber;
1779 GIC 2215 : return bistate;
1780 ECB : }
1781 :
1782 : /*
1783 : * FreeBulkInsertState - clean up after finishing a bulk insert
1784 : */
1785 : void
1786 GIC 2116 : FreeBulkInsertState(BulkInsertState bistate)
1787 : {
1788 2116 : if (bistate->current_buf != InvalidBuffer)
1789 1766 : ReleaseBuffer(bistate->current_buf);
1790 CBC 2116 : FreeAccessStrategy(bistate->strategy);
1791 GIC 2116 : pfree(bistate);
1792 2116 : }
1793 ECB :
1794 : /*
1795 : * ReleaseBulkInsertStatePin - release a buffer currently held in bistate
1796 : */
1797 : void
1798 GIC 50755 : ReleaseBulkInsertStatePin(BulkInsertState bistate)
1799 ECB : {
1800 GIC 50755 : if (bistate->current_buf != InvalidBuffer)
1801 CBC 24 : ReleaseBuffer(bistate->current_buf);
1802 50755 : bistate->current_buf = InvalidBuffer;
1803 GIC 50755 : }
1804 ECB :
1805 :
1806 : /*
1807 : * heap_insert - insert tuple into a heap
1808 : *
1809 : * The new tuple is stamped with current transaction ID and the specified
1810 : * command ID.
1811 : *
1812 : * See table_tuple_insert for comments about most of the input flags, except
1813 : * that this routine directly takes a tuple rather than a slot.
1814 : *
1815 : * There's corresponding HEAP_INSERT_ options to all the TABLE_INSERT_
1816 : * options, and there additionally is HEAP_INSERT_SPECULATIVE which is used to
1817 : * implement table_tuple_insert_speculative().
1818 : *
1819 : * On return the header fields of *tup are updated to match the stored tuple;
1820 : * in particular tup->t_self receives the actual TID where the tuple was
1821 : * stored. But note that any toasting of fields within the tuple data is NOT
1822 : * reflected into *tup.
1823 EUB : */
1824 : void
1825 GIC 11835861 : heap_insert(Relation relation, HeapTuple tup, CommandId cid,
1826 : int options, BulkInsertState bistate)
1827 ECB : {
1828 CBC 11835861 : TransactionId xid = GetCurrentTransactionId();
1829 ECB : HeapTuple heaptup;
1830 : Buffer buffer;
1831 CBC 11835861 : Buffer vmbuffer = InvalidBuffer;
1832 11835861 : bool all_visible_cleared = false;
1833 :
1834 ECB : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
1835 CBC 11835861 : Assert(HeapTupleHeaderGetNatts(tup->t_data) <=
1836 ECB : RelationGetNumberOfAttributes(relation));
1837 :
1838 : /*
1839 : * Fill in tuple header fields and toast the tuple if necessary.
1840 : *
1841 : * Note: below this point, heaptup is the data we actually intend to store
1842 : * into the relation; tup is the caller's original untoasted data.
1843 : */
1844 GIC 11835861 : heaptup = heap_prepare_insert(relation, tup, xid, cid, options);
1845 :
1846 ECB : /*
1847 : * Find buffer to insert this tuple into. If the page is all visible,
1848 : * this will also pin the requisite visibility map page.
1849 : */
1850 CBC 11835861 : buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
1851 : InvalidBuffer, options, bistate,
1852 : &vmbuffer, NULL,
1853 : 0);
1854 :
1855 : /*
1856 : * We're about to do the actual insert -- but check for conflict first, to
1857 : * avoid possibly having to roll back work we've just done.
1858 : *
1859 : * This is safe without a recheck as long as there is no possibility of
1860 : * another process scanning the page between this check and the insert
1861 : * being visible to the scan (i.e., an exclusive buffer content lock is
1862 ECB : * continuously held from this point until the tuple insert is visible).
1863 : *
1864 : * For a heap insert, we only need to check for table-level SSI locks. Our
1865 : * new tuple can't possibly conflict with existing tuple locks, and heap
1866 : * page locks are only consolidated versions of tuple locks; they do not
1867 : * lock "gaps" as index page locks do. So we don't need to specify a
1868 : * buffer when making the call, which makes for a faster check.
1869 : */
1870 GIC 11835861 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
1871 ECB :
1872 : /* NO EREPORT(ERROR) from here till changes are logged */
1873 CBC 11835849 : START_CRIT_SECTION();
1874 ECB :
1875 GIC 11835849 : RelationPutHeapTuple(relation, buffer, heaptup,
1876 CBC 11835849 : (options & HEAP_INSERT_SPECULATIVE) != 0);
1877 :
1878 GIC 11835849 : if (PageIsAllVisible(BufferGetPage(buffer)))
1879 ECB : {
1880 GIC 7837 : all_visible_cleared = true;
1881 7837 : PageClearAllVisible(BufferGetPage(buffer));
1882 7837 : visibilitymap_clear(relation,
1883 7837 : ItemPointerGetBlockNumber(&(heaptup->t_self)),
1884 : vmbuffer, VISIBILITYMAP_VALID_BITS);
1885 : }
1886 :
1887 : /*
1888 : * XXX Should we set PageSetPrunable on this page ?
1889 : *
1890 : * The inserting transaction may eventually abort thus making this tuple
1891 : * DEAD and hence available for pruning. Though we don't want to optimize
1892 : * for aborts, if no other tuple in this page is UPDATEd/DELETEd, the
1893 : * aborted tuple will never be pruned until next vacuum is triggered.
1894 ECB : *
1895 : * If you do add PageSetPrunable here, add it in heap_xlog_insert too.
1896 : */
1897 :
1898 GIC 11835849 : MarkBufferDirty(buffer);
1899 :
1900 : /* XLOG stuff */
1901 11835849 : if (RelationNeedsWAL(relation))
1902 : {
1903 ECB : xl_heap_insert xlrec;
1904 : xl_heap_header xlhdr;
1905 : XLogRecPtr recptr;
1906 CBC 10832412 : Page page = BufferGetPage(buffer);
1907 10832412 : uint8 info = XLOG_HEAP_INSERT;
1908 10832412 : int bufflags = 0;
1909 ECB :
1910 : /*
1911 : * If this is a catalog, we need to transmit combo CIDs to properly
1912 : * decode, so log that as well.
1913 : */
1914 GIC 10832412 : if (RelationIsAccessibleInLogicalDecoding(relation))
1915 CBC 4420 : log_heap_new_cid(relation, heaptup);
1916 ECB :
1917 : /*
1918 : * If this is the single and first tuple on page, we can reinit the
1919 : * page instead of restoring the whole thing. Set flag, and hide
1920 : * buffer references from XLogInsert.
1921 : */
1922 GIC 11002229 : if (ItemPointerGetOffsetNumber(&(heaptup->t_self)) == FirstOffsetNumber &&
1923 169817 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
1924 : {
1925 CBC 168908 : info |= XLOG_HEAP_INIT_PAGE;
1926 168908 : bufflags |= REGBUF_WILL_INIT;
1927 ECB : }
1928 :
1929 GIC 10832412 : xlrec.offnum = ItemPointerGetOffsetNumber(&heaptup->t_self);
1930 10832412 : xlrec.flags = 0;
1931 10832412 : if (all_visible_cleared)
1932 7834 : xlrec.flags |= XLH_INSERT_ALL_VISIBLE_CLEARED;
1933 10832412 : if (options & HEAP_INSERT_SPECULATIVE)
1934 2007 : xlrec.flags |= XLH_INSERT_IS_SPECULATIVE;
1935 10832412 : Assert(ItemPointerGetBlockNumber(&heaptup->t_self) == BufferGetBlockNumber(buffer));
1936 :
1937 : /*
1938 : * For logical decoding, we need the tuple even if we're doing a full
1939 : * page write, so make sure it's included even if we take a full-page
1940 : * image. (XXX We could alternatively store a pointer into the FPW).
1941 : */
1942 10832412 : if (RelationIsLogicallyLogged(relation) &&
1943 304647 : !(options & HEAP_INSERT_NO_LOGICAL))
1944 : {
1945 304620 : xlrec.flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
1946 304620 : bufflags |= REGBUF_KEEP_DATA;
1947 :
1948 304620 : if (IsToastRelation(relation))
1949 1681 : xlrec.flags |= XLH_INSERT_ON_TOAST_RELATION;
1950 : }
1951 :
1952 10832412 : XLogBeginInsert();
1953 10832412 : XLogRegisterData((char *) &xlrec, SizeOfHeapInsert);
1954 :
1955 CBC 10832412 : xlhdr.t_infomask2 = heaptup->t_data->t_infomask2;
1956 GIC 10832412 : xlhdr.t_infomask = heaptup->t_data->t_infomask;
1957 CBC 10832412 : xlhdr.t_hoff = heaptup->t_data->t_hoff;
1958 ECB :
1959 : /*
1960 : * note we mark xlhdr as belonging to buffer; if XLogInsert decides to
1961 : * write the whole page to the xlog, we don't need to store
1962 : * xl_heap_header in the xlog.
1963 : */
1964 GIC 10832412 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
1965 CBC 10832412 : XLogRegisterBufData(0, (char *) &xlhdr, SizeOfHeapHeader);
1966 : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
1967 GIC 10832412 : XLogRegisterBufData(0,
1968 10832412 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
1969 10832412 : heaptup->t_len - SizeofHeapTupleHeader);
1970 :
1971 : /* filtering by origin on a row level is much more efficient */
1972 10832412 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
1973 :
1974 10832412 : recptr = XLogInsert(RM_HEAP_ID, info);
1975 :
1976 10832412 : PageSetLSN(page, recptr);
1977 ECB : }
1978 :
1979 CBC 11835849 : END_CRIT_SECTION();
1980 :
1981 11835849 : UnlockReleaseBuffer(buffer);
1982 GIC 11835849 : if (vmbuffer != InvalidBuffer)
1983 7946 : ReleaseBuffer(vmbuffer);
1984 ECB :
1985 : /*
1986 : * If tuple is cachable, mark it for invalidation from the caches in case
1987 : * we abort. Note it is OK to do this after releasing the buffer, because
1988 : * the heaptup data structure is all in local memory, not in the shared
1989 : * buffer.
1990 : */
1991 GIC 11835849 : CacheInvalidateHeapTuple(relation, heaptup, NULL);
1992 :
1993 ECB : /* Note: speculative insertions are counted too, even if aborted later */
1994 GIC 11835849 : pgstat_count_heap_insert(relation, 1);
1995 :
1996 : /*
1997 ECB : * If heaptup is a private copy, release it. Don't forget to copy t_self
1998 : * back to the caller's image, too.
1999 : */
2000 GIC 11835849 : if (heaptup != tup)
2001 ECB : {
2002 CBC 60551 : tup->t_self = heaptup->t_self;
2003 GIC 60551 : heap_freetuple(heaptup);
2004 : }
2005 CBC 11835849 : }
2006 :
2007 : /*
2008 : * Subroutine for heap_insert(). Prepares a tuple for insertion. This sets the
2009 : * tuple header fields and toasts the tuple if necessary. Returns a toasted
2010 : * version of the tuple if it was toasted, or the original tuple if not. Note
2011 ECB : * that in any case, the header fields are also set in the original tuple.
2012 : */
2013 : static HeapTuple
2014 GIC 14837378 : heap_prepare_insert(Relation relation, HeapTuple tup, TransactionId xid,
2015 : CommandId cid, int options)
2016 : {
2017 ECB : /*
2018 : * To allow parallel inserts, we need to ensure that they are safe to be
2019 : * performed in workers. We have the infrastructure to allow parallel
2020 : * inserts in general except for the cases where inserts generate a new
2021 : * CommandId (eg. inserts into a table having a foreign key column).
2022 : */
2023 GIC 14837378 : if (IsParallelWorker())
2024 LBC 0 : ereport(ERROR,
2025 ECB : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2026 : errmsg("cannot insert tuples in a parallel worker")));
2027 :
2028 GIC 14837378 : tup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
2029 14837378 : tup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
2030 14837378 : tup->t_data->t_infomask |= HEAP_XMAX_INVALID;
2031 14837378 : HeapTupleHeaderSetXmin(tup->t_data, xid);
2032 14837378 : if (options & HEAP_INSERT_FROZEN)
2033 CBC 100337 : HeapTupleHeaderSetXminFrozen(tup->t_data);
2034 ECB :
2035 GIC 14837378 : HeapTupleHeaderSetCmin(tup->t_data, cid);
2036 14837378 : HeapTupleHeaderSetXmax(tup->t_data, 0); /* for cleanliness */
2037 14837378 : tup->t_tableOid = RelationGetRelid(relation);
2038 :
2039 : /*
2040 : * If the new tuple is too big for storage or contains already toasted
2041 : * out-of-line attributes from some other relation, invoke the toaster.
2042 : */
2043 14837378 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
2044 CBC 108856 : relation->rd_rel->relkind != RELKIND_MATVIEW)
2045 : {
2046 ECB : /* toast table entries should never be recursively toasted */
2047 CBC 107203 : Assert(!HeapTupleHasExternal(tup));
2048 107203 : return tup;
2049 : }
2050 GIC 14730175 : else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
2051 60586 : return heap_toast_insert_or_update(relation, tup, NULL, options);
2052 ECB : else
2053 CBC 14669589 : return tup;
2054 : }
2055 :
2056 : /*
2057 : * Helper for heap_multi_insert() that computes the number of entire pages
2058 : * that inserting the remaining heaptuples requires. Used to determine how
2059 : * much the relation needs to be extended by.
2060 : */
2061 : static int
2062 GNC 827189 : heap_multi_insert_pages(HeapTuple *heaptuples, int done, int ntuples, Size saveFreeSpace)
2063 : {
2064 827189 : size_t page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2065 827189 : int npages = 1;
2066 :
2067 4585093 : for (int i = done; i < ntuples; i++)
2068 : {
2069 3757904 : size_t tup_sz = sizeof(ItemIdData) + MAXALIGN(heaptuples[i]->t_len);
2070 :
2071 3757904 : if (page_avail < tup_sz)
2072 : {
2073 18305 : npages++;
2074 18305 : page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2075 : }
2076 3757904 : page_avail -= tup_sz;
2077 : }
2078 :
2079 827189 : return npages;
2080 : }
2081 :
2082 : /*
2083 : * heap_multi_insert - insert multiple tuples into a heap
2084 : *
2085 ECB : * This is like heap_insert(), but inserts multiple tuples in one operation.
2086 : * That's faster than calling heap_insert() in a loop, because when multiple
2087 : * tuples can be inserted on a single page, we can write just a single WAL
2088 : * record covering all of them, and only need to lock/unlock the page once.
2089 : *
2090 : * Note: this leaks memory into the current memory context. You can create a
2091 : * temporary context before calling this, if that's a problem.
2092 : */
2093 : void
2094 GIC 810261 : heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
2095 ECB : CommandId cid, int options, BulkInsertState bistate)
2096 : {
2097 CBC 810261 : TransactionId xid = GetCurrentTransactionId();
2098 : HeapTuple *heaptuples;
2099 : int i;
2100 : int ndone;
2101 : PGAlignedBlock scratch;
2102 : Page page;
2103 810261 : Buffer vmbuffer = InvalidBuffer;
2104 : bool needwal;
2105 : Size saveFreeSpace;
2106 810261 : bool need_tuple_data = RelationIsLogicallyLogged(relation);
2107 810261 : bool need_cids = RelationIsAccessibleInLogicalDecoding(relation);
2108 GNC 810261 : bool starting_with_empty_page = false;
2109 810261 : int npages = 0;
2110 810261 : int npages_used = 0;
2111 :
2112 : /* currently not needed (thus unsupported) for heap_multi_insert() */
2113 810261 : Assert(!(options & HEAP_INSERT_NO_LOGICAL));
2114 :
2115 GIC 810261 : needwal = RelationNeedsWAL(relation);
2116 810261 : saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
2117 : HEAP_DEFAULT_FILLFACTOR);
2118 ECB :
2119 : /* Toast and set header data in all the slots */
2120 GIC 810261 : heaptuples = palloc(ntuples * sizeof(HeapTuple));
2121 3811778 : for (i = 0; i < ntuples; i++)
2122 ECB : {
2123 : HeapTuple tuple;
2124 :
2125 CBC 3001517 : tuple = ExecFetchSlotHeapTuple(slots[i], true, NULL);
2126 GIC 3001517 : slots[i]->tts_tableOid = RelationGetRelid(relation);
2127 CBC 3001517 : tuple->t_tableOid = slots[i]->tts_tableOid;
2128 3001517 : heaptuples[i] = heap_prepare_insert(relation, tuple, xid, cid,
2129 ECB : options);
2130 : }
2131 :
2132 : /*
2133 : * We're about to do the actual inserts -- but check for conflict first,
2134 : * to minimize the possibility of having to roll back work we've just
2135 : * done.
2136 : *
2137 : * A check here does not definitively prevent a serialization anomaly;
2138 : * that check MUST be done at least past the point of acquiring an
2139 : * exclusive buffer content lock on every buffer that will be affected,
2140 : * and MAY be done after all inserts are reflected in the buffers and
2141 : * those locks are released; otherwise there is a race condition. Since
2142 : * multiple buffers can be locked and unlocked in the loop below, and it
2143 : * would not be feasible to identify and lock all of those buffers before
2144 : * the loop, we must do a final check at the end.
2145 : *
2146 : * The check here could be omitted with no loss of correctness; it is
2147 : * present strictly as an optimization.
2148 : *
2149 : * For heap inserts, we only need to check for table-level SSI locks. Our
2150 : * new tuples can't possibly conflict with existing tuple locks, and heap
2151 : * page locks are only consolidated versions of tuple locks; they do not
2152 : * lock "gaps" as index page locks do. So we don't need to specify a
2153 : * buffer when making the call, which makes for a faster check.
2154 : */
2155 GIC 810261 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2156 ECB :
2157 CBC 810261 : ndone = 0;
2158 1647735 : while (ndone < ntuples)
2159 : {
2160 ECB : Buffer buffer;
2161 GIC 837474 : bool all_visible_cleared = false;
2162 CBC 837474 : bool all_frozen_set = false;
2163 ECB : int nthispage;
2164 :
2165 CBC 837474 : CHECK_FOR_INTERRUPTS();
2166 ECB :
2167 : /*
2168 : * Compute number of pages needed to fit the to-be-inserted tuples in
2169 : * the worst case. This will be used to determine how much to extend
2170 : * the relation by in RelationGetBufferForTuple(), if needed. If we
2171 : * filled a prior page from scratch, we can just update our last
2172 : * computation, but if we started with a partially filled page,
2173 : * recompute from scratch, the number of potentially required pages
2174 : * can vary due to tuples needing to fit onto the page, page headers
2175 : * etc.
2176 : */
2177 GNC 837474 : if (ndone == 0 || !starting_with_empty_page)
2178 : {
2179 827189 : npages = heap_multi_insert_pages(heaptuples, ndone, ntuples,
2180 : saveFreeSpace);
2181 827189 : npages_used = 0;
2182 : }
2183 : else
2184 10285 : npages_used++;
2185 :
2186 : /*
2187 : * Find buffer where at least the next tuple will fit. If the page is
2188 : * all-visible, this will also pin the requisite visibility map page.
2189 : *
2190 : * Also pin visibility map page if COPY FREEZE inserts tuples into an
2191 : * empty page. See all_frozen_set below.
2192 ECB : */
2193 CBC 837474 : buffer = RelationGetBufferForTuple(relation, heaptuples[ndone]->t_len,
2194 : InvalidBuffer, options, bistate,
2195 : &vmbuffer, NULL,
2196 : npages - npages_used);
2197 GIC 837474 : page = BufferGetPage(buffer);
2198 ECB :
2199 CBC 837474 : starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
2200 :
2201 GIC 837474 : if (starting_with_empty_page && (options & HEAP_INSERT_FROZEN))
2202 1658 : all_frozen_set = true;
2203 :
2204 : /* NO EREPORT(ERROR) from here till changes are logged */
2205 837474 : START_CRIT_SECTION();
2206 ECB :
2207 : /*
2208 : * RelationGetBufferForTuple has ensured that the first tuple fits.
2209 : * Put that on the page, and then as many other tuples as fit.
2210 : */
2211 CBC 837474 : RelationPutHeapTuple(relation, buffer, heaptuples[ndone], false);
2212 :
2213 ECB : /*
2214 : * For logical decoding we need combo CIDs to properly decode the
2215 : * catalog.
2216 : */
2217 GIC 837474 : if (needwal && need_cids)
2218 CBC 6295 : log_heap_new_cid(relation, heaptuples[ndone]);
2219 :
2220 3001517 : for (nthispage = 1; ndone + nthispage < ntuples; nthispage++)
2221 : {
2222 GIC 2191256 : HeapTuple heaptup = heaptuples[ndone + nthispage];
2223 ECB :
2224 GIC 2191256 : if (PageGetHeapFreeSpace(page) < MAXALIGN(heaptup->t_len) + saveFreeSpace)
2225 27213 : break;
2226 :
2227 2164043 : RelationPutHeapTuple(relation, buffer, heaptup, false);
2228 :
2229 ECB : /*
2230 : * For logical decoding we need combo CIDs to properly decode the
2231 : * catalog.
2232 : */
2233 GIC 2164043 : if (needwal && need_cids)
2234 5756 : log_heap_new_cid(relation, heaptup);
2235 : }
2236 :
2237 : /*
2238 : * If the page is all visible, need to clear that, unless we're only
2239 ECB : * going to add further frozen rows to it.
2240 : *
2241 : * If we're only adding already frozen rows to a previously empty
2242 : * page, mark it as all-visible.
2243 : */
2244 GIC 837474 : if (PageIsAllVisible(page) && !(options & HEAP_INSERT_FROZEN))
2245 ECB : {
2246 CBC 2492 : all_visible_cleared = true;
2247 GIC 2492 : PageClearAllVisible(page);
2248 2492 : visibilitymap_clear(relation,
2249 : BufferGetBlockNumber(buffer),
2250 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2251 : }
2252 834982 : else if (all_frozen_set)
2253 1658 : PageSetAllVisible(page);
2254 :
2255 : /*
2256 ECB : * XXX Should we set PageSetPrunable on this page ? See heap_insert()
2257 : */
2258 :
2259 GIC 837474 : MarkBufferDirty(buffer);
2260 :
2261 : /* XLOG stuff */
2262 837474 : if (needwal)
2263 : {
2264 : XLogRecPtr recptr;
2265 : xl_heap_multi_insert *xlrec;
2266 833127 : uint8 info = XLOG_HEAP2_MULTI_INSERT;
2267 : char *tupledata;
2268 : int totaldatalen;
2269 833127 : char *scratchptr = scratch.data;
2270 : bool init;
2271 833127 : int bufflags = 0;
2272 :
2273 ECB : /*
2274 : * If the page was previously empty, we can reinit the page
2275 : * instead of restoring the whole thing.
2276 : */
2277 GIC 833127 : init = starting_with_empty_page;
2278 :
2279 : /* allocate xl_heap_multi_insert struct from the scratch area */
2280 833127 : xlrec = (xl_heap_multi_insert *) scratchptr;
2281 CBC 833127 : scratchptr += SizeOfHeapMultiInsert;
2282 :
2283 ECB : /*
2284 : * Allocate offsets array. Unless we're reinitializing the page,
2285 : * in that case the tuples are stored in order starting at
2286 : * FirstOffsetNumber and we don't need to store the offsets
2287 : * explicitly.
2288 : */
2289 CBC 833127 : if (!init)
2290 GIC 800833 : scratchptr += nthispage * sizeof(OffsetNumber);
2291 ECB :
2292 : /* the rest of the scratch space is used for tuple data */
2293 GIC 833127 : tupledata = scratchptr;
2294 :
2295 : /* check that the mutually exclusive flags are not both set */
2296 833127 : Assert(!(all_visible_cleared && all_frozen_set));
2297 :
2298 833127 : xlrec->flags = 0;
2299 833127 : if (all_visible_cleared)
2300 2492 : xlrec->flags = XLH_INSERT_ALL_VISIBLE_CLEARED;
2301 833127 : if (all_frozen_set)
2302 14 : xlrec->flags = XLH_INSERT_ALL_FROZEN_SET;
2303 :
2304 CBC 833127 : xlrec->ntuples = nthispage;
2305 :
2306 ECB : /*
2307 : * Write out an xl_multi_insert_tuple and the tuple data itself
2308 : * for each tuple.
2309 : */
2310 GIC 3529283 : for (i = 0; i < nthispage; i++)
2311 : {
2312 2696156 : HeapTuple heaptup = heaptuples[ndone + i];
2313 : xl_multi_insert_tuple *tuphdr;
2314 : int datalen;
2315 :
2316 2696156 : if (!init)
2317 CBC 1884242 : xlrec->offsets[i] = ItemPointerGetOffsetNumber(&heaptup->t_self);
2318 : /* xl_multi_insert_tuple needs two-byte alignment. */
2319 GIC 2696156 : tuphdr = (xl_multi_insert_tuple *) SHORTALIGN(scratchptr);
2320 CBC 2696156 : scratchptr = ((char *) tuphdr) + SizeOfMultiInsertTuple;
2321 ECB :
2322 CBC 2696156 : tuphdr->t_infomask2 = heaptup->t_data->t_infomask2;
2323 GIC 2696156 : tuphdr->t_infomask = heaptup->t_data->t_infomask;
2324 CBC 2696156 : tuphdr->t_hoff = heaptup->t_data->t_hoff;
2325 :
2326 ECB : /* write bitmap [+ padding] [+ oid] + data */
2327 GIC 2696156 : datalen = heaptup->t_len - SizeofHeapTupleHeader;
2328 2696156 : memcpy(scratchptr,
2329 2696156 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2330 : datalen);
2331 2696156 : tuphdr->datalen = datalen;
2332 2696156 : scratchptr += datalen;
2333 : }
2334 833127 : totaldatalen = scratchptr - tupledata;
2335 833127 : Assert((scratchptr - scratch.data) < BLCKSZ);
2336 :
2337 833127 : if (need_tuple_data)
2338 167 : xlrec->flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2339 ECB :
2340 : /*
2341 : * Signal that this is the last xl_heap_multi_insert record
2342 : * emitted by this call to heap_multi_insert(). Needed for logical
2343 : * decoding so it knows when to cleanup temporary data.
2344 : */
2345 CBC 833127 : if (ndone + nthispage == ntuples)
2346 GIC 809765 : xlrec->flags |= XLH_INSERT_LAST_IN_MULTI;
2347 ECB :
2348 GIC 833127 : if (init)
2349 : {
2350 32294 : info |= XLOG_HEAP_INIT_PAGE;
2351 32294 : bufflags |= REGBUF_WILL_INIT;
2352 : }
2353 :
2354 : /*
2355 : * If we're doing logical decoding, include the new tuple data
2356 : * even if we take a full-page image of the page.
2357 : */
2358 833127 : if (need_tuple_data)
2359 167 : bufflags |= REGBUF_KEEP_DATA;
2360 :
2361 833127 : XLogBeginInsert();
2362 CBC 833127 : XLogRegisterData((char *) xlrec, tupledata - scratch.data);
2363 GIC 833127 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2364 :
2365 833127 : XLogRegisterBufData(0, tupledata, totaldatalen);
2366 :
2367 ECB : /* filtering by origin on a row level is much more efficient */
2368 GIC 833127 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2369 :
2370 833127 : recptr = XLogInsert(RM_HEAP2_ID, info);
2371 :
2372 833127 : PageSetLSN(page, recptr);
2373 ECB : }
2374 :
2375 GIC 837474 : END_CRIT_SECTION();
2376 :
2377 ECB : /*
2378 : * If we've frozen everything on the page, update the visibilitymap.
2379 : * We're already holding pin on the vmbuffer.
2380 : */
2381 CBC 837474 : if (all_frozen_set)
2382 : {
2383 1658 : Assert(PageIsAllVisible(page));
2384 GIC 1658 : Assert(visibilitymap_pin_ok(BufferGetBlockNumber(buffer), vmbuffer));
2385 :
2386 : /*
2387 : * It's fine to use InvalidTransactionId here - this is only used
2388 : * when HEAP_INSERT_FROZEN is specified, which intentionally
2389 : * violates visibility rules.
2390 ECB : */
2391 GBC 1658 : visibilitymap_set(relation, BufferGetBlockNumber(buffer), buffer,
2392 : InvalidXLogRecPtr, vmbuffer,
2393 : InvalidTransactionId,
2394 : VISIBILITYMAP_ALL_VISIBLE | VISIBILITYMAP_ALL_FROZEN);
2395 ECB : }
2396 :
2397 CBC 837474 : UnlockReleaseBuffer(buffer);
2398 GIC 837474 : ndone += nthispage;
2399 :
2400 : /*
2401 : * NB: Only release vmbuffer after inserting all tuples - it's fairly
2402 : * likely that we'll insert into subsequent heap pages that are likely
2403 : * to use the same vm page.
2404 : */
2405 ECB : }
2406 :
2407 : /* We're done with inserting all tuples, so release the last vmbuffer. */
2408 CBC 810261 : if (vmbuffer != InvalidBuffer)
2409 GIC 2597 : ReleaseBuffer(vmbuffer);
2410 ECB :
2411 : /*
2412 : * We're done with the actual inserts. Check for conflicts again, to
2413 : * ensure that all rw-conflicts in to these inserts are detected. Without
2414 : * this final check, a sequential scan of the heap may have locked the
2415 : * table after the "before" check, missing one opportunity to detect the
2416 : * conflict, and then scanned the table before the new tuples were there,
2417 : * missing the other chance to detect the conflict.
2418 : *
2419 : * For heap inserts, we only need to check for table-level SSI locks. Our
2420 : * new tuples can't possibly conflict with existing tuple locks, and heap
2421 : * page locks are only consolidated versions of tuple locks; they do not
2422 : * lock "gaps" as index page locks do. So we don't need to specify a
2423 : * buffer when making the call.
2424 : */
2425 GIC 810261 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2426 ECB :
2427 : /*
2428 EUB : * If tuples are cachable, mark them for invalidation from the caches in
2429 : * case we abort. Note it is OK to do this after releasing the buffer,
2430 : * because the heaptuples data structure is all in local memory, not in
2431 : * the shared buffer.
2432 : */
2433 CBC 810261 : if (IsCatalogRelation(relation))
2434 : {
2435 2916673 : for (i = 0; i < ntuples; i++)
2436 GIC 2107904 : CacheInvalidateHeapTuple(relation, heaptuples[i], NULL);
2437 EUB : }
2438 :
2439 : /* copy t_self fields back to the caller's slots */
2440 GIC 3811778 : for (i = 0; i < ntuples; i++)
2441 3001517 : slots[i]->tts_tid = heaptuples[i]->t_self;
2442 ECB :
2443 GIC 810261 : pgstat_count_heap_insert(relation, ntuples);
2444 810261 : }
2445 :
2446 : /*
2447 : * simple_heap_insert - insert a tuple
2448 ECB : *
2449 : * Currently, this routine differs from heap_insert only in supplying
2450 : * a default command ID and not allowing access to the speedup options.
2451 : *
2452 : * This should be used rather than using heap_insert directly in most places
2453 : * where we are modifying system catalogs.
2454 : */
2455 : void
2456 GIC 4591015 : simple_heap_insert(Relation relation, HeapTuple tup)
2457 : {
2458 4591015 : heap_insert(relation, tup, GetCurrentCommandId(true), 0, NULL);
2459 4591015 : }
2460 :
2461 : /*
2462 : * Given infomask/infomask2, compute the bits that must be saved in the
2463 : * "infobits" field of xl_heap_delete, xl_heap_update, xl_heap_lock,
2464 ECB : * xl_heap_lock_updated WAL records.
2465 : *
2466 : * See fix_infomask_from_infobits.
2467 : */
2468 : static uint8
2469 GIC 2034567 : compute_infobits(uint16 infomask, uint16 infomask2)
2470 : {
2471 ECB : return
2472 GIC 2034567 : ((infomask & HEAP_XMAX_IS_MULTI) != 0 ? XLHL_XMAX_IS_MULTI : 0) |
2473 2034567 : ((infomask & HEAP_XMAX_LOCK_ONLY) != 0 ? XLHL_XMAX_LOCK_ONLY : 0) |
2474 2034567 : ((infomask & HEAP_XMAX_EXCL_LOCK) != 0 ? XLHL_XMAX_EXCL_LOCK : 0) |
2475 : /* note we ignore HEAP_XMAX_SHR_LOCK here */
2476 4069134 : ((infomask & HEAP_XMAX_KEYSHR_LOCK) != 0 ? XLHL_XMAX_KEYSHR_LOCK : 0) |
2477 ECB : ((infomask2 & HEAP_KEYS_UPDATED) != 0 ?
2478 CBC 2034567 : XLHL_KEYS_UPDATED : 0);
2479 : }
2480 :
2481 : /*
2482 ECB : * Given two versions of the same t_infomask for a tuple, compare them and
2483 : * return whether the relevant status for a tuple Xmax has changed. This is
2484 : * used after a buffer lock has been released and reacquired: we want to ensure
2485 : * that the tuple state continues to be the same it was when we previously
2486 : * examined it.
2487 : *
2488 : * Note the Xmax field itself must be compared separately.
2489 : */
2490 : static inline bool
2491 GIC 5303 : xmax_infomask_changed(uint16 new_infomask, uint16 old_infomask)
2492 : {
2493 5303 : const uint16 interesting =
2494 : HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY | HEAP_LOCK_MASK;
2495 ECB :
2496 CBC 5303 : if ((new_infomask & interesting) != (old_infomask & interesting))
2497 14 : return true;
2498 :
2499 GBC 5289 : return false;
2500 : }
2501 :
2502 : /*
2503 : * heap_delete - delete a tuple
2504 : *
2505 : * See table_tuple_delete() for an explanation of the parameters, except that
2506 : * this routine directly takes a tuple rather than a slot.
2507 : *
2508 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
2509 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
2510 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
2511 : * generated by another transaction).
2512 ECB : */
2513 : TM_Result
2514 GIC 1420155 : heap_delete(Relation relation, ItemPointer tid,
2515 : CommandId cid, Snapshot crosscheck, bool wait,
2516 : TM_FailureData *tmfd, bool changingPart)
2517 : {
2518 ECB : TM_Result result;
2519 CBC 1420155 : TransactionId xid = GetCurrentTransactionId();
2520 : ItemId lp;
2521 ECB : HeapTupleData tp;
2522 : Page page;
2523 : BlockNumber block;
2524 : Buffer buffer;
2525 GIC 1420155 : Buffer vmbuffer = InvalidBuffer;
2526 : TransactionId new_xmax;
2527 : uint16 new_infomask,
2528 : new_infomask2;
2529 1420155 : bool have_tuple_lock = false;
2530 : bool iscombo;
2531 1420155 : bool all_visible_cleared = false;
2532 CBC 1420155 : HeapTuple old_key_tuple = NULL; /* replica identity of the tuple */
2533 1420155 : bool old_key_copied = false;
2534 ECB :
2535 GIC 1420155 : Assert(ItemPointerIsValid(tid));
2536 ECB :
2537 : /*
2538 : * Forbid this during a parallel operation, lest it allocate a combo CID.
2539 : * Other workers might need that combo CID for visibility checks, and we
2540 : * have no provision for broadcasting it to them.
2541 : */
2542 GIC 1420155 : if (IsInParallelMode())
2543 UIC 0 : ereport(ERROR,
2544 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2545 : errmsg("cannot delete tuples during a parallel operation")));
2546 ECB :
2547 CBC 1420155 : block = ItemPointerGetBlockNumber(tid);
2548 1420155 : buffer = ReadBuffer(relation, block);
2549 1420155 : page = BufferGetPage(buffer);
2550 ECB :
2551 : /*
2552 : * Before locking the buffer, pin the visibility map page if it appears to
2553 : * be necessary. Since we haven't got the lock yet, someone else might be
2554 : * in the middle of changing this, so we'll need to recheck after we have
2555 : * the lock.
2556 : */
2557 GIC 1420155 : if (PageIsAllVisible(page))
2558 1022 : visibilitymap_pin(relation, block, &vmbuffer);
2559 EUB :
2560 GBC 1420155 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2561 :
2562 GIC 1420155 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
2563 CBC 1420155 : Assert(ItemIdIsNormal(lp));
2564 :
2565 1420155 : tp.t_tableOid = RelationGetRelid(relation);
2566 GIC 1420155 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2567 1420155 : tp.t_len = ItemIdGetLength(lp);
2568 1420155 : tp.t_self = *tid;
2569 ECB :
2570 CBC 1 : l1:
2571 :
2572 : /*
2573 ECB : * If we didn't pin the visibility map page and the page has become all
2574 : * visible while we were busy locking the buffer, we'll have to unlock and
2575 : * re-lock, to avoid holding the buffer lock across an I/O. That's a bit
2576 : * unfortunate, but hopefully shouldn't happen often.
2577 : */
2578 CBC 1420156 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
2579 ECB : {
2580 LBC 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2581 0 : visibilitymap_pin(relation, block, &vmbuffer);
2582 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2583 EUB : }
2584 ECB :
2585 GIC 1420156 : result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
2586 :
2587 1420156 : if (result == TM_Invisible)
2588 : {
2589 UIC 0 : UnlockReleaseBuffer(buffer);
2590 0 : ereport(ERROR,
2591 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2592 : errmsg("attempted to delete invisible tuple")));
2593 : }
2594 GIC 1420156 : else if (result == TM_BeingModified && wait)
2595 : {
2596 ECB : TransactionId xwait;
2597 : uint16 infomask;
2598 :
2599 : /* must copy state data before unlocking buffer */
2600 GIC 40518 : xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
2601 40518 : infomask = tp.t_data->t_infomask;
2602 :
2603 : /*
2604 : * Sleep until concurrent transaction ends -- except when there's a
2605 ECB : * single locker and it's our own transaction. Note we don't care
2606 : * which lock mode the locker has, because we need the strongest one.
2607 : *
2608 : * Before sleeping, we need to acquire tuple lock to establish our
2609 : * priority for the tuple (see heap_lock_tuple). LockTuple will
2610 : * release us when we are next-in-line for the tuple.
2611 : *
2612 : * If we are forced to "start over" below, we keep the tuple lock;
2613 : * this arranges that we stay at the head of the line while rechecking
2614 : * tuple state.
2615 : */
2616 GIC 40518 : if (infomask & HEAP_XMAX_IS_MULTI)
2617 ECB : {
2618 CBC 8 : bool current_is_member = false;
2619 :
2620 GIC 8 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
2621 : LockTupleExclusive, ¤t_is_member))
2622 ECB : {
2623 GIC 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2624 :
2625 : /*
2626 : * Acquire the lock, if necessary (but skip it when we're
2627 : * requesting a lock and already have one; avoids deadlock).
2628 : */
2629 8 : if (!current_is_member)
2630 6 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2631 ECB : LockWaitBlock, &have_tuple_lock);
2632 :
2633 : /* wait for multixact */
2634 GIC 8 : MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
2635 ECB : relation, &(tp.t_self), XLTW_Delete,
2636 : NULL);
2637 CBC 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2638 :
2639 : /*
2640 : * If xwait had just locked the tuple then some other xact
2641 : * could update this tuple before we get to this point. Check
2642 ECB : * for xmax change, and start over if so.
2643 : *
2644 : * We also must start over if we didn't pin the VM page, and
2645 : * the page has become all visible.
2646 : */
2647 CBC 16 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
2648 8 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2649 GIC 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2650 ECB : xwait))
2651 UIC 0 : goto l1;
2652 : }
2653 ECB :
2654 : /*
2655 : * You might think the multixact is necessarily done here, but not
2656 : * so: it could have surviving members, namely our own xact or
2657 : * other subxacts of this backend. It is legal for us to delete
2658 : * the tuple in either case, however (the latter case is
2659 : * essentially a situation of upgrading our former shared lock to
2660 : * exclusive). We don't bother changing the on-disk hint bits
2661 : * since we are about to overwrite the xmax altogether.
2662 : */
2663 : }
2664 CBC 40510 : else if (!TransactionIdIsCurrentTransactionId(xwait))
2665 : {
2666 : /*
2667 : * Wait for regular transaction to end; but first, acquire tuple
2668 : * lock.
2669 : */
2670 GIC 36 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2671 36 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2672 : LockWaitBlock, &have_tuple_lock);
2673 36 : XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
2674 CBC 32 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2675 ECB :
2676 : /*
2677 : * xwait is done, but if xwait had just locked the tuple then some
2678 : * other xact could update this tuple before we get to this point.
2679 : * Check for xmax change, and start over if so.
2680 : *
2681 : * We also must start over if we didn't pin the VM page, and the
2682 : * page has become all visible.
2683 : */
2684 CBC 64 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
2685 32 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2686 GIC 31 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2687 ECB : xwait))
2688 GIC 1 : goto l1;
2689 ECB :
2690 : /* Otherwise check if it committed or aborted */
2691 GIC 31 : UpdateXmaxHintBits(tp.t_data, buffer, xwait);
2692 ECB : }
2693 :
2694 : /*
2695 : * We may overwrite if previous xmax aborted, or if it committed but
2696 : * only locked the tuple without updating it.
2697 : */
2698 CBC 40513 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
2699 GIC 40524 : HEAP_XMAX_IS_LOCKED_ONLY(tp.t_data->t_infomask) ||
2700 23 : HeapTupleHeaderIsOnlyLocked(tp.t_data))
2701 40494 : result = TM_Ok;
2702 19 : else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
2703 CBC 15 : result = TM_Updated;
2704 : else
2705 4 : result = TM_Deleted;
2706 ECB : }
2707 :
2708 GIC 1420151 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
2709 ECB : {
2710 : /* Perform additional check for transaction-snapshot mode RI updates */
2711 UIC 0 : if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
2712 LBC 0 : result = TM_Updated;
2713 : }
2714 :
2715 GIC 1420151 : if (result != TM_Ok)
2716 : {
2717 CBC 39 : Assert(result == TM_SelfModified ||
2718 : result == TM_Updated ||
2719 ECB : result == TM_Deleted ||
2720 : result == TM_BeingModified);
2721 CBC 39 : Assert(!(tp.t_data->t_infomask & HEAP_XMAX_INVALID));
2722 GIC 39 : Assert(result != TM_Updated ||
2723 : !ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
2724 CBC 39 : tmfd->ctid = tp.t_data->t_ctid;
2725 GIC 39 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
2726 CBC 39 : if (result == TM_SelfModified)
2727 GIC 18 : tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
2728 ECB : else
2729 CBC 21 : tmfd->cmax = InvalidCommandId;
2730 GIC 39 : UnlockReleaseBuffer(buffer);
2731 39 : if (have_tuple_lock)
2732 19 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
2733 39 : if (vmbuffer != InvalidBuffer)
2734 UIC 0 : ReleaseBuffer(vmbuffer);
2735 GIC 39 : return result;
2736 : }
2737 ECB :
2738 : /*
2739 : * We're about to do the actual delete -- check for conflict first, to
2740 : * avoid possibly having to roll back work we've just done.
2741 : *
2742 : * This is safe without a recheck as long as there is no possibility of
2743 : * another process scanning the page between this check and the delete
2744 : * being visible to the scan (i.e., an exclusive buffer content lock is
2745 : * continuously held from this point until the tuple delete is visible).
2746 : */
2747 GIC 1420112 : CheckForSerializableConflictIn(relation, tid, BufferGetBlockNumber(buffer));
2748 :
2749 : /* replace cid with a combo CID if necessary */
2750 1420098 : HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
2751 ECB :
2752 : /*
2753 : * Compute replica identity tuple before entering the critical section so
2754 : * we don't PANIC upon a memory allocation failure.
2755 : */
2756 GIC 1420098 : old_key_tuple = ExtractReplicaIdentity(relation, &tp, true, &old_key_copied);
2757 :
2758 : /*
2759 ECB : * If this is the first possibly-multixact-able operation in the current
2760 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
2761 : * certain that the transaction will never become a member of any older
2762 : * MultiXactIds than that. (We have to do this even if we end up just
2763 : * using our own TransactionId below, since some other backend could
2764 : * incorporate our XID into a MultiXact immediately afterwards.)
2765 : */
2766 GIC 1420098 : MultiXactIdSetOldestMember();
2767 ECB :
2768 GIC 1420098 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(tp.t_data),
2769 1420098 : tp.t_data->t_infomask, tp.t_data->t_infomask2,
2770 : xid, LockTupleExclusive, true,
2771 : &new_xmax, &new_infomask, &new_infomask2);
2772 :
2773 1420098 : START_CRIT_SECTION();
2774 :
2775 : /*
2776 : * If this transaction commits, the tuple will become DEAD sooner or
2777 : * later. Set flag that this page is a candidate for pruning once our xid
2778 : * falls below the OldestXmin horizon. If the transaction finally aborts,
2779 ECB : * the subsequent page pruning will be a no-op and the hint will be
2780 : * cleared.
2781 : */
2782 GIC 1420098 : PageSetPrunable(page, xid);
2783 :
2784 CBC 1420098 : if (PageIsAllVisible(page))
2785 : {
2786 GIC 1022 : all_visible_cleared = true;
2787 1022 : PageClearAllVisible(page);
2788 CBC 1022 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
2789 : vmbuffer, VISIBILITYMAP_VALID_BITS);
2790 EUB : }
2791 :
2792 : /* store transaction information of xact deleting the tuple */
2793 GIC 1420098 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
2794 1420098 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
2795 CBC 1420098 : tp.t_data->t_infomask |= new_infomask;
2796 GIC 1420098 : tp.t_data->t_infomask2 |= new_infomask2;
2797 CBC 1420098 : HeapTupleHeaderClearHotUpdated(tp.t_data);
2798 GIC 1420098 : HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
2799 GBC 1420098 : HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
2800 EUB : /* Make sure there is no forward chain link in t_ctid */
2801 GIC 1420098 : tp.t_data->t_ctid = tp.t_self;
2802 :
2803 EUB : /* Signal that this is actually a move into another partition */
2804 GBC 1420098 : if (changingPart)
2805 GIC 404 : HeapTupleHeaderSetMovedPartitions(tp.t_data);
2806 :
2807 GBC 1420098 : MarkBufferDirty(buffer);
2808 EUB :
2809 : /*
2810 : * XLOG stuff
2811 ECB : *
2812 : * NB: heap_abort_speculative() uses the same xlog record and replay
2813 : * routines.
2814 : */
2815 GIC 1420098 : if (RelationNeedsWAL(relation))
2816 : {
2817 : xl_heap_delete xlrec;
2818 : xl_heap_header xlhdr;
2819 : XLogRecPtr recptr;
2820 :
2821 : /*
2822 : * For logical decode we need combo CIDs to properly decode the
2823 : * catalog
2824 : */
2825 CBC 1359543 : if (RelationIsAccessibleInLogicalDecoding(relation))
2826 GIC 6005 : log_heap_new_cid(relation, &tp);
2827 :
2828 1359543 : xlrec.flags = 0;
2829 1359543 : if (all_visible_cleared)
2830 1022 : xlrec.flags |= XLH_DELETE_ALL_VISIBLE_CLEARED;
2831 CBC 1359543 : if (changingPart)
2832 GIC 404 : xlrec.flags |= XLH_DELETE_IS_PARTITION_MOVE;
2833 2719086 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
2834 1359543 : tp.t_data->t_infomask2);
2835 1359543 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
2836 1359543 : xlrec.xmax = new_xmax;
2837 :
2838 1359543 : if (old_key_tuple != NULL)
2839 : {
2840 82320 : if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
2841 CBC 221 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_TUPLE;
2842 ECB : else
2843 GIC 82099 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_KEY;
2844 : }
2845 :
2846 1359543 : XLogBeginInsert();
2847 1359543 : XLogRegisterData((char *) &xlrec, SizeOfHeapDelete);
2848 ECB :
2849 CBC 1359543 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
2850 :
2851 : /*
2852 : * Log replica identity of the deleted tuple if there is one
2853 ECB : */
2854 GIC 1359543 : if (old_key_tuple != NULL)
2855 ECB : {
2856 CBC 82320 : xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
2857 GIC 82320 : xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
2858 CBC 82320 : xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
2859 ECB :
2860 GIC 82320 : XLogRegisterData((char *) &xlhdr, SizeOfHeapHeader);
2861 82320 : XLogRegisterData((char *) old_key_tuple->t_data
2862 ECB : + SizeofHeapTupleHeader,
2863 GIC 82320 : old_key_tuple->t_len
2864 : - SizeofHeapTupleHeader);
2865 : }
2866 :
2867 : /* filtering by origin on a row level is much more efficient */
2868 1359543 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2869 :
2870 CBC 1359543 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
2871 :
2872 GIC 1359543 : PageSetLSN(page, recptr);
2873 ECB : }
2874 :
2875 GIC 1420098 : END_CRIT_SECTION();
2876 :
2877 1420098 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2878 :
2879 1420098 : if (vmbuffer != InvalidBuffer)
2880 1022 : ReleaseBuffer(vmbuffer);
2881 ECB :
2882 EUB : /*
2883 : * If the tuple has toasted out-of-line attributes, we need to delete
2884 : * those items too. We have to do this before releasing the buffer
2885 : * because we need to look at the contents of the tuple, but it's OK to
2886 : * release the content lock on the buffer first.
2887 : */
2888 GIC 1420098 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
2889 1569 : relation->rd_rel->relkind != RELKIND_MATVIEW)
2890 : {
2891 : /* toast table entries should never be recursively toasted */
2892 1559 : Assert(!HeapTupleHasExternal(&tp));
2893 : }
2894 1418539 : else if (HeapTupleHasExternal(&tp))
2895 233 : heap_toast_delete(relation, &tp, false);
2896 :
2897 : /*
2898 : * Mark tuple for invalidation from system caches at next command
2899 : * boundary. We have to do this before releasing the buffer because we
2900 : * need to look at the contents of the tuple.
2901 : */
2902 CBC 1420098 : CacheInvalidateHeapTuple(relation, &tp, NULL);
2903 :
2904 ECB : /* Now we can release the buffer */
2905 GIC 1420098 : ReleaseBuffer(buffer);
2906 ECB :
2907 : /*
2908 : * Release the lmgr tuple lock, if we had it.
2909 : */
2910 CBC 1420098 : if (have_tuple_lock)
2911 18 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
2912 ECB :
2913 CBC 1420098 : pgstat_count_heap_delete(relation);
2914 :
2915 1420098 : if (old_key_tuple != NULL && old_key_copied)
2916 82100 : heap_freetuple(old_key_tuple);
2917 ECB :
2918 GIC 1420098 : return TM_Ok;
2919 : }
2920 :
2921 : /*
2922 : * simple_heap_delete - delete a tuple
2923 : *
2924 : * This routine may be used to delete a tuple when concurrent updates of
2925 ECB : * the target tuple are not expected (for example, because we have a lock
2926 : * on the relation associated with the tuple). Any failure is reported
2927 : * via ereport().
2928 : */
2929 : void
2930 CBC 536249 : simple_heap_delete(Relation relation, ItemPointer tid)
2931 ECB : {
2932 : TM_Result result;
2933 : TM_FailureData tmfd;
2934 :
2935 GIC 536249 : result = heap_delete(relation, tid,
2936 : GetCurrentCommandId(true), InvalidSnapshot,
2937 ECB : true /* wait for commit */ ,
2938 : &tmfd, false /* changingPart */ );
2939 CBC 536249 : switch (result)
2940 ECB : {
2941 UIC 0 : case TM_SelfModified:
2942 : /* Tuple was already updated in current command? */
2943 LBC 0 : elog(ERROR, "tuple already updated by self");
2944 : break;
2945 :
2946 GIC 536249 : case TM_Ok:
2947 : /* done successfully */
2948 536249 : break;
2949 :
2950 UIC 0 : case TM_Updated:
2951 0 : elog(ERROR, "tuple concurrently updated");
2952 : break;
2953 ECB :
2954 UIC 0 : case TM_Deleted:
2955 0 : elog(ERROR, "tuple concurrently deleted");
2956 : break;
2957 :
2958 0 : default:
2959 0 : elog(ERROR, "unrecognized heap_delete status: %u", result);
2960 : break;
2961 : }
2962 GIC 536249 : }
2963 :
2964 : /*
2965 : * heap_update - replace a tuple
2966 : *
2967 : * See table_tuple_update() for an explanation of the parameters, except that
2968 ECB : * this routine directly takes a tuple rather than a slot.
2969 : *
2970 : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
2971 : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
2972 : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
2973 : * generated by another transaction).
2974 : */
2975 : TM_Result
2976 GIC 418994 : heap_update(Relation relation, ItemPointer otid, HeapTuple newtup,
2977 : CommandId cid, Snapshot crosscheck, bool wait,
2978 : TM_FailureData *tmfd, LockTupleMode *lockmode,
2979 : TU_UpdateIndexes *update_indexes)
2980 : {
2981 : TM_Result result;
2982 418994 : TransactionId xid = GetCurrentTransactionId();
2983 : Bitmapset *hot_attrs;
2984 : Bitmapset *sum_attrs;
2985 ECB : Bitmapset *key_attrs;
2986 : Bitmapset *id_attrs;
2987 : Bitmapset *interesting_attrs;
2988 : Bitmapset *modified_attrs;
2989 : ItemId lp;
2990 : HeapTupleData oldtup;
2991 : HeapTuple heaptup;
2992 GIC 418994 : HeapTuple old_key_tuple = NULL;
2993 418994 : bool old_key_copied = false;
2994 : Page page;
2995 : BlockNumber block;
2996 : MultiXactStatus mxact_status;
2997 : Buffer buffer,
2998 : newbuf,
2999 418994 : vmbuffer = InvalidBuffer,
3000 418994 : vmbuffer_new = InvalidBuffer;
3001 ECB : bool need_toast;
3002 : Size newtupsize,
3003 : pagefree;
3004 CBC 418994 : bool have_tuple_lock = false;
3005 : bool iscombo;
3006 GIC 418994 : bool use_hot_update = false;
3007 GNC 418994 : bool summarized_update = false;
3008 ECB : bool key_intact;
3009 GIC 418994 : bool all_visible_cleared = false;
3010 CBC 418994 : bool all_visible_cleared_new = false;
3011 : bool checked_lockers;
3012 EUB : bool locker_remains;
3013 GBC 418994 : bool id_has_external = false;
3014 : TransactionId xmax_new_tuple,
3015 : xmax_old_tuple;
3016 : uint16 infomask_old_tuple,
3017 ECB : infomask2_old_tuple,
3018 : infomask_new_tuple,
3019 : infomask2_new_tuple;
3020 :
3021 CBC 418994 : Assert(ItemPointerIsValid(otid));
3022 :
3023 : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
3024 GIC 418994 : Assert(HeapTupleHeaderGetNatts(newtup->t_data) <=
3025 : RelationGetNumberOfAttributes(relation));
3026 :
3027 : /*
3028 : * Forbid this during a parallel operation, lest it allocate a combo CID.
3029 : * Other workers might need that combo CID for visibility checks, and we
3030 : * have no provision for broadcasting it to them.
3031 : */
3032 418994 : if (IsInParallelMode())
3033 UIC 0 : ereport(ERROR,
3034 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
3035 : errmsg("cannot update tuples during a parallel operation")));
3036 :
3037 ECB : /*
3038 : * Fetch the list of attributes to be checked for various operations.
3039 : *
3040 : * For HOT considerations, this is wasted effort if we fail to update or
3041 : * have to put the new tuple on a different page. But we must compute the
3042 : * list before obtaining buffer lock --- in the worst case, if we are
3043 : * doing an update on one of the relevant system catalogs, we could
3044 : * deadlock if we try to fetch the list later. In any case, the relcache
3045 : * caches the data so this is usually pretty cheap.
3046 : *
3047 : * We also need columns used by the replica identity and columns that are
3048 : * considered the "key" of rows in the table.
3049 : *
3050 : * Note that we get copies of each bitmap, so we need not worry about
3051 : * relcache flush happening midway through.
3052 : */
3053 GNC 418994 : hot_attrs = RelationGetIndexAttrBitmap(relation,
3054 : INDEX_ATTR_BITMAP_HOT_BLOCKING);
3055 418994 : sum_attrs = RelationGetIndexAttrBitmap(relation,
3056 : INDEX_ATTR_BITMAP_SUMMARIZED);
3057 GIC 418994 : key_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_KEY);
3058 418994 : id_attrs = RelationGetIndexAttrBitmap(relation,
3059 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
3060 418994 : interesting_attrs = NULL;
3061 418994 : interesting_attrs = bms_add_members(interesting_attrs, hot_attrs);
3062 GNC 418994 : interesting_attrs = bms_add_members(interesting_attrs, sum_attrs);
3063 GIC 418994 : interesting_attrs = bms_add_members(interesting_attrs, key_attrs);
3064 418994 : interesting_attrs = bms_add_members(interesting_attrs, id_attrs);
3065 :
3066 CBC 418994 : block = ItemPointerGetBlockNumber(otid);
3067 GIC 418994 : buffer = ReadBuffer(relation, block);
3068 418994 : page = BufferGetPage(buffer);
3069 :
3070 ECB : /*
3071 : * Before locking the buffer, pin the visibility map page if it appears to
3072 : * be necessary. Since we haven't got the lock yet, someone else might be
3073 : * in the middle of changing this, so we'll need to recheck after we have
3074 : * the lock.
3075 : */
3076 GIC 418994 : if (PageIsAllVisible(page))
3077 961 : visibilitymap_pin(relation, block, &vmbuffer);
3078 :
3079 418994 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3080 :
3081 CBC 418994 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(otid));
3082 GBC 418994 : Assert(ItemIdIsNormal(lp));
3083 :
3084 : /*
3085 : * Fill in enough data in oldtup for HeapDetermineColumnsInfo to work
3086 ECB : * properly.
3087 : */
3088 GIC 418994 : oldtup.t_tableOid = RelationGetRelid(relation);
3089 CBC 418994 : oldtup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
3090 418994 : oldtup.t_len = ItemIdGetLength(lp);
3091 418994 : oldtup.t_self = *otid;
3092 :
3093 : /* the new tuple is ready, except for this: */
3094 GIC 418994 : newtup->t_tableOid = RelationGetRelid(relation);
3095 :
3096 : /*
3097 : * Determine columns modified by the update. Additionally, identify
3098 ECB : * whether any of the unmodified replica identity key attributes in the
3099 : * old tuple is externally stored or not. This is required because for
3100 : * such attributes the flattened value won't be WAL logged as part of the
3101 : * new tuple so we must include it as part of the old_key_tuple. See
3102 EUB : * ExtractReplicaIdentity.
3103 : */
3104 GIC 418994 : modified_attrs = HeapDetermineColumnsInfo(relation, interesting_attrs,
3105 : id_attrs, &oldtup,
3106 : newtup, &id_has_external);
3107 :
3108 : /*
3109 : * If we're not updating any "key" column, we can grab a weaker lock type.
3110 : * This allows for more concurrency when we are running simultaneously
3111 : * with foreign key checks.
3112 : *
3113 : * Note that if a column gets detoasted while executing the update, but
3114 : * the value ends up being the same, this test will fail and we will use
3115 : * the stronger lock. This is acceptable; the important case to optimize
3116 : * is updates that don't manipulate key columns, not those that
3117 : * serendipitously arrive at the same key values.
3118 : */
3119 418994 : if (!bms_overlap(modified_attrs, key_attrs))
3120 : {
3121 415546 : *lockmode = LockTupleNoKeyExclusive;
3122 415546 : mxact_status = MultiXactStatusNoKeyUpdate;
3123 415546 : key_intact = true;
3124 :
3125 ECB : /*
3126 : * If this is the first possibly-multixact-able operation in the
3127 : * current transaction, set my per-backend OldestMemberMXactId
3128 : * setting. We can be certain that the transaction will never become a
3129 : * member of any older MultiXactIds than that. (We have to do this
3130 : * even if we end up just using our own TransactionId below, since
3131 : * some other backend could incorporate our XID into a MultiXact
3132 : * immediately afterwards.)
3133 : */
3134 GIC 415546 : MultiXactIdSetOldestMember();
3135 ECB : }
3136 : else
3137 : {
3138 GIC 3448 : *lockmode = LockTupleExclusive;
3139 CBC 3448 : mxact_status = MultiXactStatusUpdate;
3140 GIC 3448 : key_intact = false;
3141 : }
3142 :
3143 : /*
3144 : * Note: beyond this point, use oldtup not otid to refer to old tuple.
3145 ECB : * otid may very well point at newtup->t_self, which we will overwrite
3146 : * with the new tuple's location, so there's great risk of confusion if we
3147 : * use otid anymore.
3148 : */
3149 :
3150 GIC 418994 : l2:
3151 418995 : checked_lockers = false;
3152 418995 : locker_remains = false;
3153 418995 : result = HeapTupleSatisfiesUpdate(&oldtup, cid, buffer);
3154 :
3155 : /* see below about the "no wait" case */
3156 CBC 418995 : Assert(result != TM_BeingModified || wait);
3157 ECB :
3158 CBC 418995 : if (result == TM_Invisible)
3159 : {
3160 UIC 0 : UnlockReleaseBuffer(buffer);
3161 0 : ereport(ERROR,
3162 : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3163 : errmsg("attempted to update invisible tuple")));
3164 : }
3165 GIC 418995 : else if (result == TM_BeingModified && wait)
3166 ECB : {
3167 : TransactionId xwait;
3168 : uint16 infomask;
3169 CBC 35847 : bool can_continue = false;
3170 :
3171 ECB : /*
3172 : * XXX note that we don't consider the "no wait" case here. This
3173 : * isn't a problem currently because no caller uses that case, but it
3174 : * should be fixed if such a caller is introduced. It wasn't a
3175 : * problem previously because this code would always wait, but now
3176 : * that some tuple locks do not conflict with one of the lock modes we
3177 : * use, it is possible that this case is interesting to handle
3178 : * specially.
3179 : *
3180 : * This may cause failures with third-party code that calls
3181 : * heap_update directly.
3182 : */
3183 :
3184 : /* must copy state data before unlocking buffer */
3185 CBC 35847 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3186 35847 : infomask = oldtup.t_data->t_infomask;
3187 ECB :
3188 : /*
3189 : * Now we have to do something about the existing locker. If it's a
3190 : * multi, sleep on it; we might be awakened before it is completely
3191 : * gone (or even not sleep at all in some cases); we need to preserve
3192 : * it as locker, unless it is gone completely.
3193 : *
3194 : * If it's not a multi, we need to check for sleeping conditions
3195 : * before actually going to sleep. If the update doesn't conflict
3196 : * with the locks, we just continue without sleeping (but making sure
3197 : * it is preserved).
3198 : *
3199 : * Before sleeping, we need to acquire tuple lock to establish our
3200 : * priority for the tuple (see heap_lock_tuple). LockTuple will
3201 EUB : * release us when we are next-in-line for the tuple. Note we must
3202 : * not acquire the tuple lock until we're sure we're going to sleep;
3203 : * otherwise we're open for race conditions with other transactions
3204 : * holding the tuple lock which sleep on us.
3205 : *
3206 : * If we are forced to "start over" below, we keep the tuple lock;
3207 : * this arranges that we stay at the head of the line while rechecking
3208 ECB : * tuple state.
3209 : */
3210 CBC 35847 : if (infomask & HEAP_XMAX_IS_MULTI)
3211 : {
3212 : TransactionId update_xact;
3213 : int remain;
3214 60 : bool current_is_member = false;
3215 ECB :
3216 GIC 60 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
3217 ECB : *lockmode, ¤t_is_member))
3218 : {
3219 CBC 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3220 ECB :
3221 : /*
3222 : * Acquire the lock, if necessary (but skip it when we're
3223 : * requesting a lock and already have one; avoids deadlock).
3224 : */
3225 CBC 8 : if (!current_is_member)
3226 LBC 0 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3227 EUB : LockWaitBlock, &have_tuple_lock);
3228 ECB :
3229 : /* wait for multixact */
3230 CBC 8 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
3231 ECB : relation, &oldtup.t_self, XLTW_Update,
3232 : &remain);
3233 CBC 8 : checked_lockers = true;
3234 8 : locker_remains = remain != 0;
3235 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3236 ECB :
3237 : /*
3238 : * If xwait had just locked the tuple then some other xact
3239 : * could update this tuple before we get to this point. Check
3240 : * for xmax change, and start over if so.
3241 : */
3242 GIC 8 : if (xmax_infomask_changed(oldtup.t_data->t_infomask,
3243 8 : infomask) ||
3244 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3245 : xwait))
3246 UIC 0 : goto l2;
3247 : }
3248 ECB :
3249 : /*
3250 EUB : * Note that the multixact may not be done by now. It could have
3251 : * surviving members; our own xact or other subxacts of this
3252 : * backend, and also any other concurrent transaction that locked
3253 : * the tuple with LockTupleKeyShare if we only got
3254 : * LockTupleNoKeyExclusive. If this is the case, we have to be
3255 : * careful to mark the updated tuple with the surviving members in
3256 : * Xmax.
3257 : *
3258 : * Note that there could have been another update in the
3259 : * MultiXact. In that case, we need to check whether it committed
3260 : * or aborted. If it aborted we are safe to update it again;
3261 : * otherwise there is an update conflict, and we have to return
3262 ECB : * TableTuple{Deleted, Updated} below.
3263 : *
3264 : * In the LockTupleExclusive case, we still need to preserve the
3265 : * surviving members: those would include the tuple locks we had
3266 : * before this one, which are important to keep in case this
3267 : * subxact aborts.
3268 : */
3269 GIC 60 : if (!HEAP_XMAX_IS_LOCKED_ONLY(oldtup.t_data->t_infomask))
3270 8 : update_xact = HeapTupleGetUpdateXid(oldtup.t_data);
3271 : else
3272 52 : update_xact = InvalidTransactionId;
3273 :
3274 : /*
3275 : * There was no UPDATE in the MultiXact; or it aborted. No
3276 ECB : * TransactionIdIsInProgress() call needed here, since we called
3277 : * MultiXactIdWait() above.
3278 : */
3279 CBC 68 : if (!TransactionIdIsValid(update_xact) ||
3280 GIC 8 : TransactionIdDidAbort(update_xact))
3281 CBC 53 : can_continue = true;
3282 : }
3283 35787 : else if (TransactionIdIsCurrentTransactionId(xwait))
3284 : {
3285 ECB : /*
3286 : * The only locker is ourselves; we can avoid grabbing the tuple
3287 : * lock here, but must preserve our locking information.
3288 : */
3289 GIC 35706 : checked_lockers = true;
3290 35706 : locker_remains = true;
3291 35706 : can_continue = true;
3292 : }
3293 81 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) && key_intact)
3294 : {
3295 : /*
3296 ECB : * If it's just a key-share locker, and we're not changing the key
3297 : * columns, we don't need to wait for it to end; but we need to
3298 : * preserve it as locker.
3299 : */
3300 GIC 29 : checked_lockers = true;
3301 29 : locker_remains = true;
3302 29 : can_continue = true;
3303 ECB : }
3304 : else
3305 : {
3306 : /*
3307 : * Wait for regular transaction to end; but first, acquire tuple
3308 : * lock.
3309 : */
3310 GIC 52 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3311 52 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3312 ECB : LockWaitBlock, &have_tuple_lock);
3313 CBC 52 : XactLockTableWait(xwait, relation, &oldtup.t_self,
3314 ECB : XLTW_Update);
3315 CBC 52 : checked_lockers = true;
3316 52 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3317 ECB :
3318 : /*
3319 : * xwait is done, but if xwait had just locked the tuple then some
3320 : * other xact could update this tuple before we get to this point.
3321 : * Check for xmax change, and start over if so.
3322 : */
3323 GIC 52 : if (xmax_infomask_changed(oldtup.t_data->t_infomask, infomask) ||
3324 CBC 51 : !TransactionIdEquals(xwait,
3325 : HeapTupleHeaderGetRawXmax(oldtup.t_data)))
3326 GIC 1 : goto l2;
3327 :
3328 : /* Otherwise check if it committed or aborted */
3329 51 : UpdateXmaxHintBits(oldtup.t_data, buffer, xwait);
3330 51 : if (oldtup.t_data->t_infomask & HEAP_XMAX_INVALID)
3331 12 : can_continue = true;
3332 : }
3333 :
3334 35846 : if (can_continue)
3335 35800 : result = TM_Ok;
3336 46 : else if (!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid))
3337 CBC 41 : result = TM_Updated;
3338 EUB : else
3339 GIC 5 : result = TM_Deleted;
3340 : }
3341 EUB :
3342 GBC 418994 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3343 EUB : {
3344 : /* Perform additional check for transaction-snapshot mode RI updates */
3345 UIC 0 : if (!HeapTupleSatisfiesVisibility(&oldtup, crosscheck, buffer))
3346 ECB : {
3347 LBC 0 : result = TM_Updated;
3348 0 : Assert(!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
3349 : }
3350 ECB : }
3351 :
3352 CBC 418994 : if (result != TM_Ok)
3353 : {
3354 114 : Assert(result == TM_SelfModified ||
3355 ECB : result == TM_Updated ||
3356 : result == TM_Deleted ||
3357 : result == TM_BeingModified);
3358 GIC 114 : Assert(!(oldtup.t_data->t_infomask & HEAP_XMAX_INVALID));
3359 CBC 114 : Assert(result != TM_Updated ||
3360 : !ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
3361 GIC 114 : tmfd->ctid = oldtup.t_data->t_ctid;
3362 114 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(oldtup.t_data);
3363 114 : if (result == TM_SelfModified)
3364 45 : tmfd->cmax = HeapTupleHeaderGetCmax(oldtup.t_data);
3365 : else
3366 69 : tmfd->cmax = InvalidCommandId;
3367 114 : UnlockReleaseBuffer(buffer);
3368 114 : if (have_tuple_lock)
3369 39 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
3370 114 : if (vmbuffer != InvalidBuffer)
3371 UIC 0 : ReleaseBuffer(vmbuffer);
3372 GNC 114 : *update_indexes = TU_None;
3373 :
3374 GIC 114 : bms_free(hot_attrs);
3375 GNC 114 : bms_free(sum_attrs);
3376 GIC 114 : bms_free(key_attrs);
3377 114 : bms_free(id_attrs);
3378 114 : bms_free(modified_attrs);
3379 114 : bms_free(interesting_attrs);
3380 114 : return result;
3381 : }
3382 ECB :
3383 : /*
3384 : * If we didn't pin the visibility map page and the page has become all
3385 : * visible while we were busy locking the buffer, or during some
3386 : * subsequent window during which we had it unlocked, we'll have to unlock
3387 : * and re-lock, to avoid holding the buffer lock across an I/O. That's a
3388 : * bit unfortunate, especially since we'll now have to recheck whether the
3389 : * tuple has been locked or updated under us, but hopefully it won't
3390 : * happen very often.
3391 : */
3392 GIC 418880 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3393 : {
3394 LBC 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3395 0 : visibilitymap_pin(relation, block, &vmbuffer);
3396 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3397 UIC 0 : goto l2;
3398 ECB : }
3399 :
3400 : /* Fill in transaction status data */
3401 :
3402 : /*
3403 : * If the tuple we're updating is locked, we need to preserve the locking
3404 : * info in the old tuple's Xmax. Prepare a new Xmax value for this.
3405 : */
3406 GIC 418880 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3407 418880 : oldtup.t_data->t_infomask,
3408 418880 : oldtup.t_data->t_infomask2,
3409 : xid, *lockmode, true,
3410 : &xmax_old_tuple, &infomask_old_tuple,
3411 : &infomask2_old_tuple);
3412 :
3413 ECB : /*
3414 : * And also prepare an Xmax value for the new copy of the tuple. If there
3415 : * was no xmax previously, or there was one but all lockers are now gone,
3416 : * then use InvalidTransactionId; otherwise, get the xmax from the old
3417 : * tuple. (In rare cases that might also be InvalidTransactionId and yet
3418 : * not have the HEAP_XMAX_INVALID bit set; that's fine.)
3419 : */
3420 CBC 418880 : if ((oldtup.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3421 GIC 35788 : HEAP_LOCKED_UPGRADED(oldtup.t_data->t_infomask) ||
3422 35736 : (checked_lockers && !locker_remains))
3423 383092 : xmax_new_tuple = InvalidTransactionId;
3424 : else
3425 CBC 35788 : xmax_new_tuple = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3426 ECB :
3427 GIC 418880 : if (!TransactionIdIsValid(xmax_new_tuple))
3428 ECB : {
3429 CBC 383092 : infomask_new_tuple = HEAP_XMAX_INVALID;
3430 383092 : infomask2_new_tuple = 0;
3431 ECB : }
3432 : else
3433 : {
3434 : /*
3435 : * If we found a valid Xmax for the new tuple, then the infomask bits
3436 : * to use on the new tuple depend on what was there on the old one.
3437 : * Note that since we're doing an update, the only possibility is that
3438 : * the lockers had FOR KEY SHARE lock.
3439 : */
3440 GIC 35788 : if (oldtup.t_data->t_infomask & HEAP_XMAX_IS_MULTI)
3441 ECB : {
3442 GIC 53 : GetMultiXactIdHintBits(xmax_new_tuple, &infomask_new_tuple,
3443 : &infomask2_new_tuple);
3444 : }
3445 : else
3446 : {
3447 35735 : infomask_new_tuple = HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_LOCK_ONLY;
3448 35735 : infomask2_new_tuple = 0;
3449 : }
3450 ECB : }
3451 :
3452 : /*
3453 : * Prepare the new tuple with the appropriate initial values of Xmin and
3454 : * Xmax, as well as initial infomask bits as computed above.
3455 : */
3456 GIC 418880 : newtup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
3457 CBC 418880 : newtup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
3458 GIC 418880 : HeapTupleHeaderSetXmin(newtup->t_data, xid);
3459 418880 : HeapTupleHeaderSetCmin(newtup->t_data, cid);
3460 418880 : newtup->t_data->t_infomask |= HEAP_UPDATED | infomask_new_tuple;
3461 418880 : newtup->t_data->t_infomask2 |= infomask2_new_tuple;
3462 418880 : HeapTupleHeaderSetXmax(newtup->t_data, xmax_new_tuple);
3463 :
3464 : /*
3465 : * Replace cid with a combo CID if necessary. Note that we already put
3466 : * the plain cid into the new tuple.
3467 : */
3468 418880 : HeapTupleHeaderAdjustCmax(oldtup.t_data, &cid, &iscombo);
3469 :
3470 : /*
3471 : * If the toaster needs to be activated, OR if the new tuple will not fit
3472 : * on the same page as the old, then we need to release the content lock
3473 : * (but not the pin!) on the old tuple's buffer while we are off doing
3474 : * TOAST and/or table-file-extension work. We must mark the old tuple to
3475 : * show that it's locked, else other processes may try to update it
3476 : * themselves.
3477 : *
3478 : * We need to invoke the toaster if there are already any out-of-line
3479 : * toasted values present, or if the new tuple is over-threshold.
3480 : */
3481 418880 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3482 UIC 0 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3483 : {
3484 ECB : /* toast table entries should never be recursively toasted */
3485 UIC 0 : Assert(!HeapTupleHasExternal(&oldtup));
3486 0 : Assert(!HeapTupleHasExternal(newtup));
3487 LBC 0 : need_toast = false;
3488 : }
3489 : else
3490 GIC 418880 : need_toast = (HeapTupleHasExternal(&oldtup) ||
3491 837523 : HeapTupleHasExternal(newtup) ||
3492 CBC 418643 : newtup->t_len > TOAST_TUPLE_THRESHOLD);
3493 :
3494 GIC 418880 : pagefree = PageGetHeapFreeSpace(page);
3495 ECB :
3496 GBC 418880 : newtupsize = MAXALIGN(newtup->t_len);
3497 :
3498 CBC 418880 : if (need_toast || newtupsize > pagefree)
3499 GIC 196455 : {
3500 ECB : TransactionId xmax_lock_old_tuple;
3501 : uint16 infomask_lock_old_tuple,
3502 : infomask2_lock_old_tuple;
3503 GIC 196455 : bool cleared_all_frozen = false;
3504 :
3505 : /*
3506 : * To prevent concurrent sessions from updating the tuple, we have to
3507 : * temporarily mark it locked, while we release the page-level lock.
3508 : *
3509 ECB : * To satisfy the rule that any xid potentially appearing in a buffer
3510 : * written out to disk, we unfortunately have to WAL log this
3511 : * temporary modification. We can reuse xl_heap_lock for this
3512 : * purpose. If we crash/error before following through with the
3513 : * actual update, xmax will be of an aborted transaction, allowing
3514 : * other sessions to proceed.
3515 : */
3516 :
3517 : /*
3518 : * Compute xmax / infomask appropriate for locking the tuple. This has
3519 : * to be done separately from the combo that's going to be used for
3520 : * updating, because the potentially created multixact would otherwise
3521 : * be wrong.
3522 : */
3523 CBC 196455 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3524 GIC 196455 : oldtup.t_data->t_infomask,
3525 196455 : oldtup.t_data->t_infomask2,
3526 : xid, *lockmode, false,
3527 : &xmax_lock_old_tuple, &infomask_lock_old_tuple,
3528 : &infomask2_lock_old_tuple);
3529 :
3530 196455 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(infomask_lock_old_tuple));
3531 :
3532 196455 : START_CRIT_SECTION();
3533 :
3534 : /* Clear obsolete visibility flags ... */
3535 196455 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3536 196455 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3537 196455 : HeapTupleClearHotUpdated(&oldtup);
3538 : /* ... and store info about transaction updating this tuple */
3539 196455 : Assert(TransactionIdIsValid(xmax_lock_old_tuple));
3540 CBC 196455 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_lock_old_tuple);
3541 GIC 196455 : oldtup.t_data->t_infomask |= infomask_lock_old_tuple;
3542 196455 : oldtup.t_data->t_infomask2 |= infomask2_lock_old_tuple;
3543 196455 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
3544 :
3545 : /* temporarily make it look not-updated, but locked */
3546 196455 : oldtup.t_data->t_ctid = oldtup.t_self;
3547 :
3548 : /*
3549 ECB : * Clear all-frozen bit on visibility map if needed. We could
3550 : * immediately reset ALL_VISIBLE, but given that the WAL logging
3551 : * overhead would be unchanged, that doesn't seem necessarily
3552 : * worthwhile.
3553 : */
3554 GIC 196959 : if (PageIsAllVisible(page) &&
3555 504 : visibilitymap_clear(relation, block, vmbuffer,
3556 ECB : VISIBILITYMAP_ALL_FROZEN))
3557 GIC 351 : cleared_all_frozen = true;
3558 ECB :
3559 GIC 196455 : MarkBufferDirty(buffer);
3560 :
3561 196455 : if (RelationNeedsWAL(relation))
3562 : {
3563 : xl_heap_lock xlrec;
3564 : XLogRecPtr recptr;
3565 :
3566 186329 : XLogBeginInsert();
3567 CBC 186329 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3568 ECB :
3569 GIC 186329 : xlrec.offnum = ItemPointerGetOffsetNumber(&oldtup.t_self);
3570 186329 : xlrec.locking_xid = xmax_lock_old_tuple;
3571 372658 : xlrec.infobits_set = compute_infobits(oldtup.t_data->t_infomask,
3572 186329 : oldtup.t_data->t_infomask2);
3573 186329 : xlrec.flags =
3574 CBC 186329 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
3575 GIC 186329 : XLogRegisterData((char *) &xlrec, SizeOfHeapLock);
3576 186329 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
3577 186329 : PageSetLSN(page, recptr);
3578 : }
3579 :
3580 196455 : END_CRIT_SECTION();
3581 :
3582 196455 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3583 :
3584 ECB : /*
3585 : * Let the toaster do its thing, if needed.
3586 : *
3587 : * Note: below this point, heaptup is the data we actually intend to
3588 : * store into the relation; newtup is the caller's original untoasted
3589 : * data.
3590 : */
3591 GIC 196455 : if (need_toast)
3592 : {
3593 : /* Note we always use WAL and FSM during updates */
3594 3085 : heaptup = heap_toast_insert_or_update(relation, newtup, &oldtup, 0);
3595 3085 : newtupsize = MAXALIGN(heaptup->t_len);
3596 : }
3597 : else
3598 193370 : heaptup = newtup;
3599 :
3600 : /*
3601 : * Now, do we need a new page for the tuple, or not? This is a bit
3602 : * tricky since someone else could have added tuples to the page while
3603 : * we weren't looking. We have to recheck the available space after
3604 ECB : * reacquiring the buffer lock. But don't bother to do that if the
3605 : * former amount of free space is still not enough; it's unlikely
3606 : * there's more free now than before.
3607 : *
3608 : * What's more, if we need to get a new page, we will need to acquire
3609 : * buffer locks on both old and new pages. To avoid deadlock against
3610 : * some other backend trying to get the same two locks in the other
3611 : * order, we must be consistent about the order we get the locks in.
3612 : * We use the rule "lock the lower-numbered page of the relation
3613 : * first". To implement this, we must do RelationGetBufferForTuple
3614 : * while not holding the lock on the old page, and we must rely on it
3615 : * to get the locks on both pages in the correct order.
3616 : *
3617 : * Another consideration is that we need visibility map page pin(s) if
3618 : * we will have to clear the all-visible flag on either page. If we
3619 : * call RelationGetBufferForTuple, we rely on it to acquire any such
3620 : * pins; but if we don't, we have to handle that here. Hence we need
3621 : * a loop.
3622 : */
3623 : for (;;)
3624 : {
3625 GIC 196456 : if (newtupsize > pagefree)
3626 : {
3627 ECB : /* It doesn't fit, must use RelationGetBufferForTuple. */
3628 CBC 196178 : newbuf = RelationGetBufferForTuple(relation, heaptup->t_len,
3629 : buffer, 0, NULL,
3630 : &vmbuffer_new, &vmbuffer,
3631 : 0);
3632 ECB : /* We're all done. */
3633 CBC 196178 : break;
3634 ECB : }
3635 : /* Acquire VM page pin if needed and we don't have it. */
3636 GIC 278 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3637 UIC 0 : visibilitymap_pin(relation, block, &vmbuffer);
3638 ECB : /* Re-acquire the lock on the old tuple's page. */
3639 GIC 278 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3640 : /* Re-check using the up-to-date free space */
3641 CBC 278 : pagefree = PageGetHeapFreeSpace(page);
3642 GIC 278 : if (newtupsize > pagefree ||
3643 CBC 277 : (vmbuffer == InvalidBuffer && PageIsAllVisible(page)))
3644 ECB : {
3645 : /*
3646 : * Rats, it doesn't fit anymore, or somebody just now set the
3647 : * all-visible flag. We must now unlock and loop to avoid
3648 : * deadlock. Fortunately, this path should seldom be taken.
3649 : */
3650 CBC 1 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3651 ECB : }
3652 : else
3653 : {
3654 : /* We're all done. */
3655 GIC 277 : newbuf = buffer;
3656 CBC 277 : break;
3657 ECB : }
3658 : }
3659 : }
3660 : else
3661 : {
3662 : /* No TOAST work needed, and it'll fit on same page */
3663 GIC 222425 : newbuf = buffer;
3664 222425 : heaptup = newtup;
3665 : }
3666 :
3667 : /*
3668 : * We're about to do the actual update -- check for conflict first, to
3669 ECB : * avoid possibly having to roll back work we've just done.
3670 : *
3671 : * This is safe without a recheck as long as there is no possibility of
3672 : * another process scanning the pages between this check and the update
3673 : * being visible to the scan (i.e., exclusive buffer content lock(s) are
3674 : * continuously held from this point until the tuple update is visible).
3675 : *
3676 : * For the new tuple the only check needed is at the relation level, but
3677 : * since both tuples are in the same relation and the check for oldtup
3678 : * will include checking the relation level, there is no benefit to a
3679 : * separate check for the new tuple.
3680 : */
3681 GIC 418880 : CheckForSerializableConflictIn(relation, &oldtup.t_self,
3682 ECB : BufferGetBlockNumber(buffer));
3683 :
3684 : /*
3685 : * At this point newbuf and buffer are both pinned and locked, and newbuf
3686 : * has enough space for the new tuple. If they are the same buffer, only
3687 : * one pin is held.
3688 : */
3689 :
3690 CBC 418868 : if (newbuf == buffer)
3691 ECB : {
3692 : /*
3693 : * Since the new tuple is going into the same page, we might be able
3694 : * to do a HOT update. Check if any of the index columns have been
3695 : * changed.
3696 : */
3697 GIC 222690 : if (!bms_overlap(modified_attrs, hot_attrs))
3698 : {
3699 213765 : use_hot_update = true;
3700 :
3701 : /*
3702 : * If none of the columns that are used in hot-blocking indexes
3703 : * were updated, we can apply HOT, but we do still need to check
3704 : * if we need to update the summarizing indexes, and update those
3705 : * indexes if the columns were updated, or we may fail to detect
3706 : * e.g. value bound changes in BRIN minmax indexes.
3707 : */
3708 GNC 213765 : if (bms_overlap(modified_attrs, sum_attrs))
3709 1641 : summarized_update = true;
3710 : }
3711 : }
3712 : else
3713 ECB : {
3714 : /* Set a hint that the old page could use prune/defrag */
3715 GIC 196178 : PageSetFull(page);
3716 ECB : }
3717 :
3718 : /*
3719 : * Compute replica identity tuple before entering the critical section so
3720 : * we don't PANIC upon a memory allocation failure.
3721 : * ExtractReplicaIdentity() will return NULL if nothing needs to be
3722 : * logged. Pass old key required as true only if the replica identity key
3723 : * columns are modified or it has external data.
3724 : */
3725 GIC 418868 : old_key_tuple = ExtractReplicaIdentity(relation, &oldtup,
3726 418868 : bms_overlap(modified_attrs, id_attrs) ||
3727 ECB : id_has_external,
3728 : &old_key_copied);
3729 :
3730 : /* NO EREPORT(ERROR) from here till changes are logged */
3731 GIC 418868 : START_CRIT_SECTION();
3732 :
3733 : /*
3734 : * If this transaction commits, the old tuple will become DEAD sooner or
3735 : * later. Set flag that this page is a candidate for pruning once our xid
3736 ECB : * falls below the OldestXmin horizon. If the transaction finally aborts,
3737 : * the subsequent page pruning will be a no-op and the hint will be
3738 : * cleared.
3739 : *
3740 : * XXX Should we set hint on newbuf as well? If the transaction aborts,
3741 : * there would be a prunable tuple in the newbuf; but for now we choose
3742 : * not to optimize for aborts. Note that heap_xlog_update must be kept in
3743 : * sync if this decision changes.
3744 : */
3745 GIC 418868 : PageSetPrunable(page, xid);
3746 :
3747 418868 : if (use_hot_update)
3748 ECB : {
3749 : /* Mark the old tuple as HOT-updated */
3750 CBC 213765 : HeapTupleSetHotUpdated(&oldtup);
3751 ECB : /* And mark the new tuple as heap-only */
3752 GIC 213765 : HeapTupleSetHeapOnly(heaptup);
3753 ECB : /* Mark the caller's copy too, in case different from heaptup */
3754 GIC 213765 : HeapTupleSetHeapOnly(newtup);
3755 : }
3756 ECB : else
3757 : {
3758 : /* Make sure tuples are correctly marked as not-HOT */
3759 CBC 205103 : HeapTupleClearHotUpdated(&oldtup);
3760 GIC 205103 : HeapTupleClearHeapOnly(heaptup);
3761 CBC 205103 : HeapTupleClearHeapOnly(newtup);
3762 ECB : }
3763 :
3764 CBC 418868 : RelationPutHeapTuple(relation, newbuf, heaptup, false); /* insert new tuple */
3765 ECB :
3766 :
3767 : /* Clear obsolete visibility flags, possibly set by ourselves above... */
3768 CBC 418868 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3769 GIC 418868 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3770 : /* ... and store info about transaction updating this tuple */
3771 418868 : Assert(TransactionIdIsValid(xmax_old_tuple));
3772 418868 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_old_tuple);
3773 418868 : oldtup.t_data->t_infomask |= infomask_old_tuple;
3774 418868 : oldtup.t_data->t_infomask2 |= infomask2_old_tuple;
3775 418868 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
3776 ECB :
3777 : /* record address of new tuple in t_ctid of old one */
3778 GIC 418868 : oldtup.t_data->t_ctid = heaptup->t_self;
3779 :
3780 : /* clear PD_ALL_VISIBLE flags, reset all visibilitymap bits */
3781 418868 : if (PageIsAllVisible(BufferGetPage(buffer)))
3782 : {
3783 961 : all_visible_cleared = true;
3784 961 : PageClearAllVisible(BufferGetPage(buffer));
3785 CBC 961 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
3786 ECB : vmbuffer, VISIBILITYMAP_VALID_BITS);
3787 : }
3788 GIC 418868 : if (newbuf != buffer && PageIsAllVisible(BufferGetPage(newbuf)))
3789 : {
3790 362 : all_visible_cleared_new = true;
3791 CBC 362 : PageClearAllVisible(BufferGetPage(newbuf));
3792 362 : visibilitymap_clear(relation, BufferGetBlockNumber(newbuf),
3793 : vmbuffer_new, VISIBILITYMAP_VALID_BITS);
3794 : }
3795 :
3796 GIC 418868 : if (newbuf != buffer)
3797 196178 : MarkBufferDirty(newbuf);
3798 418868 : MarkBufferDirty(buffer);
3799 :
3800 : /* XLOG stuff */
3801 418868 : if (RelationNeedsWAL(relation))
3802 : {
3803 ECB : XLogRecPtr recptr;
3804 :
3805 : /*
3806 EUB : * For logical decoding we need combo CIDs to properly decode the
3807 : * catalog.
3808 : */
3809 GIC 407557 : if (RelationIsAccessibleInLogicalDecoding(relation))
3810 ECB : {
3811 CBC 1873 : log_heap_new_cid(relation, &oldtup);
3812 1873 : log_heap_new_cid(relation, heaptup);
3813 : }
3814 :
3815 GIC 407557 : recptr = log_heap_update(relation, buffer,
3816 : newbuf, &oldtup, heaptup,
3817 : old_key_tuple,
3818 : all_visible_cleared,
3819 : all_visible_cleared_new);
3820 407557 : if (newbuf != buffer)
3821 : {
3822 186055 : PageSetLSN(BufferGetPage(newbuf), recptr);
3823 : }
3824 407557 : PageSetLSN(BufferGetPage(buffer), recptr);
3825 : }
3826 :
3827 CBC 418868 : END_CRIT_SECTION();
3828 :
3829 GIC 418868 : if (newbuf != buffer)
3830 196178 : LockBuffer(newbuf, BUFFER_LOCK_UNLOCK);
3831 418868 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3832 :
3833 : /*
3834 ECB : * Mark old tuple for invalidation from system caches at next command
3835 : * boundary, and mark the new tuple for invalidation in case we abort. We
3836 : * have to do this before releasing the buffer because oldtup is in the
3837 : * buffer. (heaptup is all in local memory, but it's necessary to process
3838 : * both tuple versions in one call to inval.c so we can avoid redundant
3839 : * sinval messages.)
3840 : */
3841 CBC 418868 : CacheInvalidateHeapTuple(relation, &oldtup, heaptup);
3842 :
3843 : /* Now we can release the buffer(s) */
3844 GIC 418868 : if (newbuf != buffer)
3845 196178 : ReleaseBuffer(newbuf);
3846 418868 : ReleaseBuffer(buffer);
3847 418868 : if (BufferIsValid(vmbuffer_new))
3848 363 : ReleaseBuffer(vmbuffer_new);
3849 418868 : if (BufferIsValid(vmbuffer))
3850 961 : ReleaseBuffer(vmbuffer);
3851 :
3852 ECB : /*
3853 : * Release the lmgr tuple lock, if we had it.
3854 EUB : */
3855 CBC 418868 : if (have_tuple_lock)
3856 GIC 12 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
3857 :
3858 GNC 418868 : pgstat_count_heap_update(relation, use_hot_update, newbuf != buffer);
3859 :
3860 : /*
3861 : * If heaptup is a private copy, release it. Don't forget to copy t_self
3862 : * back to the caller's image, too.
3863 ECB : */
3864 GIC 418868 : if (heaptup != newtup)
3865 EUB : {
3866 GIC 3038 : newtup->t_self = heaptup->t_self;
3867 GBC 3038 : heap_freetuple(heaptup);
3868 EUB : }
3869 :
3870 : /*
3871 : * If it is a HOT update, the update may still need to update summarized
3872 : * indexes, lest we fail to update those summaries and get incorrect
3873 : * results (for example, minmax bounds of the block may change with this
3874 : * update).
3875 : */
3876 GNC 418868 : if (use_hot_update)
3877 : {
3878 213765 : if (summarized_update)
3879 1641 : *update_indexes = TU_Summarizing;
3880 : else
3881 212124 : *update_indexes = TU_None;
3882 : }
3883 : else
3884 205103 : *update_indexes = TU_All;
3885 :
3886 GIC 418868 : if (old_key_tuple != NULL && old_key_copied)
3887 141 : heap_freetuple(old_key_tuple);
3888 :
3889 418868 : bms_free(hot_attrs);
3890 GNC 418868 : bms_free(sum_attrs);
3891 GIC 418868 : bms_free(key_attrs);
3892 418868 : bms_free(id_attrs);
3893 418868 : bms_free(modified_attrs);
3894 418868 : bms_free(interesting_attrs);
3895 ECB :
3896 CBC 418868 : return TM_Ok;
3897 : }
3898 ECB :
3899 : /*
3900 : * Check if the specified attribute's values are the same. Subroutine for
3901 : * HeapDetermineColumnsInfo.
3902 : */
3903 : static bool
3904 GIC 1234805 : heap_attr_equals(TupleDesc tupdesc, int attrnum, Datum value1, Datum value2,
3905 : bool isnull1, bool isnull2)
3906 : {
3907 : Form_pg_attribute att;
3908 :
3909 ECB : /*
3910 : * If one value is NULL and other is not, then they are certainly not
3911 : * equal
3912 : */
3913 GIC 1234805 : if (isnull1 != isnull2)
3914 3 : return false;
3915 :
3916 : /*
3917 ECB : * If both are NULL, they can be considered equal.
3918 : */
3919 CBC 1234802 : if (isnull1)
3920 GIC 4991 : return true;
3921 :
3922 ECB : /*
3923 : * We do simple binary comparison of the two datums. This may be overly
3924 : * strict because there can be multiple binary representations for the
3925 : * same logical value. But we should be OK as long as there are no false
3926 : * positives. Using a type-specific equality operator is messy because
3927 : * there could be multiple notions of equality in different operator
3928 : * classes; furthermore, we cannot safely invoke user-defined functions
3929 : * while holding exclusive buffer lock.
3930 : */
3931 GIC 1229811 : if (attrnum <= 0)
3932 : {
3933 : /* The only allowed system columns are OIDs, so do this */
3934 LBC 0 : return (DatumGetObjectId(value1) == DatumGetObjectId(value2));
3935 : }
3936 : else
3937 : {
3938 GIC 1229811 : Assert(attrnum <= tupdesc->natts);
3939 1229811 : att = TupleDescAttr(tupdesc, attrnum - 1);
3940 1229811 : return datumIsEqual(value1, value2, att->attbyval, att->attlen);
3941 ECB : }
3942 : }
3943 :
3944 : /*
3945 : * Check which columns are being updated.
3946 : *
3947 EUB : * Given an updated tuple, determine (and return into the output bitmapset),
3948 : * from those listed as interesting, the set of columns that changed.
3949 : *
3950 : * has_external indicates if any of the unmodified attributes (from those
3951 : * listed as interesting) of the old tuple is a member of external_cols and is
3952 ECB : * stored externally.
3953 EUB : */
3954 : static Bitmapset *
3955 GIC 418994 : HeapDetermineColumnsInfo(Relation relation,
3956 : Bitmapset *interesting_cols,
3957 EUB : Bitmapset *external_cols,
3958 : HeapTuple oldtup, HeapTuple newtup,
3959 : bool *has_external)
3960 : {
3961 : int attidx;
3962 GBC 418994 : Bitmapset *modified = NULL;
3963 GIC 418994 : TupleDesc tupdesc = RelationGetDescr(relation);
3964 :
3965 GNC 418994 : attidx = -1;
3966 1653799 : while ((attidx = bms_next_member(interesting_cols, attidx)) >= 0)
3967 : {
3968 : /* attidx is zero-based, attrnum is the normal attribute number */
3969 1234805 : AttrNumber attrnum = attidx + FirstLowInvalidHeapAttributeNumber;
3970 : Datum value1,
3971 : value2;
3972 : bool isnull1,
3973 : isnull2;
3974 :
3975 : /*
3976 : * If it's a whole-tuple reference, say "not equal". It's not really
3977 ECB : * worth supporting this case, since it could only succeed after a
3978 : * no-op update, which is hardly a case worth optimizing for.
3979 : */
3980 CBC 1234805 : if (attrnum == 0)
3981 : {
3982 UNC 0 : modified = bms_add_member(modified, attidx);
3983 GIC 1170399 : continue;
3984 ECB : }
3985 :
3986 : /*
3987 : * Likewise, automatically say "not equal" for any system attribute
3988 : * other than tableOID; we cannot expect these to be consistent in a
3989 : * HOT chain, or even to be set correctly yet in the new tuple.
3990 : */
3991 GIC 1234805 : if (attrnum < 0)
3992 : {
3993 UIC 0 : if (attrnum != TableOidAttributeNumber)
3994 : {
3995 UNC 0 : modified = bms_add_member(modified, attidx);
3996 UIC 0 : continue;
3997 : }
3998 : }
3999 :
4000 : /*
4001 : * Extract the corresponding values. XXX this is pretty inefficient
4002 : * if there are many indexed columns. Should we do a single
4003 : * heap_deform_tuple call on each tuple, instead? But that doesn't
4004 : * work for system columns ...
4005 : */
4006 GIC 1234805 : value1 = heap_getattr(oldtup, attrnum, tupdesc, &isnull1);
4007 1234805 : value2 = heap_getattr(newtup, attrnum, tupdesc, &isnull2);
4008 :
4009 1234805 : if (!heap_attr_equals(tupdesc, attrnum, value1,
4010 : value2, isnull1, isnull2))
4011 : {
4012 GNC 18483 : modified = bms_add_member(modified, attidx);
4013 GIC 18483 : continue;
4014 : }
4015 ECB :
4016 : /*
4017 : * No need to check attributes that can't be stored externally. Note
4018 : * that system attributes can't be stored externally.
4019 : */
4020 GIC 1216322 : if (attrnum < 0 || isnull1 ||
4021 CBC 1211331 : TupleDescAttr(tupdesc, attrnum - 1)->attlen != -1)
4022 GIC 1151916 : continue;
4023 :
4024 ECB : /*
4025 : * Check if the old tuple's attribute is stored externally and is a
4026 : * member of external_cols.
4027 : */
4028 GIC 64411 : if (VARATT_IS_EXTERNAL((struct varlena *) DatumGetPointer(value1)) &&
4029 GNC 5 : bms_is_member(attidx, external_cols))
4030 CBC 2 : *has_external = true;
4031 ECB : }
4032 :
4033 CBC 418994 : return modified;
4034 : }
4035 ECB :
4036 : /*
4037 : * simple_heap_update - replace a tuple
4038 : *
4039 : * This routine may be used to update a tuple when concurrent updates of
4040 : * the target tuple are not expected (for example, because we have a lock
4041 : * on the relation associated with the tuple). Any failure is reported
4042 : * via ereport().
4043 : */
4044 : void
4045 GNC 201216 : simple_heap_update(Relation relation, ItemPointer otid, HeapTuple tup,
4046 : TU_UpdateIndexes *update_indexes)
4047 : {
4048 ECB : TM_Result result;
4049 : TM_FailureData tmfd;
4050 : LockTupleMode lockmode;
4051 :
4052 CBC 201216 : result = heap_update(relation, otid, tup,
4053 : GetCurrentCommandId(true), InvalidSnapshot,
4054 ECB : true /* wait for commit */ ,
4055 : &tmfd, &lockmode, update_indexes);
4056 CBC 201216 : switch (result)
4057 : {
4058 LBC 0 : case TM_SelfModified:
4059 ECB : /* Tuple was already updated in current command? */
4060 UIC 0 : elog(ERROR, "tuple already updated by self");
4061 ECB : break;
4062 :
4063 GIC 201216 : case TM_Ok:
4064 : /* done successfully */
4065 201216 : break;
4066 :
4067 UIC 0 : case TM_Updated:
4068 0 : elog(ERROR, "tuple concurrently updated");
4069 ECB : break;
4070 :
4071 UIC 0 : case TM_Deleted:
4072 LBC 0 : elog(ERROR, "tuple concurrently deleted");
4073 ECB : break;
4074 :
4075 UIC 0 : default:
4076 0 : elog(ERROR, "unrecognized heap_update status: %u", result);
4077 : break;
4078 : }
4079 GIC 201216 : }
4080 :
4081 :
4082 : /*
4083 ECB : * Return the MultiXactStatus corresponding to the given tuple lock mode.
4084 : */
4085 : static MultiXactStatus
4086 CBC 1188 : get_mxact_status_for_lock(LockTupleMode mode, bool is_update)
4087 : {
4088 ECB : int retval;
4089 :
4090 GIC 1188 : if (is_update)
4091 96 : retval = tupleLockExtraInfo[mode].updstatus;
4092 : else
4093 1092 : retval = tupleLockExtraInfo[mode].lockstatus;
4094 :
4095 1188 : if (retval == -1)
4096 UIC 0 : elog(ERROR, "invalid lock tuple mode %d/%s", mode,
4097 : is_update ? "true" : "false");
4098 :
4099 GIC 1188 : return (MultiXactStatus) retval;
4100 : }
4101 :
4102 ECB : /*
4103 : * heap_lock_tuple - lock a tuple in shared or exclusive mode
4104 : *
4105 : * Note that this acquires a buffer pin, which the caller must release.
4106 : *
4107 : * Input parameters:
4108 : * relation: relation containing tuple (caller must hold suitable lock)
4109 : * tid: TID of tuple to lock
4110 : * cid: current command ID (used for visibility test, and stored into
4111 : * tuple's cmax if lock is successful)
4112 : * mode: indicates if shared or exclusive tuple lock is desired
4113 : * wait_policy: what to do if tuple lock is not available
4114 : * follow_updates: if true, follow the update chain to also lock descendant
4115 : * tuples.
4116 : *
4117 : * Output parameters:
4118 : * *tuple: all fields filled in
4119 : * *buffer: set to buffer holding tuple (pinned but not locked at exit)
4120 : * *tmfd: filled in failure cases (see below)
4121 : *
4122 : * Function results are the same as the ones for table_tuple_lock().
4123 : *
4124 : * In the failure cases other than TM_Invisible, the routine fills
4125 : * *tmfd with the tuple's t_ctid, t_xmax (resolving a possible MultiXact,
4126 : * if necessary), and t_cmax (the last only for TM_SelfModified,
4127 : * since we cannot obtain cmax from a combo CID generated by another
4128 : * transaction).
4129 : * See comments for struct TM_FailureData for additional info.
4130 : *
4131 : * See README.tuplock for a thorough explanation of this mechanism.
4132 : */
4133 : TM_Result
4134 GIC 82482 : heap_lock_tuple(Relation relation, HeapTuple tuple,
4135 : CommandId cid, LockTupleMode mode, LockWaitPolicy wait_policy,
4136 : bool follow_updates,
4137 : Buffer *buffer, TM_FailureData *tmfd)
4138 : {
4139 : TM_Result result;
4140 82482 : ItemPointer tid = &(tuple->t_self);
4141 : ItemId lp;
4142 : Page page;
4143 82482 : Buffer vmbuffer = InvalidBuffer;
4144 : BlockNumber block;
4145 : TransactionId xid,
4146 ECB : xmax;
4147 : uint16 old_infomask,
4148 : new_infomask,
4149 : new_infomask2;
4150 CBC 82482 : bool first_time = true;
4151 82482 : bool skip_tuple_lock = false;
4152 GIC 82482 : bool have_tuple_lock = false;
4153 CBC 82482 : bool cleared_all_frozen = false;
4154 :
4155 82482 : *buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
4156 GIC 82482 : block = ItemPointerGetBlockNumber(tid);
4157 ECB :
4158 : /*
4159 : * Before locking the buffer, pin the visibility map page if it appears to
4160 : * be necessary. Since we haven't got the lock yet, someone else might be
4161 : * in the middle of changing this, so we'll need to recheck after we have
4162 : * the lock.
4163 : */
4164 CBC 82482 : if (PageIsAllVisible(BufferGetPage(*buffer)))
4165 1651 : visibilitymap_pin(relation, block, &vmbuffer);
4166 :
4167 82482 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4168 ECB :
4169 GIC 82482 : page = BufferGetPage(*buffer);
4170 CBC 82482 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
4171 82482 : Assert(ItemIdIsNormal(lp));
4172 ECB :
4173 GIC 82482 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
4174 CBC 82482 : tuple->t_len = ItemIdGetLength(lp);
4175 82482 : tuple->t_tableOid = RelationGetRelid(relation);
4176 :
4177 14 : l3:
4178 82496 : result = HeapTupleSatisfiesUpdate(tuple, cid, *buffer);
4179 ECB :
4180 CBC 82496 : if (result == TM_Invisible)
4181 : {
4182 ECB : /*
4183 : * This is possible, but only when locking a tuple for ON CONFLICT
4184 : * UPDATE. We return this value here rather than throwing an error in
4185 : * order to give that case the opportunity to throw a more specific
4186 : * error.
4187 : */
4188 GIC 12 : result = TM_Invisible;
4189 12 : goto out_locked;
4190 : }
4191 82484 : else if (result == TM_BeingModified ||
4192 75528 : result == TM_Updated ||
4193 : result == TM_Deleted)
4194 : {
4195 ECB : TransactionId xwait;
4196 : uint16 infomask;
4197 : uint16 infomask2;
4198 : bool require_sleep;
4199 : ItemPointerData t_ctid;
4200 :
4201 : /* must copy state data before unlocking buffer */
4202 GIC 6958 : xwait = HeapTupleHeaderGetRawXmax(tuple->t_data);
4203 6958 : infomask = tuple->t_data->t_infomask;
4204 6958 : infomask2 = tuple->t_data->t_infomask2;
4205 6958 : ItemPointerCopy(&tuple->t_data->t_ctid, &t_ctid);
4206 :
4207 6958 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4208 :
4209 : /*
4210 : * If any subtransaction of the current top transaction already holds
4211 : * a lock as strong as or stronger than what we're requesting, we
4212 : * effectively hold the desired lock already. We *must* succeed
4213 : * without trying to take the tuple lock, else we will deadlock
4214 : * against anyone wanting to acquire a stronger lock.
4215 : *
4216 : * Note we only do this the first time we loop on the HTSU result;
4217 : * there is no point in testing in subsequent passes, because
4218 : * evidently our own transaction cannot have acquired a new lock after
4219 ECB : * the first time we checked.
4220 : */
4221 GIC 6958 : if (first_time)
4222 : {
4223 CBC 6949 : first_time = false;
4224 :
4225 GIC 6949 : if (infomask & HEAP_XMAX_IS_MULTI)
4226 : {
4227 : int i;
4228 : int nmembers;
4229 ECB : MultiXactMember *members;
4230 :
4231 : /*
4232 : * We don't need to allow old multixacts here; if that had
4233 : * been the case, HeapTupleSatisfiesUpdate would have returned
4234 : * MayBeUpdated and we wouldn't be here.
4235 : */
4236 : nmembers =
4237 GIC 84 : GetMultiXactIdMembers(xwait, &members, false,
4238 CBC 84 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
4239 :
4240 251 : for (i = 0; i < nmembers; i++)
4241 ECB : {
4242 : /* only consider members of our own transaction */
4243 GIC 181 : if (!TransactionIdIsCurrentTransactionId(members[i].xid))
4244 132 : continue;
4245 ECB :
4246 GIC 49 : if (TUPLOCK_from_mxstatus(members[i].status) >= mode)
4247 : {
4248 14 : pfree(members);
4249 14 : result = TM_Ok;
4250 14 : goto out_unlocked;
4251 : }
4252 : else
4253 : {
4254 ECB : /*
4255 : * Disable acquisition of the heavyweight tuple lock.
4256 : * Otherwise, when promoting a weaker lock, we might
4257 : * deadlock with another locker that has acquired the
4258 : * heavyweight tuple lock and is waiting for our
4259 : * transaction to finish.
4260 : *
4261 : * Note that in this case we still need to wait for
4262 : * the multixact if required, to avoid acquiring
4263 : * conflicting locks.
4264 : */
4265 GIC 35 : skip_tuple_lock = true;
4266 : }
4267 : }
4268 :
4269 70 : if (members)
4270 70 : pfree(members);
4271 ECB : }
4272 GIC 6865 : else if (TransactionIdIsCurrentTransactionId(xwait))
4273 : {
4274 5697 : switch (mode)
4275 : {
4276 123 : case LockTupleKeyShare:
4277 CBC 123 : Assert(HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) ||
4278 ECB : HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4279 : HEAP_XMAX_IS_EXCL_LOCKED(infomask));
4280 CBC 123 : result = TM_Ok;
4281 GIC 123 : goto out_unlocked;
4282 116 : case LockTupleShare:
4283 116 : if (HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4284 6 : HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4285 : {
4286 CBC 110 : result = TM_Ok;
4287 110 : goto out_unlocked;
4288 EUB : }
4289 CBC 6 : break;
4290 GIC 55 : case LockTupleNoKeyExclusive:
4291 55 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4292 ECB : {
4293 GIC 44 : result = TM_Ok;
4294 44 : goto out_unlocked;
4295 : }
4296 11 : break;
4297 5403 : case LockTupleExclusive:
4298 5403 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask) &&
4299 CBC 364 : infomask2 & HEAP_KEYS_UPDATED)
4300 : {
4301 349 : result = TM_Ok;
4302 GIC 349 : goto out_unlocked;
4303 : }
4304 5054 : break;
4305 : }
4306 : }
4307 : }
4308 ECB :
4309 : /*
4310 : * Initially assume that we will have to wait for the locking
4311 : * transaction(s) to finish. We check various cases below in which
4312 EUB : * this can be turned off.
4313 : */
4314 GIC 6318 : require_sleep = true;
4315 CBC 6318 : if (mode == LockTupleKeyShare)
4316 : {
4317 : /*
4318 ECB : * If we're requesting KeyShare, and there's no update present, we
4319 : * don't need to wait. Even if there is an update, we can still
4320 : * continue if the key hasn't been modified.
4321 : *
4322 : * However, if there are updates, we need to walk the update chain
4323 : * to mark future versions of the row as locked, too. That way,
4324 : * if somebody deletes that future version, we're protected
4325 : * against the key going away. This locking of future versions
4326 EUB : * could block momentarily, if a concurrent transaction is
4327 : * deleting a key; or it could return a value to the effect that
4328 ECB : * the transaction deleting the key has already committed. So we
4329 : * do this before re-locking the buffer; otherwise this would be
4330 : * prone to deadlocks.
4331 : *
4332 : * Note that the TID we're locking was grabbed before we unlocked
4333 : * the buffer. For it to change while we're not looking, the
4334 : * other properties we're testing for below after re-locking the
4335 : * buffer would also change, in which case we would restart this
4336 : * loop above.
4337 : */
4338 GIC 572 : if (!(infomask2 & HEAP_KEYS_UPDATED))
4339 : {
4340 : bool updated;
4341 :
4342 541 : updated = !HEAP_XMAX_IS_LOCKED_ONLY(infomask);
4343 :
4344 ECB : /*
4345 : * If there are updates, follow the update chain; bail out if
4346 : * that cannot be done.
4347 : */
4348 CBC 541 : if (follow_updates && updated)
4349 ECB : {
4350 : TM_Result res;
4351 :
4352 GBC 50 : res = heap_lock_updated_tuple(relation, tuple, &t_ctid,
4353 ECB : GetCurrentTransactionId(),
4354 : mode);
4355 GIC 50 : if (res != TM_Ok)
4356 : {
4357 6 : result = res;
4358 : /* recovery code expects to have buffer lock held */
4359 6 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4360 138 : goto failed;
4361 : }
4362 : }
4363 :
4364 535 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4365 :
4366 : /*
4367 : * Make sure it's still an appropriate lock, else start over.
4368 ECB : * Also, if it wasn't updated before we released the lock, but
4369 : * is updated now, we start over too; the reason is that we
4370 : * now need to follow the update chain to lock the new
4371 : * versions.
4372 : */
4373 CBC 535 : if (!HeapTupleHeaderIsOnlyLocked(tuple->t_data) &&
4374 GIC 43 : ((tuple->t_data->t_infomask2 & HEAP_KEYS_UPDATED) ||
4375 43 : !updated))
4376 14 : goto l3;
4377 :
4378 : /* Things look okay, so we can skip sleeping */
4379 535 : require_sleep = false;
4380 :
4381 : /*
4382 : * Note we allow Xmax to change here; other updaters/lockers
4383 : * could have modified it before we grabbed the buffer lock.
4384 : * However, this is not a problem, because with the recheck we
4385 ECB : * just did we ensure that they still don't conflict with the
4386 : * lock we want.
4387 : */
4388 : }
4389 : }
4390 GIC 5746 : else if (mode == LockTupleShare)
4391 : {
4392 : /*
4393 ECB : * If we're requesting Share, we can similarly avoid sleeping if
4394 : * there's no update and no exclusive lock present.
4395 : */
4396 CBC 439 : if (HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
4397 GIC 439 : !HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4398 : {
4399 CBC 433 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4400 :
4401 ECB : /*
4402 : * Make sure it's still an appropriate lock, else start over.
4403 : * See above about allowing xmax to change.
4404 : */
4405 GBC 433 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
4406 GIC 433 : HEAP_XMAX_IS_EXCL_LOCKED(tuple->t_data->t_infomask))
4407 UIC 0 : goto l3;
4408 CBC 433 : require_sleep = false;
4409 : }
4410 ECB : }
4411 CBC 5307 : else if (mode == LockTupleNoKeyExclusive)
4412 : {
4413 ECB : /*
4414 : * If we're requesting NoKeyExclusive, we might also be able to
4415 : * avoid sleeping; just ensure that there no conflicting lock
4416 : * already acquired.
4417 : */
4418 GIC 123 : if (infomask & HEAP_XMAX_IS_MULTI)
4419 ECB : {
4420 GIC 26 : if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
4421 ECB : mode, NULL))
4422 : {
4423 : /*
4424 EUB : * No conflict, but if the xmax changed under us in the
4425 ECB : * meantime, start over.
4426 : */
4427 GIC 13 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4428 13 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4429 CBC 13 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4430 : xwait))
4431 UIC 0 : goto l3;
4432 :
4433 : /* otherwise, we're good */
4434 GBC 13 : require_sleep = false;
4435 : }
4436 : }
4437 GIC 97 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
4438 : {
4439 15 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4440 :
4441 : /* if the xmax changed in the meantime, start over */
4442 15 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4443 15 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4444 : xwait))
4445 UIC 0 : goto l3;
4446 : /* otherwise, we're good */
4447 GIC 15 : require_sleep = false;
4448 : }
4449 : }
4450 ECB :
4451 : /*
4452 : * As a check independent from those above, we can also avoid sleeping
4453 : * if the current transaction is the sole locker of the tuple. Note
4454 : * that the strength of the lock already held is irrelevant; this is
4455 : * not about recording the lock in Xmax (which will be done regardless
4456 : * of this optimization, below). Also, note that the cases where we
4457 : * hold a lock stronger than we are requesting are already handled
4458 : * above by not doing anything.
4459 : *
4460 : * Note we only deal with the non-multixact case here; MultiXactIdWait
4461 : * is well equipped to deal with this situation on its own.
4462 : */
4463 GIC 11587 : if (require_sleep && !(infomask & HEAP_XMAX_IS_MULTI) &&
4464 GBC 5275 : TransactionIdIsCurrentTransactionId(xwait))
4465 ECB : {
4466 : /* ... but if the xmax changed in the meantime, start over */
4467 CBC 5054 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4468 GIC 5054 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4469 5054 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4470 : xwait))
4471 UBC 0 : goto l3;
4472 GIC 5054 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask));
4473 5054 : require_sleep = false;
4474 : }
4475 :
4476 ECB : /*
4477 : * Time to sleep on the other transaction/multixact, if necessary.
4478 : *
4479 : * If the other transaction is an update/delete that's already
4480 : * committed, then sleeping cannot possibly do any good: if we're
4481 : * required to sleep, get out to raise an error instead.
4482 : *
4483 : * By here, we either have already acquired the buffer exclusive lock,
4484 : * or we must wait for the locking transaction or multixact; so below
4485 : * we ensure that we grab buffer lock after the sleep.
4486 : */
4487 CBC 6312 : if (require_sleep && (result == TM_Updated || result == TM_Deleted))
4488 ECB : {
4489 GIC 94 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4490 94 : goto failed;
4491 : }
4492 CBC 6218 : else if (require_sleep)
4493 : {
4494 : /*
4495 : * Acquire tuple lock to establish our priority for the tuple, or
4496 : * die trying. LockTuple will release us when we are next-in-line
4497 : * for the tuple. We must do this even if we are share-locking,
4498 : * but not if we already have a weaker lock on the tuple.
4499 ECB : *
4500 : * If we are forced to "start over" below, we keep the tuple lock;
4501 : * this arranges that we stay at the head of the line while
4502 : * rechecking tuple state.
4503 : */
4504 CBC 168 : if (!skip_tuple_lock &&
4505 GIC 152 : !heap_acquire_tuplock(relation, tid, mode, wait_policy,
4506 : &have_tuple_lock))
4507 : {
4508 : /*
4509 : * This can only happen if wait_policy is Skip and the lock
4510 : * couldn't be obtained.
4511 : */
4512 1 : result = TM_WouldBlock;
4513 : /* recovery code expects to have buffer lock held */
4514 CBC 1 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4515 GIC 1 : goto failed;
4516 : }
4517 :
4518 166 : if (infomask & HEAP_XMAX_IS_MULTI)
4519 : {
4520 40 : MultiXactStatus status = get_mxact_status_for_lock(mode, false);
4521 :
4522 : /* We only ever lock tuples, never update them */
4523 40 : if (status >= MultiXactStatusNoKeyUpdate)
4524 UIC 0 : elog(ERROR, "invalid lock mode in heap_lock_tuple");
4525 ECB :
4526 : /* wait for multixact to end, or die trying */
4527 CBC 40 : switch (wait_policy)
4528 ECB : {
4529 CBC 36 : case LockWaitBlock:
4530 36 : MultiXactIdWait((MultiXactId) xwait, status, infomask,
4531 ECB : relation, &tuple->t_self, XLTW_Lock, NULL);
4532 GIC 36 : break;
4533 CBC 2 : case LockWaitSkip:
4534 GIC 2 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
4535 : status, infomask, relation,
4536 ECB : NULL))
4537 : {
4538 GIC 2 : result = TM_WouldBlock;
4539 ECB : /* recovery code expects to have buffer lock held */
4540 GIC 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4541 2 : goto failed;
4542 : }
4543 UIC 0 : break;
4544 GIC 2 : case LockWaitError:
4545 2 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
4546 : status, infomask, relation,
4547 : NULL))
4548 2 : ereport(ERROR,
4549 ECB : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
4550 : errmsg("could not obtain lock on row in relation \"%s\"",
4551 : RelationGetRelationName(relation))));
4552 :
4553 LBC 0 : break;
4554 ECB : }
4555 :
4556 : /*
4557 : * Of course, the multixact might not be done here: if we're
4558 : * requesting a light lock mode, other transactions with light
4559 : * locks could still be alive, as well as locks owned by our
4560 : * own xact or other subxacts of this backend. We need to
4561 : * preserve the surviving MultiXact members. Note that it
4562 : * isn't absolutely necessary in the latter case, but doing so
4563 : * is simpler.
4564 : */
4565 : }
4566 : else
4567 : {
4568 : /* wait for regular transaction to end, or die trying */
4569 GIC 126 : switch (wait_policy)
4570 : {
4571 CBC 87 : case LockWaitBlock:
4572 GIC 87 : XactLockTableWait(xwait, relation, &tuple->t_self,
4573 EUB : XLTW_Lock);
4574 GBC 87 : break;
4575 33 : case LockWaitSkip:
4576 33 : if (!ConditionalXactLockTableWait(xwait))
4577 : {
4578 GIC 33 : result = TM_WouldBlock;
4579 ECB : /* recovery code expects to have buffer lock held */
4580 CBC 33 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4581 GIC 33 : goto failed;
4582 : }
4583 UIC 0 : break;
4584 GIC 6 : case LockWaitError:
4585 6 : if (!ConditionalXactLockTableWait(xwait))
4586 6 : ereport(ERROR,
4587 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
4588 : errmsg("could not obtain lock on row in relation \"%s\"",
4589 : RelationGetRelationName(relation))));
4590 LBC 0 : break;
4591 : }
4592 : }
4593 :
4594 : /* if there are updates, follow the update chain */
4595 GIC 123 : if (follow_updates && !HEAP_XMAX_IS_LOCKED_ONLY(infomask))
4596 : {
4597 ECB : TM_Result res;
4598 :
4599 GIC 38 : res = heap_lock_updated_tuple(relation, tuple, &t_ctid,
4600 : GetCurrentTransactionId(),
4601 ECB : mode);
4602 GIC 38 : if (res != TM_Ok)
4603 : {
4604 2 : result = res;
4605 : /* recovery code expects to have buffer lock held */
4606 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4607 2 : goto failed;
4608 : }
4609 : }
4610 :
4611 121 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4612 :
4613 ECB : /*
4614 : * xwait is done, but if xwait had just locked the tuple then some
4615 : * other xact could update this tuple before we get to this point.
4616 : * Check for xmax change, and start over if so.
4617 : */
4618 CBC 121 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4619 109 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4620 : xwait))
4621 GIC 14 : goto l3;
4622 :
4623 107 : if (!(infomask & HEAP_XMAX_IS_MULTI))
4624 : {
4625 : /*
4626 : * Otherwise check if it committed or aborted. Note we cannot
4627 : * be here if the tuple was only locked by somebody who didn't
4628 ECB : * conflict with us; that would have been handled above. So
4629 : * that transaction must necessarily be gone by now. But
4630 : * don't check for this in the multixact case, because some
4631 : * locker transactions might still be running.
4632 : */
4633 CBC 76 : UpdateXmaxHintBits(tuple->t_data, *buffer, xwait);
4634 : }
4635 ECB : }
4636 :
4637 : /* By here, we're certain that we hold buffer exclusive lock again */
4638 :
4639 : /*
4640 : * We may lock if previous xmax aborted, or if it committed but only
4641 : * locked the tuple without updating it; or if we didn't have to wait
4642 : * at all for whatever reason.
4643 : */
4644 GIC 6157 : if (!require_sleep ||
4645 107 : (tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
4646 139 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
4647 62 : HeapTupleHeaderIsOnlyLocked(tuple->t_data))
4648 6101 : result = TM_Ok;
4649 56 : else if (!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
4650 45 : result = TM_Updated;
4651 : else
4652 CBC 11 : result = TM_Deleted;
4653 : }
4654 :
4655 GIC 75526 : failed:
4656 81821 : if (result != TM_Ok)
4657 ECB : {
4658 CBC 200 : Assert(result == TM_SelfModified || result == TM_Updated ||
4659 : result == TM_Deleted || result == TM_WouldBlock);
4660 ECB :
4661 : /*
4662 : * When locking a tuple under LockWaitSkip semantics and we fail with
4663 : * TM_WouldBlock above, it's possible for concurrent transactions to
4664 : * release the lock and set HEAP_XMAX_INVALID in the meantime. So
4665 : * this assert is slightly different from the equivalent one in
4666 : * heap_delete and heap_update.
4667 : */
4668 GIC 200 : Assert((result == TM_WouldBlock) ||
4669 ECB : !(tuple->t_data->t_infomask & HEAP_XMAX_INVALID));
4670 GIC 200 : Assert(result != TM_Updated ||
4671 ECB : !ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid));
4672 GIC 200 : tmfd->ctid = tuple->t_data->t_ctid;
4673 200 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
4674 CBC 200 : if (result == TM_SelfModified)
4675 GIC 6 : tmfd->cmax = HeapTupleHeaderGetCmax(tuple->t_data);
4676 ECB : else
4677 GIC 194 : tmfd->cmax = InvalidCommandId;
4678 CBC 200 : goto out_locked;
4679 ECB : }
4680 :
4681 : /*
4682 : * If we didn't pin the visibility map page and the page has become all
4683 : * visible while we were busy locking the buffer, or during some
4684 : * subsequent window during which we had it unlocked, we'll have to unlock
4685 : * and re-lock, to avoid holding the buffer lock across I/O. That's a bit
4686 : * unfortunate, especially since we'll now have to recheck whether the
4687 : * tuple has been locked or updated under us, but hopefully it won't
4688 : * happen very often.
4689 : */
4690 GIC 81621 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
4691 : {
4692 UIC 0 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4693 0 : visibilitymap_pin(relation, block, &vmbuffer);
4694 LBC 0 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4695 0 : goto l3;
4696 : }
4697 ECB :
4698 GIC 81621 : xmax = HeapTupleHeaderGetRawXmax(tuple->t_data);
4699 81621 : old_infomask = tuple->t_data->t_infomask;
4700 :
4701 : /*
4702 : * If this is the first possibly-multixact-able operation in the current
4703 : * transaction, set my per-backend OldestMemberMXactId setting. We can be
4704 : * certain that the transaction will never become a member of any older
4705 : * MultiXactIds than that. (We have to do this even if we end up just
4706 : * using our own TransactionId below, since some other backend could
4707 : * incorporate our XID into a MultiXact immediately afterwards.)
4708 : */
4709 81621 : MultiXactIdSetOldestMember();
4710 :
4711 : /*
4712 : * Compute the new xmax and infomask to store into the tuple. Note we do
4713 ECB : * not modify the tuple just yet, because that would leave it in the wrong
4714 : * state if multixact.c elogs.
4715 : */
4716 CBC 81621 : compute_new_xmax_infomask(xmax, old_infomask, tuple->t_data->t_infomask2,
4717 ECB : GetCurrentTransactionId(), mode, false,
4718 : &xid, &new_infomask, &new_infomask2);
4719 :
4720 GIC 81621 : START_CRIT_SECTION();
4721 ECB :
4722 : /*
4723 : * Store transaction information of xact locking the tuple.
4724 : *
4725 : * Note: Cmax is meaningless in this context, so don't set it; this avoids
4726 : * possibly generating a useless combo CID. Moreover, if we're locking a
4727 : * previously updated tuple, it's important to preserve the Cmax.
4728 : *
4729 : * Also reset the HOT UPDATE bit, but only if there's no update; otherwise
4730 : * we would break the HOT chain.
4731 : */
4732 CBC 81621 : tuple->t_data->t_infomask &= ~HEAP_XMAX_BITS;
4733 GIC 81621 : tuple->t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
4734 81621 : tuple->t_data->t_infomask |= new_infomask;
4735 81621 : tuple->t_data->t_infomask2 |= new_infomask2;
4736 CBC 81621 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
4737 GIC 81582 : HeapTupleHeaderClearHotUpdated(tuple->t_data);
4738 CBC 81621 : HeapTupleHeaderSetXmax(tuple->t_data, xid);
4739 :
4740 ECB : /*
4741 : * Make sure there is no forward chain link in t_ctid. Note that in the
4742 : * cases where the tuple has been updated, we must not overwrite t_ctid,
4743 : * because it was set by the updater. Moreover, if the tuple has been
4744 : * updated, we need to follow the update chain to lock the new versions of
4745 : * the tuple as well.
4746 : */
4747 GIC 81621 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
4748 81582 : tuple->t_data->t_ctid = *tid;
4749 :
4750 : /* Clear only the all-frozen bit on visibility map if needed */
4751 83272 : if (PageIsAllVisible(page) &&
4752 1651 : visibilitymap_clear(relation, block, vmbuffer,
4753 : VISIBILITYMAP_ALL_FROZEN))
4754 15 : cleared_all_frozen = true;
4755 :
4756 :
4757 81621 : MarkBufferDirty(*buffer);
4758 :
4759 : /*
4760 : * XLOG stuff. You might think that we don't need an XLOG record because
4761 : * there is no state change worth restoring after a crash. You would be
4762 ECB : * wrong however: we have just written either a TransactionId or a
4763 : * MultiXactId that may never have been seen on disk before, and we need
4764 : * to make sure that there are XLOG entries covering those ID numbers.
4765 : * Else the same IDs might be re-used after a crash, which would be
4766 : * disastrous if this page made it to disk before the crash. Essentially
4767 : * we have to enforce the WAL log-before-data rule even in this case.
4768 : * (Also, in a PITR log-shipping or 2PC environment, we have to have XLOG
4769 : * entries for everything anyway.)
4770 : */
4771 GIC 81621 : if (RelationNeedsWAL(relation))
4772 ECB : {
4773 : xl_heap_lock xlrec;
4774 : XLogRecPtr recptr;
4775 :
4776 CBC 81066 : XLogBeginInsert();
4777 81066 : XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD);
4778 :
4779 GIC 81066 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
4780 81066 : xlrec.locking_xid = xid;
4781 162132 : xlrec.infobits_set = compute_infobits(new_infomask,
4782 81066 : tuple->t_data->t_infomask2);
4783 81066 : xlrec.flags = cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
4784 81066 : XLogRegisterData((char *) &xlrec, SizeOfHeapLock);
4785 :
4786 : /* we don't decode row locks atm, so no need to log the origin */
4787 ECB :
4788 GIC 81066 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
4789 ECB :
4790 CBC 81066 : PageSetLSN(page, recptr);
4791 ECB : }
4792 :
4793 GIC 81621 : END_CRIT_SECTION();
4794 :
4795 CBC 81621 : result = TM_Ok;
4796 ECB :
4797 GIC 81833 : out_locked:
4798 CBC 81833 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4799 ECB :
4800 CBC 82473 : out_unlocked:
4801 82473 : if (BufferIsValid(vmbuffer))
4802 1651 : ReleaseBuffer(vmbuffer);
4803 ECB :
4804 : /*
4805 : * Don't update the visibility map here. Locking a tuple doesn't change
4806 : * visibility info.
4807 : */
4808 :
4809 : /*
4810 : * Now that we have successfully marked the tuple as locked, we can
4811 : * release the lmgr tuple lock, if we had it.
4812 : */
4813 CBC 82473 : if (have_tuple_lock)
4814 137 : UnlockTupleTuplock(relation, tid, mode);
4815 EUB :
4816 GBC 82473 : return result;
4817 EUB : }
4818 :
4819 : /*
4820 : * Acquire heavyweight lock on the given tuple, in preparation for acquiring
4821 ECB : * its normal, Xmax-based tuple lock.
4822 : *
4823 : * have_tuple_lock is an input and output parameter: on input, it indicates
4824 : * whether the lock has previously been acquired (and this function does
4825 : * nothing in that case). If this function returns success, have_tuple_lock
4826 : * has been flipped to true.
4827 : *
4828 : * Returns false if it was unable to obtain the lock; this can only happen if
4829 : * wait_policy is Skip.
4830 : */
4831 : static bool
4832 GIC 246 : heap_acquire_tuplock(Relation relation, ItemPointer tid, LockTupleMode mode,
4833 : LockWaitPolicy wait_policy, bool *have_tuple_lock)
4834 : {
4835 246 : if (*have_tuple_lock)
4836 9 : return true;
4837 :
4838 CBC 237 : switch (wait_policy)
4839 : {
4840 GBC 196 : case LockWaitBlock:
4841 196 : LockTupleTuplock(relation, tid, mode);
4842 196 : break;
4843 :
4844 GIC 34 : case LockWaitSkip:
4845 34 : if (!ConditionalLockTupleTuplock(relation, tid, mode))
4846 1 : return false;
4847 33 : break;
4848 :
4849 7 : case LockWaitError:
4850 7 : if (!ConditionalLockTupleTuplock(relation, tid, mode))
4851 1 : ereport(ERROR,
4852 : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
4853 : errmsg("could not obtain lock on row in relation \"%s\"",
4854 : RelationGetRelationName(relation))));
4855 6 : break;
4856 : }
4857 CBC 235 : *have_tuple_lock = true;
4858 :
4859 235 : return true;
4860 ECB : }
4861 :
4862 : /*
4863 : * Given an original set of Xmax and infomask, and a transaction (identified by
4864 : * add_to_xmax) acquiring a new lock of some mode, compute the new Xmax and
4865 : * corresponding infomasks to use on the tuple.
4866 : *
4867 : * Note that this might have side effects such as creating a new MultiXactId.
4868 : *
4869 : * Most callers will have called HeapTupleSatisfiesUpdate before this function;
4870 : * that will have set the HEAP_XMAX_INVALID bit if the xmax was a MultiXactId
4871 : * but it was not running anymore. There is a race condition, which is that the
4872 : * MultiXactId may have finished since then, but that uncommon case is handled
4873 : * either here, or within MultiXactIdExpand.
4874 : *
4875 : * There is a similar race condition possible when the old xmax was a regular
4876 : * TransactionId. We test TransactionIdIsInProgress again just to narrow the
4877 : * window, but it's still possible to end up creating an unnecessary
4878 : * MultiXactId. Fortunately this is harmless.
4879 : */
4880 : static void
4881 GIC 2117116 : compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
4882 : uint16 old_infomask2, TransactionId add_to_xmax,
4883 : LockTupleMode mode, bool is_update,
4884 : TransactionId *result_xmax, uint16 *result_infomask,
4885 : uint16 *result_infomask2)
4886 : {
4887 : TransactionId new_xmax;
4888 ECB : uint16 new_infomask,
4889 EUB : new_infomask2;
4890 :
4891 CBC 2117116 : Assert(TransactionIdIsCurrentTransactionId(add_to_xmax));
4892 :
4893 2220971 : l5:
4894 GIC 2220971 : new_infomask = 0;
4895 2220971 : new_infomask2 = 0;
4896 2220971 : if (old_infomask & HEAP_XMAX_INVALID)
4897 : {
4898 : /*
4899 : * No previous locker; we just insert our own TransactionId.
4900 ECB : *
4901 : * Note that it's critical that this case be the first one checked,
4902 : * because there are several blocks below that come back to this one
4903 : * to implement certain optimizations; old_infomask might contain
4904 : * other dirty bits in those cases, but we don't really care.
4905 : */
4906 GIC 2116000 : if (is_update)
4907 : {
4908 1838882 : new_xmax = add_to_xmax;
4909 1838882 : if (mode == LockTupleExclusive)
4910 1455536 : new_infomask2 |= HEAP_KEYS_UPDATED;
4911 : }
4912 : else
4913 : {
4914 CBC 277118 : new_infomask |= HEAP_XMAX_LOCK_ONLY;
4915 GIC 277118 : switch (mode)
4916 ECB : {
4917 CBC 2202 : case LockTupleKeyShare:
4918 2202 : new_xmax = add_to_xmax;
4919 2202 : new_infomask |= HEAP_XMAX_KEYSHR_LOCK;
4920 2202 : break;
4921 GIC 706 : case LockTupleShare:
4922 CBC 706 : new_xmax = add_to_xmax;
4923 706 : new_infomask |= HEAP_XMAX_SHR_LOCK;
4924 GIC 706 : break;
4925 CBC 179670 : case LockTupleNoKeyExclusive:
4926 GIC 179670 : new_xmax = add_to_xmax;
4927 179670 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
4928 179670 : break;
4929 94540 : case LockTupleExclusive:
4930 94540 : new_xmax = add_to_xmax;
4931 94540 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
4932 94540 : new_infomask2 |= HEAP_KEYS_UPDATED;
4933 94540 : break;
4934 UIC 0 : default:
4935 UBC 0 : new_xmax = InvalidTransactionId; /* silence compiler */
4936 0 : elog(ERROR, "invalid lock mode");
4937 EUB : }
4938 : }
4939 : }
4940 GIC 104971 : else if (old_infomask & HEAP_XMAX_IS_MULTI)
4941 : {
4942 : MultiXactStatus new_status;
4943 :
4944 ECB : /*
4945 EUB : * Currently we don't allow XMAX_COMMITTED to be set for multis, so
4946 : * cross-check.
4947 ECB : */
4948 GIC 120 : Assert(!(old_infomask & HEAP_XMAX_COMMITTED));
4949 :
4950 ECB : /*
4951 : * A multixact together with LOCK_ONLY set but neither lock bit set
4952 : * (i.e. a pg_upgraded share locked tuple) cannot possibly be running
4953 : * anymore. This check is critical for databases upgraded by
4954 : * pg_upgrade; both MultiXactIdIsRunning and MultiXactIdExpand assume
4955 : * that such multis are never passed.
4956 : */
4957 CBC 120 : if (HEAP_LOCKED_UPGRADED(old_infomask))
4958 : {
4959 UIC 0 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
4960 0 : old_infomask |= HEAP_XMAX_INVALID;
4961 0 : goto l5;
4962 : }
4963 :
4964 : /*
4965 : * If the XMAX is already a MultiXactId, then we need to expand it to
4966 : * include add_to_xmax; but if all the members were lockers and are
4967 : * all gone, we can do away with the IS_MULTI bit and just set
4968 ECB : * add_to_xmax as the only locker/updater. If all lockers are gone
4969 : * and we have an updater that aborted, we can also do without a
4970 : * multi.
4971 : *
4972 : * The cost of doing GetMultiXactIdMembers would be paid by
4973 : * MultiXactIdExpand if we weren't to do this, so this check is not
4974 : * incurring extra work anyhow.
4975 : */
4976 CBC 120 : if (!MultiXactIdIsRunning(xmax, HEAP_XMAX_IS_LOCKED_ONLY(old_infomask)))
4977 : {
4978 GIC 23 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) ||
4979 8 : !TransactionIdDidCommit(MultiXactIdGetUpdateXid(xmax,
4980 ECB : old_infomask)))
4981 : {
4982 : /*
4983 : * Reset these bits and restart; otherwise fall through to
4984 : * create a new multi below.
4985 : */
4986 CBC 23 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
4987 23 : old_infomask |= HEAP_XMAX_INVALID;
4988 GIC 23 : goto l5;
4989 : }
4990 : }
4991 :
4992 97 : new_status = get_mxact_status_for_lock(mode, is_update);
4993 :
4994 97 : new_xmax = MultiXactIdExpand((MultiXactId) xmax, add_to_xmax,
4995 ECB : new_status);
4996 GBC 97 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
4997 : }
4998 CBC 104851 : else if (old_infomask & HEAP_XMAX_COMMITTED)
4999 : {
5000 ECB : /*
5001 : * It's a committed update, so we need to preserve him as updater of
5002 : * the tuple.
5003 : */
5004 : MultiXactStatus status;
5005 : MultiXactStatus new_status;
5006 :
5007 CBC 13 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5008 LBC 0 : status = MultiXactStatusUpdate;
5009 : else
5010 GIC 13 : status = MultiXactStatusNoKeyUpdate;
5011 :
5012 13 : new_status = get_mxact_status_for_lock(mode, is_update);
5013 :
5014 : /*
5015 : * since it's not running, it's obviously impossible for the old
5016 : * updater to be identical to the current one, so we need not check
5017 : * for that case as we do in the block above.
5018 ECB : */
5019 CBC 13 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5020 GIC 13 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5021 : }
5022 CBC 104838 : else if (TransactionIdIsInProgress(xmax))
5023 ECB : {
5024 : /*
5025 : * If the XMAX is a valid, in-progress TransactionId, then we need to
5026 : * create a new MultiXactId that includes both the old locker or
5027 : * updater and our own TransactionId.
5028 : */
5029 : MultiXactStatus new_status;
5030 : MultiXactStatus old_status;
5031 : LockTupleMode old_mode;
5032 :
5033 GIC 104829 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5034 : {
5035 104803 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5036 5603 : old_status = MultiXactStatusForKeyShare;
5037 99200 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5038 429 : old_status = MultiXactStatusForShare;
5039 98771 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5040 : {
5041 98771 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5042 92666 : old_status = MultiXactStatusForUpdate;
5043 ECB : else
5044 GIC 6105 : old_status = MultiXactStatusForNoKeyUpdate;
5045 : }
5046 : else
5047 : {
5048 : /*
5049 ECB : * LOCK_ONLY can be present alone only when a page has been
5050 : * upgraded by pg_upgrade. But in that case,
5051 : * TransactionIdIsInProgress() should have returned false. We
5052 : * assume it's no longer locked in this case.
5053 : */
5054 UIC 0 : elog(WARNING, "LOCK_ONLY found for Xid in progress %u", xmax);
5055 0 : old_infomask |= HEAP_XMAX_INVALID;
5056 0 : old_infomask &= ~HEAP_XMAX_LOCK_ONLY;
5057 LBC 0 : goto l5;
5058 : }
5059 : }
5060 : else
5061 : {
5062 : /* it's an update, but which kind? */
5063 GIC 26 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5064 UBC 0 : old_status = MultiXactStatusUpdate;
5065 : else
5066 CBC 26 : old_status = MultiXactStatusNoKeyUpdate;
5067 : }
5068 :
5069 GIC 104829 : old_mode = TUPLOCK_from_mxstatus(old_status);
5070 :
5071 : /*
5072 : * If the lock to be acquired is for the same TransactionId as the
5073 : * existing lock, there's an optimization possible: consider only the
5074 ECB : * strongest of both locks as the only one present, and restart.
5075 : */
5076 GIC 104829 : if (xmax == add_to_xmax)
5077 ECB : {
5078 : /*
5079 : * Note that it's not possible for the original tuple to be
5080 : * updated: we wouldn't be here because the tuple would have been
5081 : * invisible and we wouldn't try to update it. As a subtlety,
5082 : * this code can also run when traversing an update chain to lock
5083 : * future versions of a tuple. But we wouldn't be here either,
5084 : * because the add_to_xmax would be different from the original
5085 : * updater.
5086 : */
5087 CBC 103824 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5088 ECB :
5089 : /* acquire the strongest of both */
5090 GIC 103824 : if (mode < old_mode)
5091 52130 : mode = old_mode;
5092 : /* mustn't touch is_update */
5093 :
5094 103824 : old_infomask |= HEAP_XMAX_INVALID;
5095 103824 : goto l5;
5096 : }
5097 :
5098 : /* otherwise, just fall back to creating a new multixact */
5099 1005 : new_status = get_mxact_status_for_lock(mode, is_update);
5100 1005 : new_xmax = MultiXactIdCreate(xmax, old_status,
5101 : add_to_xmax, new_status);
5102 1005 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5103 : }
5104 14 : else if (!HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) &&
5105 CBC 5 : TransactionIdDidCommit(xmax))
5106 1 : {
5107 : /*
5108 ECB : * It's a committed update, so we gotta preserve him as updater of the
5109 : * tuple.
5110 : */
5111 : MultiXactStatus status;
5112 : MultiXactStatus new_status;
5113 :
5114 GIC 1 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5115 LBC 0 : status = MultiXactStatusUpdate;
5116 : else
5117 GIC 1 : status = MultiXactStatusNoKeyUpdate;
5118 ECB :
5119 GIC 1 : new_status = get_mxact_status_for_lock(mode, is_update);
5120 :
5121 : /*
5122 EUB : * since it's not running, it's obviously impossible for the old
5123 : * updater to be identical to the current one, so we need not check
5124 : * for that case as we do in the block above.
5125 : */
5126 GIC 1 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5127 1 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5128 : }
5129 : else
5130 : {
5131 : /*
5132 : * Can get here iff the locking/updating transaction was running when
5133 : * the infomask was extracted from the tuple, but finished before
5134 ECB : * TransactionIdIsInProgress got to run. Deal with it as if there was
5135 : * no locker at all in the first place.
5136 : */
5137 GIC 8 : old_infomask |= HEAP_XMAX_INVALID;
5138 8 : goto l5;
5139 : }
5140 :
5141 2117116 : *result_infomask = new_infomask;
5142 2117116 : *result_infomask2 = new_infomask2;
5143 2117116 : *result_xmax = new_xmax;
5144 2117116 : }
5145 :
5146 : /*
5147 ECB : * Subroutine for heap_lock_updated_tuple_rec.
5148 : *
5149 : * Given a hypothetical multixact status held by the transaction identified
5150 : * with the given xid, does the current transaction need to wait, fail, or can
5151 : * it continue if it wanted to acquire a lock of the given mode? "needwait"
5152 : * is set to true if waiting is necessary; if it can continue, then TM_Ok is
5153 : * returned. If the lock is already held by the current transaction, return
5154 : * TM_SelfModified. In case of a conflict with another transaction, a
5155 : * different HeapTupleSatisfiesUpdate return code is returned.
5156 : *
5157 : * The held status is said to be hypothetical because it might correspond to a
5158 : * lock held by a single Xid, i.e. not a real MultiXactId; we express it this
5159 : * way for simplicity of API.
5160 : */
5161 : static TM_Result
5162 CBC 32 : test_lockmode_for_conflict(MultiXactStatus status, TransactionId xid,
5163 : LockTupleMode mode, HeapTuple tup,
5164 : bool *needwait)
5165 : {
5166 : MultiXactStatus wantedstatus;
5167 :
5168 GIC 32 : *needwait = false;
5169 32 : wantedstatus = get_mxact_status_for_lock(mode, false);
5170 :
5171 EUB : /*
5172 : * Note: we *must* check TransactionIdIsInProgress before
5173 : * TransactionIdDidAbort/Commit; see comment at top of heapam_visibility.c
5174 : * for an explanation.
5175 ECB : */
5176 CBC 32 : if (TransactionIdIsCurrentTransactionId(xid))
5177 : {
5178 : /*
5179 : * The tuple has already been locked by our own transaction. This is
5180 : * very rare but can happen if multiple transactions are trying to
5181 : * lock an ancient version of the same tuple.
5182 : */
5183 UIC 0 : return TM_SelfModified;
5184 ECB : }
5185 GIC 32 : else if (TransactionIdIsInProgress(xid))
5186 EUB : {
5187 : /*
5188 : * If the locking transaction is running, what we do depends on
5189 : * whether the lock modes conflict: if they do, then we must wait for
5190 ECB : * it to finish; otherwise we can fall through to lock this tuple
5191 : * version without waiting.
5192 : */
5193 GIC 16 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5194 16 : LOCKMODE_from_mxstatus(wantedstatus)))
5195 : {
5196 8 : *needwait = true;
5197 : }
5198 :
5199 : /*
5200 : * If we set needwait above, then this value doesn't matter;
5201 : * otherwise, this value signals to caller that it's okay to proceed.
5202 : */
5203 16 : return TM_Ok;
5204 : }
5205 CBC 16 : else if (TransactionIdDidAbort(xid))
5206 GIC 3 : return TM_Ok;
5207 GBC 13 : else if (TransactionIdDidCommit(xid))
5208 EUB : {
5209 : /*
5210 : * The other transaction committed. If it was only a locker, then the
5211 : * lock is completely gone now and we can return success; but if it
5212 : * was an update, then what we do depends on whether the two lock
5213 : * modes conflict. If they conflict, then we must report error to
5214 : * caller. But if they don't, we can fall through to allow the current
5215 : * transaction to lock the tuple.
5216 ECB : *
5217 : * Note: the reason we worry about ISUPDATE here is because as soon as
5218 : * a transaction ends, all its locks are gone and meaningless, and
5219 : * thus we can ignore them; whereas its updates persist. In the
5220 EUB : * TransactionIdIsInProgress case, above, we don't need to check
5221 : * because we know the lock is still "alive" and thus a conflict needs
5222 : * always be checked.
5223 : */
5224 GIC 13 : if (!ISUPDATE_from_mxstatus(status))
5225 4 : return TM_Ok;
5226 :
5227 9 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5228 9 : LOCKMODE_from_mxstatus(wantedstatus)))
5229 ECB : {
5230 : /* bummer */
5231 CBC 8 : if (!ItemPointerEquals(&tup->t_self, &tup->t_data->t_ctid))
5232 6 : return TM_Updated;
5233 : else
5234 GIC 2 : return TM_Deleted;
5235 ECB : }
5236 :
5237 CBC 1 : return TM_Ok;
5238 : }
5239 :
5240 : /* Not in progress, not aborted, not committed -- must have crashed */
5241 UIC 0 : return TM_Ok;
5242 : }
5243 :
5244 :
5245 ECB : /*
5246 : * Recursive part of heap_lock_updated_tuple
5247 : *
5248 : * Fetch the tuple pointed to by tid in rel, and mark it as locked by the given
5249 : * xid with the given mode; if this tuple is updated, recurse to lock the new
5250 : * version as well.
5251 : */
5252 : static TM_Result
5253 GIC 80 : heap_lock_updated_tuple_rec(Relation rel, ItemPointer tid, TransactionId xid,
5254 : LockTupleMode mode)
5255 : {
5256 : TM_Result result;
5257 : ItemPointerData tupid;
5258 : HeapTupleData mytup;
5259 : Buffer buf;
5260 : uint16 new_infomask,
5261 : new_infomask2,
5262 : old_infomask,
5263 : old_infomask2;
5264 : TransactionId xmax,
5265 : new_xmax;
5266 CBC 80 : TransactionId priorXmax = InvalidTransactionId;
5267 GIC 80 : bool cleared_all_frozen = false;
5268 ECB : bool pinned_desired_page;
5269 CBC 80 : Buffer vmbuffer = InvalidBuffer;
5270 ECB : BlockNumber block;
5271 :
5272 CBC 80 : ItemPointerCopy(tid, &tupid);
5273 ECB :
5274 : for (;;)
5275 : {
5276 GIC 83 : new_infomask = 0;
5277 83 : new_xmax = InvalidTransactionId;
5278 83 : block = ItemPointerGetBlockNumber(&tupid);
5279 83 : ItemPointerCopy(&tupid, &(mytup.t_self));
5280 :
5281 83 : if (!heap_fetch(rel, SnapshotAny, &mytup, &buf, false))
5282 : {
5283 : /*
5284 : * if we fail to find the updated version of the tuple, it's
5285 : * because it was vacuumed/pruned away after its creator
5286 : * transaction aborted. So behave as if we got to the end of the
5287 : * chain, and there's no further tuple to lock: return success to
5288 ECB : * caller.
5289 : */
5290 UBC 0 : result = TM_Ok;
5291 0 : goto out_unlocked;
5292 : }
5293 :
5294 CBC 83 : l4:
5295 GIC 91 : CHECK_FOR_INTERRUPTS();
5296 EUB :
5297 : /*
5298 : * Before locking the buffer, pin the visibility map page if it
5299 : * appears to be necessary. Since we haven't got the lock yet,
5300 : * someone else might be in the middle of changing this, so we'll need
5301 : * to recheck after we have the lock.
5302 : */
5303 CBC 91 : if (PageIsAllVisible(BufferGetPage(buf)))
5304 : {
5305 UBC 0 : visibilitymap_pin(rel, block, &vmbuffer);
5306 0 : pinned_desired_page = true;
5307 : }
5308 : else
5309 CBC 91 : pinned_desired_page = false;
5310 ECB :
5311 GIC 91 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5312 :
5313 : /*
5314 : * If we didn't pin the visibility map page and the page has become
5315 : * all visible while we were busy locking the buffer, we'll have to
5316 : * unlock and re-lock, to avoid holding the buffer lock across I/O.
5317 : * That's a bit unfortunate, but hopefully shouldn't happen often.
5318 : *
5319 : * Note: in some paths through this function, we will reach here
5320 ECB : * holding a pin on a vm page that may or may not be the one matching
5321 : * this page. If this page isn't all-visible, we won't use the vm
5322 : * page, but we hold onto such a pin till the end of the function.
5323 : */
5324 GBC 91 : if (!pinned_desired_page && PageIsAllVisible(BufferGetPage(buf)))
5325 EUB : {
5326 UBC 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5327 UIC 0 : visibilitymap_pin(rel, block, &vmbuffer);
5328 UBC 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5329 EUB : }
5330 :
5331 : /*
5332 : * Check the tuple XMIN against prior XMAX, if any. If we reached the
5333 : * end of the chain, we're done, so return success.
5334 : */
5335 GIC 94 : if (TransactionIdIsValid(priorXmax) &&
5336 3 : !TransactionIdEquals(HeapTupleHeaderGetXmin(mytup.t_data),
5337 : priorXmax))
5338 : {
5339 UIC 0 : result = TM_Ok;
5340 UBC 0 : goto out_locked;
5341 : }
5342 :
5343 : /*
5344 : * Also check Xmin: if this tuple was created by an aborted
5345 : * (sub)transaction, then we already locked the last live one in the
5346 ECB : * chain, thus we're done, so return success.
5347 : */
5348 GIC 91 : if (TransactionIdDidAbort(HeapTupleHeaderGetXmin(mytup.t_data)))
5349 ECB : {
5350 GIC 13 : result = TM_Ok;
5351 13 : goto out_locked;
5352 ECB : }
5353 :
5354 GIC 78 : old_infomask = mytup.t_data->t_infomask;
5355 78 : old_infomask2 = mytup.t_data->t_infomask2;
5356 78 : xmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5357 :
5358 : /*
5359 : * If this tuple version has been updated or locked by some concurrent
5360 : * transaction(s), what we do depends on whether our lock mode
5361 : * conflicts with what those other transactions hold, and also on the
5362 : * status of them.
5363 : */
5364 CBC 78 : if (!(old_infomask & HEAP_XMAX_INVALID))
5365 EUB : {
5366 : TransactionId rawxmax;
5367 ECB : bool needwait;
5368 :
5369 CBC 30 : rawxmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5370 30 : if (old_infomask & HEAP_XMAX_IS_MULTI)
5371 : {
5372 ECB : int nmembers;
5373 : int i;
5374 : MultiXactMember *members;
5375 :
5376 : /*
5377 : * We don't need a test for pg_upgrade'd tuples: this is only
5378 : * applied to tuples after the first in an update chain. Said
5379 : * first tuple in the chain may well be locked-in-9.2-and-
5380 : * pg_upgraded, but that one was already locked by our caller,
5381 : * not us; and any subsequent ones cannot be because our
5382 : * caller must necessarily have obtained a snapshot later than
5383 : * the pg_upgrade itself.
5384 : */
5385 GIC 1 : Assert(!HEAP_LOCKED_UPGRADED(mytup.t_data->t_infomask));
5386 ECB :
5387 GBC 1 : nmembers = GetMultiXactIdMembers(rawxmax, &members, false,
5388 GIC 1 : HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5389 GBC 4 : for (i = 0; i < nmembers; i++)
5390 : {
5391 CBC 3 : result = test_lockmode_for_conflict(members[i].status,
5392 GIC 3 : members[i].xid,
5393 : mode,
5394 ECB : &mytup,
5395 : &needwait);
5396 :
5397 : /*
5398 : * If the tuple was already locked by ourselves in a
5399 : * previous iteration of this (say heap_lock_tuple was
5400 : * forced to restart the locking loop because of a change
5401 : * in xmax), then we hold the lock already on this tuple
5402 : * version and we don't need to do anything; and this is
5403 : * not an error condition either. We just need to skip
5404 : * this tuple and continue locking the next version in the
5405 : * update chain.
5406 : */
5407 CBC 3 : if (result == TM_SelfModified)
5408 : {
5409 LBC 0 : pfree(members);
5410 0 : goto next;
5411 : }
5412 ECB :
5413 CBC 3 : if (needwait)
5414 ECB : {
5415 LBC 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5416 0 : XactLockTableWait(members[i].xid, rel,
5417 : &mytup.t_self,
5418 ECB : XLTW_LockUpdated);
5419 UIC 0 : pfree(members);
5420 LBC 0 : goto l4;
5421 : }
5422 CBC 3 : if (result != TM_Ok)
5423 : {
5424 UIC 0 : pfree(members);
5425 LBC 0 : goto out_locked;
5426 : }
5427 ECB : }
5428 GIC 1 : if (members)
5429 CBC 1 : pfree(members);
5430 ECB : }
5431 : else
5432 : {
5433 : MultiXactStatus status;
5434 :
5435 : /*
5436 : * For a non-multi Xmax, we first need to compute the
5437 : * corresponding MultiXactStatus by using the infomask bits.
5438 : */
5439 CBC 29 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5440 ECB : {
5441 CBC 10 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5442 GIC 10 : status = MultiXactStatusForKeyShare;
5443 UIC 0 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5444 0 : status = MultiXactStatusForShare;
5445 0 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5446 ECB : {
5447 LBC 0 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5448 UIC 0 : status = MultiXactStatusForUpdate;
5449 ECB : else
5450 LBC 0 : status = MultiXactStatusForNoKeyUpdate;
5451 EUB : }
5452 : else
5453 ECB : {
5454 : /*
5455 : * LOCK_ONLY present alone (a pg_upgraded tuple marked
5456 : * as share-locked in the old cluster) shouldn't be
5457 : * seen in the middle of an update chain.
5458 : */
5459 UIC 0 : elog(ERROR, "invalid lock status in tuple");
5460 : }
5461 : }
5462 : else
5463 : {
5464 : /* it's an update, but which kind? */
5465 GIC 19 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5466 14 : status = MultiXactStatusUpdate;
5467 : else
5468 5 : status = MultiXactStatusNoKeyUpdate;
5469 : }
5470 :
5471 29 : result = test_lockmode_for_conflict(status, rawxmax, mode,
5472 : &mytup, &needwait);
5473 :
5474 : /*
5475 : * If the tuple was already locked by ourselves in a previous
5476 : * iteration of this (say heap_lock_tuple was forced to
5477 : * restart the locking loop because of a change in xmax), then
5478 : * we hold the lock already on this tuple version and we don't
5479 ECB : * need to do anything; and this is not an error condition
5480 : * either. We just need to skip this tuple and continue
5481 : * locking the next version in the update chain.
5482 : */
5483 GIC 29 : if (result == TM_SelfModified)
5484 UIC 0 : goto next;
5485 :
5486 CBC 29 : if (needwait)
5487 ECB : {
5488 GIC 8 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5489 8 : XactLockTableWait(rawxmax, rel, &mytup.t_self,
5490 : XLTW_LockUpdated);
5491 8 : goto l4;
5492 : }
5493 21 : if (result != TM_Ok)
5494 : {
5495 8 : goto out_locked;
5496 : }
5497 : }
5498 ECB : }
5499 :
5500 : /* compute the new Xmax and infomask values for the tuple ... */
5501 GIC 62 : compute_new_xmax_infomask(xmax, old_infomask, mytup.t_data->t_infomask2,
5502 : xid, mode, false,
5503 : &new_xmax, &new_infomask, &new_infomask2);
5504 ECB :
5505 GIC 62 : if (PageIsAllVisible(BufferGetPage(buf)) &&
5506 UIC 0 : visibilitymap_clear(rel, block, vmbuffer,
5507 : VISIBILITYMAP_ALL_FROZEN))
5508 0 : cleared_all_frozen = true;
5509 :
5510 GIC 62 : START_CRIT_SECTION();
5511 :
5512 : /* ... and set them */
5513 62 : HeapTupleHeaderSetXmax(mytup.t_data, new_xmax);
5514 62 : mytup.t_data->t_infomask &= ~HEAP_XMAX_BITS;
5515 62 : mytup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5516 62 : mytup.t_data->t_infomask |= new_infomask;
5517 62 : mytup.t_data->t_infomask2 |= new_infomask2;
5518 :
5519 62 : MarkBufferDirty(buf);
5520 :
5521 : /* XLOG stuff */
5522 62 : if (RelationNeedsWAL(rel))
5523 : {
5524 ECB : xl_heap_lock_updated xlrec;
5525 : XLogRecPtr recptr;
5526 GIC 62 : Page page = BufferGetPage(buf);
5527 :
5528 62 : XLogBeginInsert();
5529 CBC 62 : XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
5530 :
5531 GIC 62 : xlrec.offnum = ItemPointerGetOffsetNumber(&mytup.t_self);
5532 CBC 62 : xlrec.xmax = new_xmax;
5533 62 : xlrec.infobits_set = compute_infobits(new_infomask, new_infomask2);
5534 62 : xlrec.flags =
5535 GIC 62 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
5536 ECB :
5537 CBC 62 : XLogRegisterData((char *) &xlrec, SizeOfHeapLockUpdated);
5538 ECB :
5539 GIC 62 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_LOCK_UPDATED);
5540 ECB :
5541 GBC 62 : PageSetLSN(page, recptr);
5542 : }
5543 ECB :
5544 GIC 62 : END_CRIT_SECTION();
5545 :
5546 CBC 62 : next:
5547 : /* if we find the end of update chain, we're done. */
5548 124 : if (mytup.t_data->t_infomask & HEAP_XMAX_INVALID ||
5549 GIC 124 : HeapTupleHeaderIndicatesMovedPartitions(mytup.t_data) ||
5550 CBC 66 : ItemPointerEquals(&mytup.t_self, &mytup.t_data->t_ctid) ||
5551 GIC 4 : HeapTupleHeaderIsOnlyLocked(mytup.t_data))
5552 : {
5553 59 : result = TM_Ok;
5554 59 : goto out_locked;
5555 : }
5556 ECB :
5557 : /* tail recursion */
5558 GIC 3 : priorXmax = HeapTupleHeaderGetUpdateXid(mytup.t_data);
5559 CBC 3 : ItemPointerCopy(&(mytup.t_data->t_ctid), &tupid);
5560 GIC 3 : UnlockReleaseBuffer(buf);
5561 : }
5562 :
5563 : result = TM_Ok;
5564 ECB :
5565 GIC 80 : out_locked:
5566 CBC 80 : UnlockReleaseBuffer(buf);
5567 :
5568 GIC 80 : out_unlocked:
5569 CBC 80 : if (vmbuffer != InvalidBuffer)
5570 UIC 0 : ReleaseBuffer(vmbuffer);
5571 ECB :
5572 CBC 80 : return result;
5573 : }
5574 ECB :
5575 : /*
5576 : * heap_lock_updated_tuple
5577 : * Follow update chain when locking an updated tuple, acquiring locks (row
5578 : * marks) on the updated versions.
5579 : *
5580 : * The initial tuple is assumed to be already locked.
5581 : *
5582 : * This function doesn't check visibility, it just unconditionally marks the
5583 : * tuple(s) as locked. If any tuple in the updated chain is being deleted
5584 : * concurrently (or updated with the key being modified), sleep until the
5585 : * transaction doing it is finished.
5586 : *
5587 : * Note that we don't acquire heavyweight tuple locks on the tuples we walk
5588 : * when we have to wait for other transactions to release them, as opposed to
5589 : * what heap_lock_tuple does. The reason is that having more than one
5590 : * transaction walking the chain is probably uncommon enough that risk of
5591 : * starvation is not likely: one of the preconditions for being here is that
5592 : * the snapshot in use predates the update that created this tuple (because we
5593 : * started at an earlier version of the tuple), but at the same time such a
5594 : * transaction cannot be using repeatable read or serializable isolation
5595 : * levels, because that would lead to a serializability failure.
5596 : */
5597 : static TM_Result
5598 GIC 88 : heap_lock_updated_tuple(Relation rel, HeapTuple tuple, ItemPointer ctid,
5599 : TransactionId xid, LockTupleMode mode)
5600 : {
5601 : /*
5602 : * If the tuple has not been updated, or has moved into another partition
5603 : * (effectively a delete) stop here.
5604 : */
5605 88 : if (!HeapTupleHeaderIndicatesMovedPartitions(tuple->t_data) &&
5606 86 : !ItemPointerEquals(&tuple->t_self, ctid))
5607 : {
5608 : /*
5609 : * If this is the first possibly-multixact-able operation in the
5610 : * current transaction, set my per-backend OldestMemberMXactId
5611 ECB : * setting. We can be certain that the transaction will never become a
5612 : * member of any older MultiXactIds than that. (We have to do this
5613 : * even if we end up just using our own TransactionId below, since
5614 : * some other backend could incorporate our XID into a MultiXact
5615 : * immediately afterwards.)
5616 : */
5617 GIC 80 : MultiXactIdSetOldestMember();
5618 :
5619 80 : return heap_lock_updated_tuple_rec(rel, ctid, xid, mode);
5620 : }
5621 ECB :
5622 : /* nothing to lock */
5623 CBC 8 : return TM_Ok;
5624 ECB : }
5625 :
5626 : /*
5627 : * heap_finish_speculative - mark speculative insertion as successful
5628 : *
5629 : * To successfully finish a speculative insertion we have to clear speculative
5630 : * token from tuple. To do so the t_ctid field, which will contain a
5631 : * speculative token value, is modified in place to point to the tuple itself,
5632 : * which is characteristic of a newly inserted ordinary tuple.
5633 : *
5634 : * NB: It is not ok to commit without either finishing or aborting a
5635 : * speculative insertion. We could treat speculative tuples of committed
5636 : * transactions implicitly as completed, but then we would have to be prepared
5637 : * to deal with speculative tokens on committed tuples. That wouldn't be
5638 : * difficult - no-one looks at the ctid field of a tuple with invalid xmax -
5639 : * but clearing the token at completion isn't very expensive either.
5640 : * An explicit confirmation WAL record also makes logical decoding simpler.
5641 : */
5642 : void
5643 GIC 2005 : heap_finish_speculative(Relation relation, ItemPointer tid)
5644 : {
5645 : Buffer buffer;
5646 : Page page;
5647 ECB : OffsetNumber offnum;
5648 GBC 2005 : ItemId lp = NULL;
5649 ECB : HeapTupleHeader htup;
5650 EUB :
5651 CBC 2005 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
5652 GIC 2005 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
5653 2005 : page = (Page) BufferGetPage(buffer);
5654 :
5655 2005 : offnum = ItemPointerGetOffsetNumber(tid);
5656 2005 : if (PageGetMaxOffsetNumber(page) >= offnum)
5657 2005 : lp = PageGetItemId(page, offnum);
5658 :
5659 CBC 2005 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
5660 UIC 0 : elog(ERROR, "invalid lp");
5661 :
5662 GIC 2005 : htup = (HeapTupleHeader) PageGetItem(page, lp);
5663 :
5664 : /* NO EREPORT(ERROR) from here till changes are logged */
5665 2005 : START_CRIT_SECTION();
5666 :
5667 CBC 2005 : Assert(HeapTupleHeaderIsSpeculative(htup));
5668 ECB :
5669 GBC 2005 : MarkBufferDirty(buffer);
5670 :
5671 ECB : /*
5672 : * Replace the speculative insertion token with a real t_ctid, pointing to
5673 : * itself like it does on regular tuples.
5674 : */
5675 CBC 2005 : htup->t_ctid = *tid;
5676 ECB :
5677 : /* XLOG stuff */
5678 GIC 2005 : if (RelationNeedsWAL(relation))
5679 : {
5680 : xl_heap_confirm xlrec;
5681 : XLogRecPtr recptr;
5682 :
5683 CBC 1999 : xlrec.offnum = ItemPointerGetOffsetNumber(tid);
5684 :
5685 GIC 1999 : XLogBeginInsert();
5686 ECB :
5687 : /* We want the same filtering on this as on a plain insert */
5688 CBC 1999 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
5689 :
5690 GIC 1999 : XLogRegisterData((char *) &xlrec, SizeOfHeapConfirm);
5691 1999 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
5692 :
5693 1999 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_CONFIRM);
5694 :
5695 1999 : PageSetLSN(page, recptr);
5696 ECB : }
5697 :
5698 GIC 2005 : END_CRIT_SECTION();
5699 :
5700 2005 : UnlockReleaseBuffer(buffer);
5701 CBC 2005 : }
5702 ECB :
5703 : /*
5704 : * heap_abort_speculative - kill a speculatively inserted tuple
5705 : *
5706 : * Marks a tuple that was speculatively inserted in the same command as dead,
5707 : * by setting its xmin as invalid. That makes it immediately appear as dead
5708 : * to all transactions, including our own. In particular, it makes
5709 : * HeapTupleSatisfiesDirty() regard the tuple as dead, so that another backend
5710 : * inserting a duplicate key value won't unnecessarily wait for our whole
5711 : * transaction to finish (it'll just wait for our speculative insertion to
5712 : * finish).
5713 : *
5714 : * Killing the tuple prevents "unprincipled deadlocks", which are deadlocks
5715 : * that arise due to a mutual dependency that is not user visible. By
5716 : * definition, unprincipled deadlocks cannot be prevented by the user
5717 : * reordering lock acquisition in client code, because the implementation level
5718 : * lock acquisitions are not under the user's direct control. If speculative
5719 : * inserters did not take this precaution, then under high concurrency they
5720 : * could deadlock with each other, which would not be acceptable.
5721 : *
5722 : * This is somewhat redundant with heap_delete, but we prefer to have a
5723 : * dedicated routine with stripped down requirements. Note that this is also
5724 : * used to delete the TOAST tuples created during speculative insertion.
5725 : *
5726 : * This routine does not affect logical decoding as it only looks at
5727 : * confirmation records.
5728 : */
5729 : void
5730 GIC 10 : heap_abort_speculative(Relation relation, ItemPointer tid)
5731 : {
5732 10 : TransactionId xid = GetCurrentTransactionId();
5733 : ItemId lp;
5734 ECB : HeapTupleData tp;
5735 : Page page;
5736 : BlockNumber block;
5737 : Buffer buffer;
5738 : TransactionId prune_xid;
5739 :
5740 GIC 10 : Assert(ItemPointerIsValid(tid));
5741 :
5742 10 : block = ItemPointerGetBlockNumber(tid);
5743 10 : buffer = ReadBuffer(relation, block);
5744 10 : page = BufferGetPage(buffer);
5745 :
5746 10 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
5747 :
5748 : /*
5749 : * Page can't be all visible, we just inserted into it, and are still
5750 : * running.
5751 : */
5752 10 : Assert(!PageIsAllVisible(page));
5753 :
5754 10 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
5755 10 : Assert(ItemIdIsNormal(lp));
5756 :
5757 10 : tp.t_tableOid = RelationGetRelid(relation);
5758 10 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
5759 10 : tp.t_len = ItemIdGetLength(lp);
5760 CBC 10 : tp.t_self = *tid;
5761 :
5762 : /*
5763 : * Sanity check that the tuple really is a speculatively inserted tuple,
5764 : * inserted by us.
5765 ECB : */
5766 GIC 10 : if (tp.t_data->t_choice.t_heap.t_xmin != xid)
5767 UIC 0 : elog(ERROR, "attempted to kill a tuple inserted by another transaction");
5768 GIC 10 : if (!(IsToastRelation(relation) || HeapTupleHeaderIsSpeculative(tp.t_data)))
5769 UIC 0 : elog(ERROR, "attempted to kill a non-speculative tuple");
5770 GIC 10 : Assert(!HeapTupleHeaderIsHeapOnly(tp.t_data));
5771 :
5772 : /*
5773 : * No need to check for serializable conflicts here. There is never a
5774 : * need for a combo CID, either. No need to extract replica identity, or
5775 : * do anything special with infomask bits.
5776 ECB : */
5777 EUB :
5778 GIC 10 : START_CRIT_SECTION();
5779 :
5780 : /*
5781 ECB : * The tuple will become DEAD immediately. Flag that this page is a
5782 : * candidate for pruning by setting xmin to TransactionXmin. While not
5783 : * immediately prunable, it is the oldest xid we can cheaply determine
5784 : * that's safe against wraparound / being older than the table's
5785 : * relfrozenxid. To defend against the unlikely case of a new relation
5786 : * having a newer relfrozenxid than our TransactionXmin, use relfrozenxid
5787 : * if so (vacuum can't subsequently move relfrozenxid to beyond
5788 : * TransactionXmin, so there's no race here).
5789 : */
5790 GBC 10 : Assert(TransactionIdIsValid(TransactionXmin));
5791 GIC 10 : if (TransactionIdPrecedes(TransactionXmin, relation->rd_rel->relfrozenxid))
5792 LBC 0 : prune_xid = relation->rd_rel->relfrozenxid;
5793 : else
5794 CBC 10 : prune_xid = TransactionXmin;
5795 10 : PageSetPrunable(page, prune_xid);
5796 ECB :
5797 EUB : /* store transaction information of xact deleting the tuple */
5798 GIC 10 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
5799 10 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5800 ECB :
5801 : /*
5802 : * Set the tuple header xmin to InvalidTransactionId. This makes the
5803 : * tuple immediately invisible everyone. (In particular, to any
5804 : * transactions waiting on the speculative token, woken up later.)
5805 : */
5806 CBC 10 : HeapTupleHeaderSetXmin(tp.t_data, InvalidTransactionId);
5807 :
5808 : /* Clear the speculative insertion token too */
5809 10 : tp.t_data->t_ctid = tp.t_self;
5810 :
5811 GIC 10 : MarkBufferDirty(buffer);
5812 :
5813 : /*
5814 ECB : * XLOG stuff
5815 : *
5816 : * The WAL records generated here match heap_delete(). The same recovery
5817 : * routines are used.
5818 : */
5819 CBC 10 : if (RelationNeedsWAL(relation))
5820 ECB : {
5821 : xl_heap_delete xlrec;
5822 : XLogRecPtr recptr;
5823 :
5824 CBC 10 : xlrec.flags = XLH_DELETE_IS_SUPER;
5825 GIC 20 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
5826 CBC 10 : tp.t_data->t_infomask2);
5827 GIC 10 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
5828 10 : xlrec.xmax = xid;
5829 ECB :
5830 GIC 10 : XLogBeginInsert();
5831 CBC 10 : XLogRegisterData((char *) &xlrec, SizeOfHeapDelete);
5832 GIC 10 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
5833 :
5834 : /* No replica identity & replication origin logged */
5835 :
5836 10 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
5837 :
5838 10 : PageSetLSN(page, recptr);
5839 ECB : }
5840 :
5841 CBC 10 : END_CRIT_SECTION();
5842 :
5843 GIC 10 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
5844 :
5845 10 : if (HeapTupleHasExternal(&tp))
5846 : {
5847 1 : Assert(!IsToastRelation(relation));
5848 1 : heap_toast_delete(relation, &tp, true);
5849 : }
5850 :
5851 : /*
5852 : * Never need to mark tuple for invalidation, since catalogs don't support
5853 : * speculative insertion
5854 : */
5855 :
5856 : /* Now we can release the buffer */
5857 10 : ReleaseBuffer(buffer);
5858 :
5859 : /* count deletion, as we counted the insertion too */
5860 10 : pgstat_count_heap_delete(relation);
5861 10 : }
5862 :
5863 : /*
5864 : * heap_inplace_update - update a tuple "in place" (ie, overwrite it)
5865 : *
5866 : * Overwriting violates both MVCC and transactional safety, so the uses
5867 : * of this function in Postgres are extremely limited. Nonetheless we
5868 : * find some places to use it.
5869 : *
5870 : * The tuple cannot change size, and therefore it's reasonable to assume
5871 : * that its null bitmap (if any) doesn't change either. So we just
5872 : * overwrite the data portion of the tuple without touching the null
5873 : * bitmap or any of the header fields.
5874 : *
5875 : * tuple is an in-memory tuple structure containing the data to be written
5876 : * over the target tuple. Also, tuple->t_self identifies the target tuple.
5877 : *
5878 : * Note that the tuple updated here had better not come directly from the
5879 : * syscache if the relation has a toast relation as this tuple could
5880 : * include toast values that have been expanded, causing a failure here.
5881 : */
5882 : void
5883 184243 : heap_inplace_update(Relation relation, HeapTuple tuple)
5884 : {
5885 : Buffer buffer;
5886 : Page page;
5887 : OffsetNumber offnum;
5888 184243 : ItemId lp = NULL;
5889 : HeapTupleHeader htup;
5890 : uint32 oldlen;
5891 : uint32 newlen;
5892 :
5893 : /*
5894 : * For now, we don't allow parallel updates. Unlike a regular update,
5895 : * this should never create a combo CID, so it might be possible to relax
5896 ECB : * this restriction, but not without more thought and testing. It's not
5897 : * clear that it would be useful, anyway.
5898 : */
5899 GIC 184243 : if (IsInParallelMode())
5900 UIC 0 : ereport(ERROR,
5901 : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
5902 : errmsg("cannot update tuples during a parallel operation")));
5903 :
5904 GIC 184243 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&(tuple->t_self)));
5905 184243 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
5906 184243 : page = (Page) BufferGetPage(buffer);
5907 :
5908 184243 : offnum = ItemPointerGetOffsetNumber(&(tuple->t_self));
5909 184243 : if (PageGetMaxOffsetNumber(page) >= offnum)
5910 184243 : lp = PageGetItemId(page, offnum);
5911 ECB :
5912 GIC 184243 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
5913 UIC 0 : elog(ERROR, "invalid lp");
5914 ECB :
5915 GIC 184243 : htup = (HeapTupleHeader) PageGetItem(page, lp);
5916 ECB :
5917 CBC 184243 : oldlen = ItemIdGetLength(lp) - htup->t_hoff;
5918 GIC 184243 : newlen = tuple->t_len - tuple->t_data->t_hoff;
5919 GBC 184243 : if (oldlen != newlen || htup->t_hoff != tuple->t_data->t_hoff)
5920 UBC 0 : elog(ERROR, "wrong tuple length");
5921 EUB :
5922 : /* NO EREPORT(ERROR) from here till changes are logged */
5923 CBC 184243 : START_CRIT_SECTION();
5924 EUB :
5925 GIC 184243 : memcpy((char *) htup + htup->t_hoff,
5926 184243 : (char *) tuple->t_data + tuple->t_data->t_hoff,
5927 : newlen);
5928 ECB :
5929 GIC 184243 : MarkBufferDirty(buffer);
5930 :
5931 : /* XLOG stuff */
5932 184243 : if (RelationNeedsWAL(relation))
5933 : {
5934 : xl_heap_inplace xlrec;
5935 : XLogRecPtr recptr;
5936 :
5937 184235 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
5938 ECB :
5939 CBC 184235 : XLogBeginInsert();
5940 GBC 184235 : XLogRegisterData((char *) &xlrec, SizeOfHeapInplace);
5941 :
5942 GIC 184235 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
5943 184235 : XLogRegisterBufData(0, (char *) htup + htup->t_hoff, newlen);
5944 :
5945 ECB : /* inplace updates aren't decoded atm, don't log the origin */
5946 :
5947 CBC 184235 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_INPLACE);
5948 ECB :
5949 CBC 184235 : PageSetLSN(page, recptr);
5950 : }
5951 :
5952 GIC 184243 : END_CRIT_SECTION();
5953 EUB :
5954 GBC 184243 : UnlockReleaseBuffer(buffer);
5955 EUB :
5956 : /*
5957 : * Send out shared cache inval if necessary. Note that because we only
5958 : * pass the new version of the tuple, this mustn't be used for any
5959 : * operations that could change catcache lookup keys. But we aren't
5960 : * bothering with index updates either, so that's true a fortiori.
5961 : */
5962 GIC 184243 : if (!IsBootstrapProcessingMode())
5963 93963 : CacheInvalidateHeapTuple(relation, tuple, NULL);
5964 184243 : }
5965 :
5966 : #define FRM_NOOP 0x0001
5967 EUB : #define FRM_INVALIDATE_XMAX 0x0002
5968 : #define FRM_RETURN_IS_XID 0x0004
5969 : #define FRM_RETURN_IS_MULTI 0x0008
5970 : #define FRM_MARK_COMMITTED 0x0010
5971 :
5972 : /*
5973 : * FreezeMultiXactId
5974 : * Determine what to do during freezing when a tuple is marked by a
5975 : * MultiXactId.
5976 : *
5977 : * "flags" is an output value; it's used to tell caller what to do on return.
5978 : * "pagefrz" is an input/output value, used to manage page level freezing.
5979 : *
5980 : * Possible values that we can set in "flags":
5981 : * FRM_NOOP
5982 : * don't do anything -- keep existing Xmax
5983 : * FRM_INVALIDATE_XMAX
5984 : * mark Xmax as InvalidTransactionId and set XMAX_INVALID flag.
5985 : * FRM_RETURN_IS_XID
5986 : * The Xid return value is a single update Xid to set as xmax.
5987 : * FRM_MARK_COMMITTED
5988 ECB : * Xmax can be marked as HEAP_XMAX_COMMITTED
5989 : * FRM_RETURN_IS_MULTI
5990 : * The return value is a new MultiXactId to set as new Xmax.
5991 : * (caller must obtain proper infomask bits using GetMultiXactIdHintBits)
5992 : *
5993 : * Caller delegates control of page freezing to us. In practice we always
5994 : * force freezing of caller's page unless FRM_NOOP processing is indicated.
5995 : * We help caller ensure that XIDs < FreezeLimit and MXIDs < MultiXactCutoff
5996 : * can never be left behind. We freely choose when and how to process each
5997 : * Multi, without ever violating the cutoff postconditions for freezing.
5998 EUB : *
5999 : * It's useful to remove Multis on a proactive timeline (relative to freezing
6000 : * XIDs) to keep MultiXact member SLRU buffer misses to a minimum. It can also
6001 : * be cheaper in the short run, for us, since we too can avoid SLRU buffer
6002 : * misses through eager processing.
6003 : *
6004 : * NB: Creates a _new_ MultiXactId when FRM_RETURN_IS_MULTI is set, though only
6005 : * when FreezeLimit and/or MultiXactCutoff cutoffs leave us with no choice.
6006 : * This can usually be put off, which is usually enough to avoid it altogether.
6007 : * Allocating new multis during VACUUM should be avoided on general principle;
6008 : * only VACUUM can advance relminmxid, so allocating new Multis here comes with
6009 : * its own special risks.
6010 : *
6011 : * NB: Caller must maintain "no freeze" NewRelfrozenXid/NewRelminMxid trackers
6012 : * using heap_tuple_should_freeze when we haven't forced page-level freezing.
6013 : *
6014 : * NB: Caller should avoid needlessly calling heap_tuple_should_freeze when we
6015 : * have already forced page-level freezing, since that might incur the same
6016 : * SLRU buffer misses that we specifically intended to avoid by freezing.
6017 : */
6018 : static TransactionId
6019 GIC 6 : FreezeMultiXactId(MultiXactId multi, uint16 t_infomask,
6020 : const struct VacuumCutoffs *cutoffs, uint16 *flags,
6021 : HeapPageFreeze *pagefrz)
6022 : {
6023 : TransactionId newxmax;
6024 : MultiXactMember *members;
6025 : int nmembers;
6026 : bool need_replace;
6027 : int nnewmembers;
6028 : MultiXactMember *newmembers;
6029 : bool has_lockers;
6030 : TransactionId update_xid;
6031 : bool update_committed;
6032 : TransactionId FreezePageRelfrozenXid;
6033 :
6034 CBC 6 : *flags = 0;
6035 ECB :
6036 : /* We should only be called in Multis */
6037 GIC 6 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
6038 ECB :
6039 GIC 6 : if (!MultiXactIdIsValid(multi) ||
6040 CBC 6 : HEAP_LOCKED_UPGRADED(t_infomask))
6041 : {
6042 UIC 0 : *flags |= FRM_INVALIDATE_XMAX;
6043 UNC 0 : pagefrz->freeze_required = true;
6044 UIC 0 : return InvalidTransactionId;
6045 ECB : }
6046 GNC 6 : else if (MultiXactIdPrecedes(multi, cutoffs->relminmxid))
6047 UIC 0 : ereport(ERROR,
6048 ECB : (errcode(ERRCODE_DATA_CORRUPTED),
6049 EUB : errmsg_internal("found multixact %u from before relminmxid %u",
6050 : multi, cutoffs->relminmxid)));
6051 GNC 6 : else if (MultiXactIdPrecedes(multi, cutoffs->OldestMxact))
6052 : {
6053 : TransactionId update_xact;
6054 :
6055 ECB : /*
6056 : * This old multi cannot possibly have members still running, but
6057 : * verify just in case. If it was a locker only, it can be removed
6058 : * without any further consideration; but if it contained an update,
6059 : * we might need to preserve it.
6060 : */
6061 GIC 4 : if (MultiXactIdIsRunning(multi,
6062 4 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask)))
6063 UIC 0 : ereport(ERROR,
6064 : (errcode(ERRCODE_DATA_CORRUPTED),
6065 : errmsg_internal("multixact %u from before multi freeze cutoff %u found to be still running",
6066 : multi, cutoffs->OldestMxact)));
6067 ECB :
6068 GBC 4 : if (HEAP_XMAX_IS_LOCKED_ONLY(t_infomask))
6069 ECB : {
6070 CBC 4 : *flags |= FRM_INVALIDATE_XMAX;
6071 GNC 4 : pagefrz->freeze_required = true;
6072 4 : return InvalidTransactionId;
6073 : }
6074 :
6075 : /* replace multi with single XID for its updater? */
6076 UNC 0 : update_xact = MultiXactIdGetUpdateXid(multi, t_infomask);
6077 0 : if (TransactionIdPrecedes(update_xact, cutoffs->relfrozenxid))
6078 0 : ereport(ERROR,
6079 : (errcode(ERRCODE_DATA_CORRUPTED),
6080 : errmsg_internal("multixact %u contains update XID %u from before relfrozenxid %u",
6081 : multi, update_xact,
6082 : cutoffs->relfrozenxid)));
6083 0 : else if (TransactionIdPrecedes(update_xact, cutoffs->OldestXmin))
6084 : {
6085 : /*
6086 : * Updater XID has to have aborted (otherwise the tuple would have
6087 : * been pruned away instead, since updater XID is < OldestXmin).
6088 : * Just remove xmax.
6089 : */
6090 0 : if (TransactionIdDidCommit(update_xact))
6091 LBC 0 : ereport(ERROR,
6092 ECB : (errcode(ERRCODE_DATA_CORRUPTED),
6093 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6094 : multi, update_xact,
6095 : cutoffs->OldestXmin)));
6096 UNC 0 : *flags |= FRM_INVALIDATE_XMAX;
6097 0 : pagefrz->freeze_required = true;
6098 0 : return InvalidTransactionId;
6099 ECB : }
6100 :
6101 : /* Have to keep updater XID as new xmax */
6102 UNC 0 : *flags |= FRM_RETURN_IS_XID;
6103 0 : pagefrz->freeze_required = true;
6104 0 : return update_xact;
6105 : }
6106 ECB :
6107 : /*
6108 : * Some member(s) of this Multi may be below FreezeLimit xid cutoff, so we
6109 : * need to walk the whole members array to figure out what to do, if
6110 : * anything.
6111 : */
6112 : nmembers =
6113 GIC 2 : GetMultiXactIdMembers(multi, &members, false,
6114 2 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask));
6115 2 : if (nmembers <= 0)
6116 : {
6117 : /* Nothing worth keeping */
6118 UIC 0 : *flags |= FRM_INVALIDATE_XMAX;
6119 UNC 0 : pagefrz->freeze_required = true;
6120 UIC 0 : return InvalidTransactionId;
6121 : }
6122 EUB :
6123 : /*
6124 : * The FRM_NOOP case is the only case where we might need to ratchet back
6125 : * FreezePageRelfrozenXid or FreezePageRelminMxid. It is also the only
6126 : * case where our caller might ratchet back its NoFreezePageRelfrozenXid
6127 : * or NoFreezePageRelminMxid "no freeze" trackers to deal with a multi.
6128 : * FRM_NOOP handling should result in the NewRelfrozenXid/NewRelminMxid
6129 : * trackers managed by VACUUM being ratcheting back by xmax to the degree
6130 : * required to make it safe to leave xmax undisturbed, independent of
6131 : * whether or not page freezing is triggered somewhere else.
6132 : *
6133 : * Our policy is to force freezing in every case other than FRM_NOOP,
6134 : * which obviates the need to maintain either set of trackers, anywhere.
6135 : * Every other case will reliably execute a freeze plan for xmax that
6136 : * either replaces xmax with an XID/MXID >= OldestXmin/OldestMxact, or
6137 : * sets xmax to an InvalidTransactionId XID, rendering xmax fully frozen.
6138 : * (VACUUM's NewRelfrozenXid/NewRelminMxid trackers are initialized with
6139 : * OldestXmin/OldestMxact, so later values never need to be tracked here.)
6140 : */
6141 GIC 2 : need_replace = false;
6142 GNC 2 : FreezePageRelfrozenXid = pagefrz->FreezePageRelfrozenXid;
6143 4 : for (int i = 0; i < nmembers; i++)
6144 : {
6145 3 : TransactionId xid = members[i].xid;
6146 :
6147 3 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6148 :
6149 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
6150 : {
6151 : /* Can't violate the FreezeLimit postcondition */
6152 GIC 1 : need_replace = true;
6153 1 : break;
6154 : }
6155 GNC 2 : if (TransactionIdPrecedes(xid, FreezePageRelfrozenXid))
6156 UNC 0 : FreezePageRelfrozenXid = xid;
6157 EUB : }
6158 :
6159 : /* Can't violate the MultiXactCutoff postcondition, either */
6160 GNC 2 : if (!need_replace)
6161 1 : need_replace = MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff);
6162 :
6163 GIC 2 : if (!need_replace)
6164 : {
6165 : /*
6166 : * vacuumlazy.c might ratchet back NewRelminMxid, NewRelfrozenXid, or
6167 : * both together to make it safe to retain this particular multi after
6168 : * freezing its page
6169 : */
6170 1 : *flags |= FRM_NOOP;
6171 GNC 1 : pagefrz->FreezePageRelfrozenXid = FreezePageRelfrozenXid;
6172 1 : if (MultiXactIdPrecedes(multi, pagefrz->FreezePageRelminMxid))
6173 UNC 0 : pagefrz->FreezePageRelminMxid = multi;
6174 GIC 1 : pfree(members);
6175 1 : return multi;
6176 : }
6177 :
6178 EUB : /*
6179 : * Do a more thorough second pass over the multi to figure out which
6180 : * member XIDs actually need to be kept. Checking the precise status of
6181 : * individual members might even show that we don't need to keep anything.
6182 : * That is quite possible even though the Multi must be >= OldestMxact,
6183 : * since our second pass only keeps member XIDs when it's truly necessary;
6184 : * even member XIDs >= OldestXmin often won't be kept by second pass.
6185 : */
6186 GIC 1 : nnewmembers = 0;
6187 1 : newmembers = palloc(sizeof(MultiXactMember) * nmembers);
6188 GBC 1 : has_lockers = false;
6189 1 : update_xid = InvalidTransactionId;
6190 GIC 1 : update_committed = false;
6191 :
6192 : /*
6193 : * Determine whether to keep each member xid, or to ignore it instead
6194 : */
6195 GNC 3 : for (int i = 0; i < nmembers; i++)
6196 : {
6197 2 : TransactionId xid = members[i].xid;
6198 2 : MultiXactStatus mstatus = members[i].status;
6199 :
6200 2 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6201 :
6202 2 : if (!ISUPDATE_from_mxstatus(mstatus))
6203 : {
6204 : /*
6205 : * Locker XID (not updater XID). We only keep lockers that are
6206 : * still running.
6207 EUB : */
6208 GBC 4 : if (TransactionIdIsCurrentTransactionId(xid) ||
6209 GIC 2 : TransactionIdIsInProgress(xid))
6210 : {
6211 GNC 1 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6212 UNC 0 : ereport(ERROR,
6213 : (errcode(ERRCODE_DATA_CORRUPTED),
6214 : errmsg_internal("multixact %u contains running locker XID %u from before removable cutoff %u",
6215 : multi, xid,
6216 : cutoffs->OldestXmin)));
6217 CBC 1 : newmembers[nnewmembers++] = members[i];
6218 GIC 1 : has_lockers = true;
6219 : }
6220 :
6221 GNC 2 : continue;
6222 ECB : }
6223 :
6224 : /*
6225 : * Updater XID (not locker XID). Should we keep it?
6226 : *
6227 : * Since the tuple wasn't totally removed when vacuum pruned, the
6228 : * update Xid cannot possibly be older than OldestXmin cutoff unless
6229 : * the updater XID aborted. If the updater transaction is known
6230 : * aborted or crashed then it's okay to ignore it, otherwise not.
6231 : *
6232 : * In any case the Multi should never contain two updaters, whatever
6233 : * their individual commit status. Check for that first, in passing.
6234 : */
6235 UNC 0 : if (TransactionIdIsValid(update_xid))
6236 0 : ereport(ERROR,
6237 : (errcode(ERRCODE_DATA_CORRUPTED),
6238 : errmsg_internal("multixact %u has two or more updating members",
6239 : multi),
6240 : errdetail_internal("First updater XID=%u second updater XID=%u.",
6241 : update_xid, xid)));
6242 :
6243 : /*
6244 : * As with all tuple visibility routines, it's critical to test
6245 : * TransactionIdIsInProgress before TransactionIdDidCommit, because of
6246 : * race conditions explained in detail in heapam_visibility.c.
6247 : */
6248 0 : if (TransactionIdIsCurrentTransactionId(xid) ||
6249 0 : TransactionIdIsInProgress(xid))
6250 0 : update_xid = xid;
6251 0 : else if (TransactionIdDidCommit(xid))
6252 : {
6253 : /*
6254 : * The transaction committed, so we can tell caller to set
6255 : * HEAP_XMAX_COMMITTED. (We can only do this because we know the
6256 : * transaction is not running.)
6257 : */
6258 0 : update_committed = true;
6259 0 : update_xid = xid;
6260 : }
6261 : else
6262 : {
6263 : /*
6264 : * Not in progress, not committed -- must be aborted or crashed;
6265 : * we can ignore it.
6266 : */
6267 0 : continue;
6268 : }
6269 :
6270 : /*
6271 : * We determined that updater must be kept -- add it to pending new
6272 : * members list
6273 : */
6274 0 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6275 0 : ereport(ERROR,
6276 : (errcode(ERRCODE_DATA_CORRUPTED),
6277 : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6278 : multi, xid, cutoffs->OldestXmin)));
6279 0 : newmembers[nnewmembers++] = members[i];
6280 ECB : }
6281 :
6282 CBC 1 : pfree(members);
6283 ECB :
6284 : /*
6285 : * Determine what to do with caller's multi based on information gathered
6286 : * during our second pass
6287 : */
6288 GIC 1 : if (nnewmembers == 0)
6289 : {
6290 : /* Nothing worth keeping */
6291 LBC 0 : *flags |= FRM_INVALIDATE_XMAX;
6292 UNC 0 : newxmax = InvalidTransactionId;
6293 : }
6294 CBC 1 : else if (TransactionIdIsValid(update_xid) && !has_lockers)
6295 EUB : {
6296 : /*
6297 : * If there's a single member and it's an update, pass it back alone
6298 : * without creating a new Multi. (XXX we could do this when there's a
6299 : * single remaining locker, too, but that would complicate the API too
6300 : * much; moreover, the case with the single updater is more
6301 ECB : * interesting, because those are longer-lived.)
6302 : */
6303 UIC 0 : Assert(nnewmembers == 1);
6304 LBC 0 : *flags |= FRM_RETURN_IS_XID;
6305 0 : if (update_committed)
6306 UIC 0 : *flags |= FRM_MARK_COMMITTED;
6307 UNC 0 : newxmax = update_xid;
6308 : }
6309 : else
6310 : {
6311 ECB : /*
6312 : * Create a new multixact with the surviving members of the previous
6313 : * one, to set as new Xmax in the tuple
6314 EUB : */
6315 GNC 1 : newxmax = MultiXactIdCreateFromMembers(nnewmembers, newmembers);
6316 GIC 1 : *flags |= FRM_RETURN_IS_MULTI;
6317 : }
6318 :
6319 GBC 1 : pfree(newmembers);
6320 :
6321 GNC 1 : pagefrz->freeze_required = true;
6322 1 : return newxmax;
6323 : }
6324 :
6325 : /*
6326 ECB : * heap_prepare_freeze_tuple
6327 : *
6328 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
6329 : * are older than the OldestXmin and/or OldestMxact freeze cutoffs. If so,
6330 : * setup enough state (in the *frz output argument) to enable caller to
6331 : * process this tuple as part of freezing its page, and return true. Return
6332 : * false if nothing can be changed about the tuple right now.
6333 : *
6334 : * Also sets *totally_frozen to true if the tuple will be totally frozen once
6335 : * caller executes returned freeze plan (or if the tuple was already totally
6336 : * frozen by an earlier VACUUM). This indicates that there are no remaining
6337 : * XIDs or MultiXactIds that will need to be processed by a future VACUUM.
6338 : *
6339 : * VACUUM caller must assemble HeapTupleFreeze freeze plan entries for every
6340 : * tuple that we returned true for, and call heap_freeze_execute_prepared to
6341 : * execute freezing. Caller must initialize pagefrz fields for page as a
6342 : * whole before first call here for each heap page.
6343 : *
6344 : * VACUUM caller decides on whether or not to freeze the page as a whole.
6345 : * We'll often prepare freeze plans for a page that caller just discards.
6346 : * However, VACUUM doesn't always get to make a choice; it must freeze when
6347 : * pagefrz.freeze_required is set, to ensure that any XIDs < FreezeLimit (and
6348 : * MXIDs < MultiXactCutoff) can never be left behind. We help to make sure
6349 : * that VACUUM always follows that rule.
6350 : *
6351 : * We sometimes force freezing of xmax MultiXactId values long before it is
6352 : * strictly necessary to do so just to ensure the FreezeLimit postcondition.
6353 : * It's worth processing MultiXactIds proactively when it is cheap to do so,
6354 : * and it's convenient to make that happen by piggy-backing it on the "force
6355 : * freezing" mechanism. Conversely, we sometimes delay freezing MultiXactIds
6356 : * because it is expensive right now (though only when it's still possible to
6357 : * do so without violating the FreezeLimit/MultiXactCutoff postcondition).
6358 : *
6359 : * It is assumed that the caller has checked the tuple with
6360 : * HeapTupleSatisfiesVacuum() and determined that it is not HEAPTUPLE_DEAD
6361 : * (else we should be removing the tuple, not freezing it).
6362 : *
6363 : * NB: This function has side effects: it might allocate a new MultiXactId.
6364 : * It will be set as tuple's new xmax when our *frz output is processed within
6365 : * heap_execute_freeze_tuple later on. If the tuple is in a shared buffer
6366 : * then caller had better have an exclusive lock on it already.
6367 : */
6368 : bool
6369 GIC 9525743 : heap_prepare_freeze_tuple(HeapTupleHeader tuple,
6370 : const struct VacuumCutoffs *cutoffs,
6371 : HeapPageFreeze *pagefrz,
6372 : HeapTupleFreeze *frz, bool *totally_frozen)
6373 : {
6374 GNC 9525743 : bool xmin_already_frozen = false,
6375 9525743 : xmax_already_frozen = false;
6376 9525743 : bool freeze_xmin = false,
6377 9525743 : replace_xvac = false,
6378 9525743 : replace_xmax = false,
6379 9525743 : freeze_xmax = false;
6380 : TransactionId xid;
6381 :
6382 9525743 : frz->xmax = HeapTupleHeaderGetRawXmax(tuple);
6383 GIC 9525743 : frz->t_infomask2 = tuple->t_infomask2;
6384 9525743 : frz->t_infomask = tuple->t_infomask;
6385 GNC 9525743 : frz->frzflags = 0;
6386 9525743 : frz->checkflags = 0;
6387 EUB :
6388 : /*
6389 : * Process xmin, while keeping track of whether it's already frozen, or
6390 : * will become frozen iff our freeze plan is executed by caller (could be
6391 : * neither).
6392 : */
6393 GIC 9525743 : xid = HeapTupleHeaderGetXmin(tuple);
6394 9525743 : if (!TransactionIdIsNormal(xid))
6395 GNC 4010086 : xmin_already_frozen = true;
6396 ECB : else
6397 : {
6398 GNC 5515657 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
6399 UIC 0 : ereport(ERROR,
6400 : (errcode(ERRCODE_DATA_CORRUPTED),
6401 : errmsg_internal("found xmin %u from before relfrozenxid %u",
6402 : xid, cutoffs->relfrozenxid)));
6403 :
6404 : /* Will set freeze_xmin flags in freeze plan below */
6405 GNC 5515657 : freeze_xmin = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
6406 :
6407 : /* Verify that xmin committed if and when freeze plan is executed */
6408 5515657 : if (freeze_xmin)
6409 4722584 : frz->checkflags |= HEAP_FREEZE_CHECK_XMIN_COMMITTED;
6410 ECB : }
6411 :
6412 : /*
6413 : * Old-style VACUUM FULL is gone, but we have to process xvac for as long
6414 : * as we support having MOVED_OFF/MOVED_IN tuples in the database
6415 EUB : */
6416 GNC 9525743 : xid = HeapTupleHeaderGetXvac(tuple);
6417 9525743 : if (TransactionIdIsNormal(xid))
6418 : {
6419 UNC 0 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
6420 0 : Assert(TransactionIdPrecedes(xid, cutoffs->OldestXmin));
6421 :
6422 : /*
6423 : * For Xvac, we always freeze proactively. This allows totally_frozen
6424 : * tracking to ignore xvac.
6425 : */
6426 0 : replace_xvac = pagefrz->freeze_required = true;
6427 :
6428 : /* Will set replace_xvac flags in freeze plan below */
6429 : }
6430 :
6431 : /* Now process xmax */
6432 GNC 9525743 : xid = frz->xmax;
6433 GIC 9525743 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
6434 : {
6435 : /* Raw xmax is a MultiXactId */
6436 : TransactionId newxmax;
6437 ECB : uint16 flags;
6438 :
6439 : /*
6440 : * We will either remove xmax completely (in the "freeze_xmax" path),
6441 : * process xmax by replacing it (in the "replace_xmax" path), or
6442 : * perform no-op xmax processing. The only constraint is that the
6443 : * FreezeLimit/MultiXactCutoff postcondition must never be violated.
6444 : */
6445 GNC 6 : newxmax = FreezeMultiXactId(xid, tuple->t_infomask, cutoffs,
6446 : &flags, pagefrz);
6447 ECB :
6448 GNC 6 : if (flags & FRM_NOOP)
6449 : {
6450 : /*
6451 : * xmax is a MultiXactId, and nothing about it changes for now.
6452 : * This is the only case where 'freeze_required' won't have been
6453 : * set for us by FreezeMultiXactId, as well as the only case where
6454 : * neither freeze_xmax nor replace_xmax are set (given a multi).
6455 : *
6456 : * This is a no-op, but the call to FreezeMultiXactId might have
6457 : * ratcheted back NewRelfrozenXid and/or NewRelminMxid trackers
6458 : * for us (the "freeze page" variants, specifically). That'll
6459 : * make it safe for our caller to freeze the page later on, while
6460 : * leaving this particular xmax undisturbed.
6461 : *
6462 : * FreezeMultiXactId is _not_ responsible for the "no freeze"
6463 : * NewRelfrozenXid/NewRelminMxid trackers, though -- that's our
6464 : * job. A call to heap_tuple_should_freeze for this same tuple
6465 : * will take place below if 'freeze_required' isn't set already.
6466 : * (This repeats work from FreezeMultiXactId, but allows "no
6467 : * freeze" tracker maintenance to happen in only one place.)
6468 : */
6469 1 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->MultiXactCutoff));
6470 1 : Assert(MultiXactIdIsValid(newxmax) && xid == newxmax);
6471 : }
6472 5 : else if (flags & FRM_RETURN_IS_XID)
6473 : {
6474 : /*
6475 ECB : * xmax will become an updater Xid (original MultiXact's updater
6476 : * member Xid will be carried forward as a simple Xid in Xmax).
6477 EUB : */
6478 UNC 0 : Assert(!TransactionIdPrecedes(newxmax, cutoffs->OldestXmin));
6479 ECB :
6480 : /*
6481 : * NB -- some of these transformations are only valid because we
6482 : * know the return Xid is a tuple updater (i.e. not merely a
6483 : * locker.) Also note that the only reason we don't explicitly
6484 : * worry about HEAP_KEYS_UPDATED is because it lives in
6485 : * t_infomask2 rather than t_infomask.
6486 : */
6487 UIC 0 : frz->t_infomask &= ~HEAP_XMAX_BITS;
6488 0 : frz->xmax = newxmax;
6489 0 : if (flags & FRM_MARK_COMMITTED)
6490 0 : frz->t_infomask |= HEAP_XMAX_COMMITTED;
6491 UNC 0 : replace_xmax = true;
6492 EUB : }
6493 GBC 5 : else if (flags & FRM_RETURN_IS_MULTI)
6494 EUB : {
6495 : uint16 newbits;
6496 : uint16 newbits2;
6497 :
6498 ECB : /*
6499 : * xmax is an old MultiXactId that we have to replace with a new
6500 : * MultiXactId, to carry forward two or more original member XIDs.
6501 : */
6502 GNC 1 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->OldestMxact));
6503 :
6504 : /*
6505 : * We can't use GetMultiXactIdHintBits directly on the new multi
6506 : * here; that routine initializes the masks to all zeroes, which
6507 : * would lose other bits we need. Doing it this way ensures all
6508 : * unrelated bits remain untouched.
6509 ECB : */
6510 CBC 1 : frz->t_infomask &= ~HEAP_XMAX_BITS;
6511 1 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6512 1 : GetMultiXactIdHintBits(newxmax, &newbits, &newbits2);
6513 GIC 1 : frz->t_infomask |= newbits;
6514 1 : frz->t_infomask2 |= newbits2;
6515 1 : frz->xmax = newxmax;
6516 GNC 1 : replace_xmax = true;
6517 : }
6518 : else
6519 ECB : {
6520 : /*
6521 : * Freeze plan for tuple "freezes xmax" in the strictest sense:
6522 : * it'll leave nothing in xmax (neither an Xid nor a MultiXactId).
6523 : */
6524 GNC 4 : Assert(flags & FRM_INVALIDATE_XMAX);
6525 GIC 4 : Assert(!TransactionIdIsValid(newxmax));
6526 :
6527 : /* Will set freeze_xmax flags in freeze plan below */
6528 GNC 4 : freeze_xmax = true;
6529 : }
6530 :
6531 : /* MultiXactId processing forces freezing (barring FRM_NOOP case) */
6532 6 : Assert(pagefrz->freeze_required || (!freeze_xmax && !replace_xmax));
6533 : }
6534 GIC 9525737 : else if (TransactionIdIsNormal(xid))
6535 : {
6536 : /* Raw xmax is normal XID */
6537 GNC 353374 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
6538 UIC 0 : ereport(ERROR,
6539 ECB : (errcode(ERRCODE_DATA_CORRUPTED),
6540 : errmsg_internal("found xmax %u from before relfrozenxid %u",
6541 : xid, cutoffs->relfrozenxid)));
6542 :
6543 : /* Will set freeze_xmax flags in freeze plan below */
6544 GNC 353374 : freeze_xmax = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
6545 :
6546 : /*
6547 : * Verify that xmax aborted if and when freeze plan is executed,
6548 : * provided it's from an update. (A lock-only xmax can be removed
6549 : * independent of this, since the lock is released at xact end.)
6550 : */
6551 353374 : if (freeze_xmax && !HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask))
6552 219 : frz->checkflags |= HEAP_FREEZE_CHECK_XMAX_ABORTED;
6553 : }
6554 9172363 : else if (!TransactionIdIsValid(xid))
6555 : {
6556 : /* Raw xmax is InvalidTransactionId XID */
6557 9172363 : Assert((tuple->t_infomask & HEAP_XMAX_IS_MULTI) == 0);
6558 GIC 9172363 : xmax_already_frozen = true;
6559 : }
6560 ECB : else
6561 UIC 0 : ereport(ERROR,
6562 : (errcode(ERRCODE_DATA_CORRUPTED),
6563 : errmsg_internal("found raw xmax %u (infomask 0x%04x) not invalid and not multi",
6564 : xid, tuple->t_infomask)));
6565 :
6566 GNC 9525743 : if (freeze_xmin)
6567 : {
6568 4722584 : Assert(!xmin_already_frozen);
6569 :
6570 4722584 : frz->t_infomask |= HEAP_XMIN_FROZEN;
6571 : }
6572 9525743 : if (replace_xvac)
6573 : {
6574 : /*
6575 : * If a MOVED_OFF tuple is not dead, the xvac transaction must have
6576 : * failed; whereas a non-dead MOVED_IN tuple must mean the xvac
6577 : * transaction succeeded.
6578 : */
6579 UNC 0 : Assert(pagefrz->freeze_required);
6580 0 : if (tuple->t_infomask & HEAP_MOVED_OFF)
6581 0 : frz->frzflags |= XLH_INVALID_XVAC;
6582 : else
6583 0 : frz->frzflags |= XLH_FREEZE_XVAC;
6584 : }
6585 GNC 9525743 : if (replace_xmax)
6586 : {
6587 1 : Assert(!xmax_already_frozen && !freeze_xmax);
6588 1 : Assert(pagefrz->freeze_required);
6589 :
6590 : /* Already set replace_xmax flags in freeze plan earlier */
6591 : }
6592 GIC 9525743 : if (freeze_xmax)
6593 : {
6594 GNC 1058 : Assert(!xmax_already_frozen && !replace_xmax);
6595 ECB :
6596 GIC 1058 : frz->xmax = InvalidTransactionId;
6597 ECB :
6598 : /*
6599 : * The tuple might be marked either XMAX_INVALID or XMAX_COMMITTED +
6600 : * LOCKED. Normalize to INVALID just to be sure no one gets confused.
6601 : * Also get rid of the HEAP_KEYS_UPDATED bit.
6602 : */
6603 GIC 1058 : frz->t_infomask &= ~HEAP_XMAX_BITS;
6604 CBC 1058 : frz->t_infomask |= HEAP_XMAX_INVALID;
6605 GIC 1058 : frz->t_infomask2 &= ~HEAP_HOT_UPDATED;
6606 CBC 1058 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6607 ECB : }
6608 :
6609 EUB : /*
6610 : * Determine if this tuple is already totally frozen, or will become
6611 : * totally frozen (provided caller executes freeze plans for the page)
6612 : */
6613 GNC 18257355 : *totally_frozen = ((freeze_xmin || xmin_already_frozen) &&
6614 8731612 : (freeze_xmax || xmax_already_frozen));
6615 :
6616 9525743 : if (!pagefrz->freeze_required && !(xmin_already_frozen &&
6617 : xmax_already_frozen))
6618 : {
6619 : /*
6620 : * So far no previous tuple from the page made freezing mandatory.
6621 : * Does this tuple force caller to freeze the entire page?
6622 EUB : */
6623 GNC 2207959 : pagefrz->freeze_required =
6624 2207959 : heap_tuple_should_freeze(tuple, cutoffs,
6625 : &pagefrz->NoFreezePageRelfrozenXid,
6626 : &pagefrz->NoFreezePageRelminMxid);
6627 : }
6628 ECB :
6629 : /* Tell caller if this tuple has a usable freeze plan set in *frz */
6630 GNC 9525743 : return freeze_xmin || replace_xvac || replace_xmax || freeze_xmax;
6631 : }
6632 :
6633 : /*
6634 : * heap_execute_freeze_tuple
6635 : * Execute the prepared freezing of a tuple with caller's freeze plan.
6636 ECB : *
6637 : * Caller is responsible for ensuring that no other backend can access the
6638 : * storage underlying this tuple, either by holding an exclusive lock on the
6639 : * buffer containing it (which is what lazy VACUUM does), or by having it be
6640 : * in private storage (which is what CLUSTER and friends do).
6641 : */
6642 : static inline void
6643 GNC 3307177 : heap_execute_freeze_tuple(HeapTupleHeader tuple, HeapTupleFreeze *frz)
6644 : {
6645 CBC 3307177 : HeapTupleHeaderSetXmax(tuple, frz->xmax);
6646 :
6647 3307177 : if (frz->frzflags & XLH_FREEZE_XVAC)
6648 UIC 0 : HeapTupleHeaderSetXvac(tuple, FrozenTransactionId);
6649 :
6650 CBC 3307177 : if (frz->frzflags & XLH_INVALID_XVAC)
6651 LBC 0 : HeapTupleHeaderSetXvac(tuple, InvalidTransactionId);
6652 :
6653 GIC 3307177 : tuple->t_infomask = frz->t_infomask;
6654 3307177 : tuple->t_infomask2 = frz->t_infomask2;
6655 3307177 : }
6656 :
6657 : /*
6658 : * heap_freeze_execute_prepared
6659 : *
6660 : * Executes freezing of one or more heap tuples on a page on behalf of caller.
6661 : * Caller passes an array of tuple plans from heap_prepare_freeze_tuple.
6662 : * Caller must set 'offset' in each plan for us. Note that we destructively
6663 : * sort caller's tuples array in-place, so caller had better be done with it.
6664 : *
6665 : * WAL-logs the changes so that VACUUM can advance the rel's relfrozenxid
6666 : * later on without any risk of unsafe pg_xact lookups, even following a hard
6667 : * crash (or when querying from a standby). We represent freezing by setting
6668 : * infomask bits in tuple headers, but this shouldn't be thought of as a hint.
6669 : * See section on buffer access rules in src/backend/storage/buffer/README.
6670 : */
6671 : void
6672 GNC 80091 : heap_freeze_execute_prepared(Relation rel, Buffer buffer,
6673 : TransactionId snapshotConflictHorizon,
6674 : HeapTupleFreeze *tuples, int ntuples)
6675 : {
6676 80091 : Page page = BufferGetPage(buffer);
6677 :
6678 80091 : Assert(ntuples > 0);
6679 :
6680 : /*
6681 : * Perform xmin/xmax XID status sanity checks before critical section.
6682 : *
6683 : * heap_prepare_freeze_tuple doesn't perform these checks directly because
6684 : * pg_xact lookups are relatively expensive. They shouldn't be repeated
6685 : * by successive VACUUMs that each decide against freezing the same page.
6686 : */
6687 3137114 : for (int i = 0; i < ntuples; i++)
6688 : {
6689 3057023 : HeapTupleFreeze *frz = tuples + i;
6690 3057023 : ItemId itemid = PageGetItemId(page, frz->offset);
6691 : HeapTupleHeader htup;
6692 :
6693 3057023 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
6694 :
6695 : /* Deliberately avoid relying on tuple hint bits here */
6696 3057023 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMIN_COMMITTED)
6697 : {
6698 3057022 : TransactionId xmin = HeapTupleHeaderGetRawXmin(htup);
6699 :
6700 3057022 : Assert(!HeapTupleHeaderXminFrozen(htup));
6701 3057022 : if (unlikely(!TransactionIdDidCommit(xmin)))
6702 UNC 0 : ereport(ERROR,
6703 : (errcode(ERRCODE_DATA_CORRUPTED),
6704 : errmsg_internal("uncommitted xmin %u needs to be frozen",
6705 : xmin)));
6706 : }
6707 :
6708 : /*
6709 : * TransactionIdDidAbort won't work reliably in the presence of XIDs
6710 : * left behind by transactions that were in progress during a crash,
6711 : * so we can only check that xmax didn't commit
6712 : */
6713 GNC 3057023 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMAX_ABORTED)
6714 : {
6715 16 : TransactionId xmax = HeapTupleHeaderGetRawXmax(htup);
6716 :
6717 16 : Assert(TransactionIdIsNormal(xmax));
6718 16 : if (unlikely(TransactionIdDidCommit(xmax)))
6719 UNC 0 : ereport(ERROR,
6720 : (errcode(ERRCODE_DATA_CORRUPTED),
6721 : errmsg_internal("cannot freeze committed xmax %u",
6722 : xmax)));
6723 : }
6724 : }
6725 :
6726 GNC 80091 : START_CRIT_SECTION();
6727 :
6728 3137114 : for (int i = 0; i < ntuples; i++)
6729 : {
6730 3057023 : HeapTupleFreeze *frz = tuples + i;
6731 3057023 : ItemId itemid = PageGetItemId(page, frz->offset);
6732 : HeapTupleHeader htup;
6733 :
6734 3057023 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
6735 3057023 : heap_execute_freeze_tuple(htup, frz);
6736 : }
6737 :
6738 80091 : MarkBufferDirty(buffer);
6739 :
6740 : /* Now WAL-log freezing if necessary */
6741 80091 : if (RelationNeedsWAL(rel))
6742 : {
6743 : xl_heap_freeze_plan plans[MaxHeapTuplesPerPage];
6744 : OffsetNumber offsets[MaxHeapTuplesPerPage];
6745 : int nplans;
6746 : xl_heap_freeze_page xlrec;
6747 : XLogRecPtr recptr;
6748 :
6749 : /* Prepare deduplicated representation for use in WAL record */
6750 80089 : nplans = heap_log_freeze_plan(tuples, ntuples, plans, offsets);
6751 :
6752 80089 : xlrec.snapshotConflictHorizon = snapshotConflictHorizon;
6753 80089 : xlrec.isCatalogRel = RelationIsAccessibleInLogicalDecoding(rel);
6754 80089 : xlrec.nplans = nplans;
6755 :
6756 80089 : XLogBeginInsert();
6757 80089 : XLogRegisterData((char *) &xlrec, SizeOfHeapFreezePage);
6758 :
6759 : /*
6760 : * The freeze plan array and offset array are not actually in the
6761 : * buffer, but pretend that they are. When XLogInsert stores the
6762 : * whole buffer, the arrays need not be stored too.
6763 : */
6764 80089 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6765 80089 : XLogRegisterBufData(0, (char *) plans,
6766 : nplans * sizeof(xl_heap_freeze_plan));
6767 80089 : XLogRegisterBufData(0, (char *) offsets,
6768 : ntuples * sizeof(OffsetNumber));
6769 :
6770 80089 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_FREEZE_PAGE);
6771 :
6772 80089 : PageSetLSN(page, recptr);
6773 : }
6774 :
6775 80091 : END_CRIT_SECTION();
6776 80091 : }
6777 :
6778 : /*
6779 : * Comparator used to deduplicate XLOG_HEAP2_FREEZE_PAGE freeze plans
6780 : */
6781 : static int
6782 3873335 : heap_log_freeze_cmp(const void *arg1, const void *arg2)
6783 : {
6784 3873335 : HeapTupleFreeze *frz1 = (HeapTupleFreeze *) arg1;
6785 3873335 : HeapTupleFreeze *frz2 = (HeapTupleFreeze *) arg2;
6786 :
6787 3873335 : if (frz1->xmax < frz2->xmax)
6788 3 : return -1;
6789 3873332 : else if (frz1->xmax > frz2->xmax)
6790 13 : return 1;
6791 :
6792 3873319 : if (frz1->t_infomask2 < frz2->t_infomask2)
6793 8108 : return -1;
6794 3865211 : else if (frz1->t_infomask2 > frz2->t_infomask2)
6795 22015 : return 1;
6796 :
6797 3843196 : if (frz1->t_infomask < frz2->t_infomask)
6798 51398 : return -1;
6799 3791798 : else if (frz1->t_infomask > frz2->t_infomask)
6800 86119 : return 1;
6801 :
6802 3705679 : if (frz1->frzflags < frz2->frzflags)
6803 UNC 0 : return -1;
6804 GNC 3705679 : else if (frz1->frzflags > frz2->frzflags)
6805 UNC 0 : return 1;
6806 :
6807 : /*
6808 : * heap_log_freeze_eq would consider these tuple-wise plans to be equal.
6809 : * (So the tuples will share a single canonical freeze plan.)
6810 : *
6811 : * We tiebreak on page offset number to keep each freeze plan's page
6812 : * offset number array individually sorted. (Unnecessary, but be tidy.)
6813 : */
6814 GNC 3705679 : if (frz1->offset < frz2->offset)
6815 3313703 : return -1;
6816 391976 : else if (frz1->offset > frz2->offset)
6817 391976 : return 1;
6818 :
6819 UNC 0 : Assert(false);
6820 : return 0;
6821 : }
6822 :
6823 : /*
6824 : * Compare fields that describe actions required to freeze tuple with caller's
6825 : * open plan. If everything matches then the frz tuple plan is equivalent to
6826 : * caller's plan.
6827 : */
6828 : static inline bool
6829 GNC 2976932 : heap_log_freeze_eq(xl_heap_freeze_plan *plan, HeapTupleFreeze *frz)
6830 : {
6831 2976932 : if (plan->xmax == frz->xmax &&
6832 2976927 : plan->t_infomask2 == frz->t_infomask2 &&
6833 2973857 : plan->t_infomask == frz->t_infomask &&
6834 2961624 : plan->frzflags == frz->frzflags)
6835 2961624 : return true;
6836 :
6837 : /* Caller must call heap_log_freeze_new_plan again for frz */
6838 15308 : return false;
6839 : }
6840 :
6841 : /*
6842 : * Start new plan initialized using tuple-level actions. At least one tuple
6843 : * will have steps required to freeze described by caller's plan during REDO.
6844 : */
6845 : static inline void
6846 95397 : heap_log_freeze_new_plan(xl_heap_freeze_plan *plan, HeapTupleFreeze *frz)
6847 : {
6848 95397 : plan->xmax = frz->xmax;
6849 95397 : plan->t_infomask2 = frz->t_infomask2;
6850 95397 : plan->t_infomask = frz->t_infomask;
6851 95397 : plan->frzflags = frz->frzflags;
6852 95397 : plan->ntuples = 1; /* for now */
6853 95397 : }
6854 :
6855 : /*
6856 : * Deduplicate tuple-based freeze plans so that each distinct set of
6857 : * processing steps is only stored once in XLOG_HEAP2_FREEZE_PAGE records.
6858 : * Called during original execution of freezing (for logged relations).
6859 : *
6860 : * Return value is number of plans set in *plans_out for caller. Also writes
6861 : * an array of offset numbers into *offsets_out output argument for caller
6862 : * (actually there is one array per freeze plan, but that's not of immediate
6863 : * concern to our caller).
6864 : */
6865 : static int
6866 80089 : heap_log_freeze_plan(HeapTupleFreeze *tuples, int ntuples,
6867 : xl_heap_freeze_plan *plans_out,
6868 : OffsetNumber *offsets_out)
6869 : {
6870 80089 : int nplans = 0;
6871 :
6872 : /* Sort tuple-based freeze plans in the order required to deduplicate */
6873 80089 : qsort(tuples, ntuples, sizeof(HeapTupleFreeze), heap_log_freeze_cmp);
6874 :
6875 3137110 : for (int i = 0; i < ntuples; i++)
6876 : {
6877 3057021 : HeapTupleFreeze *frz = tuples + i;
6878 :
6879 3057021 : if (i == 0)
6880 : {
6881 : /* New canonical freeze plan starting with first tup */
6882 80089 : heap_log_freeze_new_plan(plans_out, frz);
6883 80089 : nplans++;
6884 : }
6885 2976932 : else if (heap_log_freeze_eq(plans_out, frz))
6886 : {
6887 : /* tup matches open canonical plan -- include tup in it */
6888 2961624 : Assert(offsets_out[i - 1] < frz->offset);
6889 2961624 : plans_out->ntuples++;
6890 : }
6891 : else
6892 : {
6893 : /* Tup doesn't match current plan -- done with it now */
6894 15308 : plans_out++;
6895 :
6896 : /* New canonical freeze plan starting with this tup */
6897 15308 : heap_log_freeze_new_plan(plans_out, frz);
6898 15308 : nplans++;
6899 : }
6900 :
6901 : /*
6902 : * Save page offset number in dedicated buffer in passing.
6903 : *
6904 : * REDO routine relies on the record's offset numbers array grouping
6905 : * offset numbers by freeze plan. The sort order within each grouping
6906 : * is ascending offset number order, just to keep things tidy.
6907 : */
6908 3057021 : offsets_out[i] = frz->offset;
6909 : }
6910 :
6911 80089 : Assert(nplans > 0 && nplans <= ntuples);
6912 :
6913 80089 : return nplans;
6914 : }
6915 :
6916 ECB : /*
6917 : * heap_freeze_tuple
6918 : * Freeze tuple in place, without WAL logging.
6919 : *
6920 : * Useful for callers like CLUSTER that perform their own WAL logging.
6921 : */
6922 : bool
6923 CBC 384627 : heap_freeze_tuple(HeapTupleHeader tuple,
6924 ECB : TransactionId relfrozenxid, TransactionId relminmxid,
6925 : TransactionId FreezeLimit, TransactionId MultiXactCutoff)
6926 : {
6927 : HeapTupleFreeze frz;
6928 : bool do_freeze;
6929 : bool totally_frozen;
6930 : struct VacuumCutoffs cutoffs;
6931 : HeapPageFreeze pagefrz;
6932 :
6933 GNC 384627 : cutoffs.relfrozenxid = relfrozenxid;
6934 384627 : cutoffs.relminmxid = relminmxid;
6935 384627 : cutoffs.OldestXmin = FreezeLimit;
6936 384627 : cutoffs.OldestMxact = MultiXactCutoff;
6937 384627 : cutoffs.FreezeLimit = FreezeLimit;
6938 384627 : cutoffs.MultiXactCutoff = MultiXactCutoff;
6939 :
6940 384627 : pagefrz.freeze_required = true;
6941 384627 : pagefrz.FreezePageRelfrozenXid = FreezeLimit;
6942 384627 : pagefrz.FreezePageRelminMxid = MultiXactCutoff;
6943 384627 : pagefrz.NoFreezePageRelfrozenXid = FreezeLimit;
6944 384627 : pagefrz.NoFreezePageRelminMxid = MultiXactCutoff;
6945 :
6946 384627 : do_freeze = heap_prepare_freeze_tuple(tuple, &cutoffs,
6947 : &pagefrz, &frz, &totally_frozen);
6948 ECB :
6949 EUB : /*
6950 : * Note that because this is not a WAL-logged operation, we don't need to
6951 : * fill in the offset in the freeze record.
6952 : */
6953 :
6954 GIC 384627 : if (do_freeze)
6955 247765 : heap_execute_freeze_tuple(tuple, &frz);
6956 384627 : return do_freeze;
6957 : }
6958 ECB :
6959 : /*
6960 : * For a given MultiXactId, return the hint bits that should be set in the
6961 : * tuple's infomask.
6962 : *
6963 EUB : * Normally this should be called for a multixact that was just created, and
6964 : * so is on our local cache, so the GetMembers call is fast.
6965 : */
6966 : static void
6967 GIC 1170 : GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
6968 : uint16 *new_infomask2)
6969 : {
6970 : int nmembers;
6971 : MultiXactMember *members;
6972 : int i;
6973 CBC 1170 : uint16 bits = HEAP_XMAX_IS_MULTI;
6974 GIC 1170 : uint16 bits2 = 0;
6975 CBC 1170 : bool has_update = false;
6976 1170 : LockTupleMode strongest = LockTupleKeyShare;
6977 ECB :
6978 : /*
6979 : * We only use this in multis we just created, so they cannot be values
6980 : * pre-pg_upgrade.
6981 : */
6982 CBC 1170 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
6983 :
6984 GIC 3579 : for (i = 0; i < nmembers; i++)
6985 : {
6986 : LockTupleMode mode;
6987 :
6988 : /*
6989 : * Remember the strongest lock mode held by any member of the
6990 ECB : * multixact.
6991 : */
6992 CBC 2409 : mode = TUPLOCK_from_mxstatus(members[i].status);
6993 2409 : if (mode > strongest)
6994 654 : strongest = mode;
6995 ECB :
6996 : /* See what other bits we need */
6997 CBC 2409 : switch (members[i].status)
6998 : {
6999 GIC 2218 : case MultiXactStatusForKeyShare:
7000 : case MultiXactStatusForShare:
7001 : case MultiXactStatusForNoKeyUpdate:
7002 2218 : break;
7003 :
7004 52 : case MultiXactStatusForUpdate:
7005 52 : bits2 |= HEAP_KEYS_UPDATED;
7006 52 : break;
7007 :
7008 129 : case MultiXactStatusNoKeyUpdate:
7009 129 : has_update = true;
7010 CBC 129 : break;
7011 :
7012 GIC 10 : case MultiXactStatusUpdate:
7013 10 : bits2 |= HEAP_KEYS_UPDATED;
7014 CBC 10 : has_update = true;
7015 GIC 10 : break;
7016 : }
7017 ECB : }
7018 :
7019 CBC 1170 : if (strongest == LockTupleExclusive ||
7020 : strongest == LockTupleNoKeyExclusive)
7021 216 : bits |= HEAP_XMAX_EXCL_LOCK;
7022 GIC 954 : else if (strongest == LockTupleShare)
7023 CBC 435 : bits |= HEAP_XMAX_SHR_LOCK;
7024 GIC 519 : else if (strongest == LockTupleKeyShare)
7025 519 : bits |= HEAP_XMAX_KEYSHR_LOCK;
7026 ECB :
7027 CBC 1170 : if (!has_update)
7028 GIC 1031 : bits |= HEAP_XMAX_LOCK_ONLY;
7029 ECB :
7030 GIC 1170 : if (nmembers > 0)
7031 1170 : pfree(members);
7032 ECB :
7033 CBC 1170 : *new_infomask = bits;
7034 GIC 1170 : *new_infomask2 = bits2;
7035 1170 : }
7036 :
7037 : /*
7038 ECB : * MultiXactIdGetUpdateXid
7039 : *
7040 : * Given a multixact Xmax and corresponding infomask, which does not have the
7041 : * HEAP_XMAX_LOCK_ONLY bit set, obtain and return the Xid of the updating
7042 : * transaction.
7043 : *
7044 : * Caller is expected to check the status of the updating transaction, if
7045 : * necessary.
7046 : */
7047 : static TransactionId
7048 GIC 526 : MultiXactIdGetUpdateXid(TransactionId xmax, uint16 t_infomask)
7049 : {
7050 526 : TransactionId update_xact = InvalidTransactionId;
7051 : MultiXactMember *members;
7052 ECB : int nmembers;
7053 :
7054 GIC 526 : Assert(!(t_infomask & HEAP_XMAX_LOCK_ONLY));
7055 CBC 526 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
7056 :
7057 ECB : /*
7058 : * Since we know the LOCK_ONLY bit is not set, this cannot be a multi from
7059 : * pre-pg_upgrade.
7060 : */
7061 GIC 526 : nmembers = GetMultiXactIdMembers(xmax, &members, false, false);
7062 :
7063 526 : if (nmembers > 0)
7064 : {
7065 : int i;
7066 :
7067 CBC 1978 : for (i = 0; i < nmembers; i++)
7068 : {
7069 : /* Ignore lockers */
7070 GIC 1452 : if (!ISUPDATE_from_mxstatus(members[i].status))
7071 926 : continue;
7072 :
7073 : /* there can be at most one updater */
7074 526 : Assert(update_xact == InvalidTransactionId);
7075 526 : update_xact = members[i].xid;
7076 : #ifndef USE_ASSERT_CHECKING
7077 ECB :
7078 : /*
7079 : * in an assert-enabled build, walk the whole array to ensure
7080 : * there's no other updater.
7081 : */
7082 : break;
7083 : #endif
7084 : }
7085 :
7086 CBC 526 : pfree(members);
7087 ECB : }
7088 :
7089 GIC 526 : return update_xact;
7090 ECB : }
7091 :
7092 : /*
7093 : * HeapTupleGetUpdateXid
7094 : * As above, but use a HeapTupleHeader
7095 : *
7096 : * See also HeapTupleHeaderGetUpdateXid, which can be used without previously
7097 : * checking the hint bits.
7098 : */
7099 : TransactionId
7100 CBC 518 : HeapTupleGetUpdateXid(HeapTupleHeader tuple)
7101 : {
7102 GIC 1036 : return MultiXactIdGetUpdateXid(HeapTupleHeaderGetRawXmax(tuple),
7103 518 : tuple->t_infomask);
7104 : }
7105 :
7106 : /*
7107 : * Does the given multixact conflict with the current transaction grabbing a
7108 : * tuple lock of the given strength?
7109 : *
7110 : * The passed infomask pairs up with the given multixact in the tuple header.
7111 ECB : *
7112 : * If current_is_member is not NULL, it is set to 'true' if the current
7113 : * transaction is a member of the given multixact.
7114 : */
7115 : static bool
7116 GIC 94 : DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
7117 ECB : LockTupleMode lockmode, bool *current_is_member)
7118 : {
7119 : int nmembers;
7120 : MultiXactMember *members;
7121 GIC 94 : bool result = false;
7122 94 : LOCKMODE wanted = tupleLockExtraInfo[lockmode].hwlock;
7123 :
7124 94 : if (HEAP_LOCKED_UPGRADED(infomask))
7125 UIC 0 : return false;
7126 ECB :
7127 GIC 94 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7128 CBC 94 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7129 GIC 94 : if (nmembers >= 0)
7130 : {
7131 : int i;
7132 :
7133 295 : for (i = 0; i < nmembers; i++)
7134 : {
7135 : TransactionId memxid;
7136 ECB : LOCKMODE memlockmode;
7137 :
7138 CBC 207 : if (result && (current_is_member == NULL || *current_is_member))
7139 : break;
7140 :
7141 201 : memlockmode = LOCKMODE_from_mxstatus(members[i].status);
7142 :
7143 ECB : /* ignore members from current xact (but track their presence) */
7144 GIC 201 : memxid = members[i].xid;
7145 201 : if (TransactionIdIsCurrentTransactionId(memxid))
7146 ECB : {
7147 GIC 91 : if (current_is_member != NULL)
7148 CBC 78 : *current_is_member = true;
7149 91 : continue;
7150 ECB : }
7151 GIC 110 : else if (result)
7152 CBC 8 : continue;
7153 ECB :
7154 : /* ignore members that don't conflict with the lock we want */
7155 GIC 102 : if (!DoLockModesConflict(memlockmode, wanted))
7156 CBC 67 : continue;
7157 ECB :
7158 CBC 35 : if (ISUPDATE_from_mxstatus(members[i].status))
7159 ECB : {
7160 : /* ignore aborted updaters */
7161 GIC 17 : if (TransactionIdDidAbort(memxid))
7162 1 : continue;
7163 ECB : }
7164 : else
7165 : {
7166 : /* ignore lockers-only that are no longer in progress */
7167 CBC 18 : if (!TransactionIdIsInProgress(memxid))
7168 5 : continue;
7169 ECB : }
7170 :
7171 : /*
7172 : * Whatever remains are either live lockers that conflict with our
7173 : * wanted lock, and updaters that are not aborted. Those conflict
7174 : * with what we want. Set up to return true, but keep going to
7175 : * look for the current transaction among the multixact members,
7176 : * if needed.
7177 : */
7178 CBC 29 : result = true;
7179 ECB : }
7180 GIC 94 : pfree(members);
7181 : }
7182 :
7183 94 : return result;
7184 : }
7185 :
7186 : /*
7187 : * Do_MultiXactIdWait
7188 : * Actual implementation for the two functions below.
7189 : *
7190 : * 'multi', 'status' and 'infomask' indicate what to sleep on (the status is
7191 : * needed to ensure we only sleep on conflicting members, and the infomask is
7192 ECB : * used to optimize multixact access in case it's a lock-only multi); 'nowait'
7193 : * indicates whether to use conditional lock acquisition, to allow callers to
7194 : * fail if lock is unavailable. 'rel', 'ctid' and 'oper' are used to set up
7195 : * context information for error messages. 'remaining', if not NULL, receives
7196 : * the number of members that are still running, including any (non-aborted)
7197 : * subtransactions of our own transaction.
7198 : *
7199 : * We do this by sleeping on each member using XactLockTableWait. Any
7200 : * members that belong to the current backend are *not* waited for, however;
7201 : * this would not merely be useless but would lead to Assert failure inside
7202 : * XactLockTableWait. By the time this returns, it is certain that all
7203 : * transactions *of other backends* that were members of the MultiXactId
7204 : * that conflict with the requested status are dead (and no new ones can have
7205 : * been added, since it is not legal to add members to an existing
7206 : * MultiXactId).
7207 : *
7208 : * But by the time we finish sleeping, someone else may have changed the Xmax
7209 : * of the containing tuple, so the caller needs to iterate on us somehow.
7210 : *
7211 : * Note that in case we return false, the number of remaining members is
7212 : * not to be trusted.
7213 : */
7214 : static bool
7215 CBC 56 : Do_MultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7216 : uint16 infomask, bool nowait,
7217 : Relation rel, ItemPointer ctid, XLTW_Oper oper,
7218 ECB : int *remaining)
7219 : {
7220 GIC 56 : bool result = true;
7221 : MultiXactMember *members;
7222 : int nmembers;
7223 56 : int remain = 0;
7224 :
7225 : /* for pre-pg_upgrade tuples, no need to sleep at all */
7226 56 : nmembers = HEAP_LOCKED_UPGRADED(infomask) ? -1 :
7227 56 : GetMultiXactIdMembers(multi, &members, false,
7228 56 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7229 :
7230 CBC 56 : if (nmembers >= 0)
7231 : {
7232 : int i;
7233 ECB :
7234 GIC 181 : for (i = 0; i < nmembers; i++)
7235 : {
7236 129 : TransactionId memxid = members[i].xid;
7237 129 : MultiXactStatus memstatus = members[i].status;
7238 :
7239 129 : if (TransactionIdIsCurrentTransactionId(memxid))
7240 : {
7241 24 : remain++;
7242 24 : continue;
7243 : }
7244 ECB :
7245 GIC 105 : if (!DoLockModesConflict(LOCKMODE_from_mxstatus(memstatus),
7246 CBC 105 : LOCKMODE_from_mxstatus(status)))
7247 ECB : {
7248 GIC 20 : if (remaining && TransactionIdIsInProgress(memxid))
7249 6 : remain++;
7250 20 : continue;
7251 : }
7252 :
7253 : /*
7254 : * This member conflicts with our multi, so we have to sleep (or
7255 : * return failure, if asked to avoid waiting.)
7256 : *
7257 : * Note that we don't set up an error context callback ourselves,
7258 : * but instead we pass the info down to XactLockTableWait. This
7259 : * might seem a bit wasteful because the context is set up and
7260 ECB : * tore down for each member of the multixact, but in reality it
7261 : * should be barely noticeable, and it avoids duplicate code.
7262 : */
7263 GIC 85 : if (nowait)
7264 : {
7265 CBC 4 : result = ConditionalXactLockTableWait(memxid);
7266 4 : if (!result)
7267 GIC 4 : break;
7268 ECB : }
7269 EUB : else
7270 GIC 81 : XactLockTableWait(memxid, rel, ctid, oper);
7271 ECB : }
7272 :
7273 CBC 56 : pfree(members);
7274 : }
7275 :
7276 GIC 56 : if (remaining)
7277 CBC 8 : *remaining = remain;
7278 :
7279 GIC 56 : return result;
7280 : }
7281 :
7282 ECB : /*
7283 : * MultiXactIdWait
7284 : * Sleep on a MultiXactId.
7285 : *
7286 : * By the time we finish sleeping, someone else may have changed the Xmax
7287 : * of the containing tuple, so the caller needs to iterate on us somehow.
7288 : *
7289 : * We return (in *remaining, if not NULL) the number of members that are still
7290 : * running, including any (non-aborted) subtransactions of our own transaction.
7291 : */
7292 : static void
7293 CBC 52 : MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
7294 : Relation rel, ItemPointer ctid, XLTW_Oper oper,
7295 ECB : int *remaining)
7296 : {
7297 GIC 52 : (void) Do_MultiXactIdWait(multi, status, infomask, false,
7298 : rel, ctid, oper, remaining);
7299 CBC 52 : }
7300 ECB :
7301 : /*
7302 : * ConditionalMultiXactIdWait
7303 : * As above, but only lock if we can get the lock without blocking.
7304 : *
7305 : * By the time we finish sleeping, someone else may have changed the Xmax
7306 : * of the containing tuple, so the caller needs to iterate on us somehow.
7307 : *
7308 : * If the multixact is now all gone, return true. Returns false if some
7309 : * transactions might still be running.
7310 : *
7311 : * We return (in *remaining, if not NULL) the number of members that are still
7312 : * running, including any (non-aborted) subtransactions of our own transaction.
7313 : */
7314 : static bool
7315 GIC 4 : ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7316 : uint16 infomask, Relation rel, int *remaining)
7317 : {
7318 4 : return Do_MultiXactIdWait(multi, status, infomask, true,
7319 : rel, NULL, XLTW_None, remaining);
7320 : }
7321 :
7322 ECB : /*
7323 : * heap_tuple_needs_eventual_freeze
7324 : *
7325 : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7326 : * will eventually require freezing (if tuple isn't removed by pruning first).
7327 : */
7328 : bool
7329 GIC 7027764 : heap_tuple_needs_eventual_freeze(HeapTupleHeader tuple)
7330 : {
7331 : TransactionId xid;
7332 :
7333 : /*
7334 : * If xmin is a normal transaction ID, this tuple is definitely not
7335 : * frozen.
7336 : */
7337 7027764 : xid = HeapTupleHeaderGetXmin(tuple);
7338 7027764 : if (TransactionIdIsNormal(xid))
7339 13376 : return true;
7340 :
7341 : /*
7342 : * If xmax is a valid xact or multixact, this tuple is also not frozen.
7343 : */
7344 7014388 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7345 : {
7346 : MultiXactId multi;
7347 :
7348 2 : multi = HeapTupleHeaderGetRawXmax(tuple);
7349 2 : if (MultiXactIdIsValid(multi))
7350 2 : return true;
7351 : }
7352 : else
7353 : {
7354 7014386 : xid = HeapTupleHeaderGetRawXmax(tuple);
7355 7014386 : if (TransactionIdIsNormal(xid))
7356 4 : return true;
7357 : }
7358 :
7359 CBC 7014382 : if (tuple->t_infomask & HEAP_MOVED)
7360 : {
7361 UIC 0 : xid = HeapTupleHeaderGetXvac(tuple);
7362 0 : if (TransactionIdIsNormal(xid))
7363 0 : return true;
7364 ECB : }
7365 :
7366 GIC 7014382 : return false;
7367 ECB : }
7368 :
7369 : /*
7370 : * heap_tuple_should_freeze
7371 : *
7372 : * Return value indicates if heap_prepare_freeze_tuple sibling function would
7373 : * (or should) force freezing of the heap page that contains caller's tuple.
7374 : * Tuple header XIDs/MXIDs < FreezeLimit/MultiXactCutoff trigger freezing.
7375 : * This includes (xmin, xmax, xvac) fields, as well as MultiXact member XIDs.
7376 : *
7377 : * The *NoFreezePageRelfrozenXid and *NoFreezePageRelminMxid input/output
7378 : * arguments help VACUUM track the oldest extant XID/MXID remaining in rel.
7379 : * Our working assumption is that caller won't decide to freeze this tuple.
7380 : * It's up to caller to only ratchet back its own top-level trackers after the
7381 : * point that it fully commits to not freezing the tuple/page in question.
7382 : */
7383 : bool
7384 GNC 2208045 : heap_tuple_should_freeze(HeapTupleHeader tuple,
7385 : const struct VacuumCutoffs *cutoffs,
7386 : TransactionId *NoFreezePageRelfrozenXid,
7387 : MultiXactId *NoFreezePageRelminMxid)
7388 ECB : {
7389 : TransactionId xid;
7390 : MultiXactId multi;
7391 GNC 2208045 : bool freeze = false;
7392 ECB :
7393 : /* First deal with xmin */
7394 CBC 2208045 : xid = HeapTupleHeaderGetXmin(tuple);
7395 2208045 : if (TransactionIdIsNormal(xid))
7396 ECB : {
7397 GNC 2207897 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7398 2207897 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7399 80949 : *NoFreezePageRelfrozenXid = xid;
7400 2207897 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7401 79971 : freeze = true;
7402 : }
7403 :
7404 : /* Now deal with xmax */
7405 GIC 2208045 : xid = InvalidTransactionId;
7406 2208045 : multi = InvalidMultiXactId;
7407 2208045 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7408 2 : multi = HeapTupleHeaderGetRawXmax(tuple);
7409 : else
7410 CBC 2208043 : xid = HeapTupleHeaderGetRawXmax(tuple);
7411 :
7412 2208045 : if (TransactionIdIsNormal(xid))
7413 ECB : {
7414 GNC 320768 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7415 ECB : /* xmax is a non-permanent XID */
7416 GNC 320768 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7417 4 : *NoFreezePageRelfrozenXid = xid;
7418 320768 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7419 1 : freeze = true;
7420 : }
7421 CBC 1887277 : else if (!MultiXactIdIsValid(multi))
7422 : {
7423 : /* xmax is a permanent XID or invalid MultiXactId/XID */
7424 ECB : }
7425 CBC 2 : else if (HEAP_LOCKED_UPGRADED(tuple->t_infomask))
7426 : {
7427 ECB : /* xmax is a pg_upgrade'd MultiXact, which can't have updater XID */
7428 UNC 0 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7429 0 : *NoFreezePageRelminMxid = multi;
7430 : /* heap_prepare_freeze_tuple always freezes pg_upgrade'd xmax */
7431 0 : freeze = true;
7432 : }
7433 : else
7434 : {
7435 : /* xmax is a MultiXactId that may have an updater XID */
7436 : MultiXactMember *members;
7437 : int nmembers;
7438 :
7439 GNC 2 : Assert(MultiXactIdPrecedesOrEquals(cutoffs->relminmxid, multi));
7440 2 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7441 2 : *NoFreezePageRelminMxid = multi;
7442 2 : if (MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff))
7443 2 : freeze = true;
7444 :
7445 : /* need to check whether any member of the mxact is old */
7446 CBC 2 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7447 GIC 2 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask));
7448 ECB :
7449 GIC 5 : for (int i = 0; i < nmembers; i++)
7450 : {
7451 3 : xid = members[i].xid;
7452 GNC 3 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7453 3 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7454 UNC 0 : *NoFreezePageRelfrozenXid = xid;
7455 GNC 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7456 UNC 0 : freeze = true;
7457 : }
7458 GIC 2 : if (nmembers > 0)
7459 1 : pfree(members);
7460 : }
7461 :
7462 2208045 : if (tuple->t_infomask & HEAP_MOVED)
7463 : {
7464 LBC 0 : xid = HeapTupleHeaderGetXvac(tuple);
7465 UIC 0 : if (TransactionIdIsNormal(xid))
7466 : {
7467 UNC 0 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7468 0 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7469 0 : *NoFreezePageRelfrozenXid = xid;
7470 : /* heap_prepare_freeze_tuple forces xvac freezing */
7471 0 : freeze = true;
7472 : }
7473 : }
7474 :
7475 GNC 2208045 : return freeze;
7476 : }
7477 :
7478 : /*
7479 : * Maintain snapshotConflictHorizon for caller by ratcheting forward its value
7480 : * using any committed XIDs contained in 'tuple', an obsolescent heap tuple
7481 : * that caller is in the process of physically removing, e.g. via HOT pruning
7482 : * or index deletion.
7483 : *
7484 : * Caller must initialize its value to InvalidTransactionId, which is
7485 : * generally interpreted as "definitely no need for a recovery conflict".
7486 : * Final value must reflect all heap tuples that caller will physically remove
7487 : * (or remove TID references to) via its ongoing pruning/deletion operation.
7488 : * ResolveRecoveryConflictWithSnapshot() is passed the final value (taken from
7489 : * caller's WAL record) by REDO routine when it replays caller's operation.
7490 : */
7491 : void
7492 1575911 : HeapTupleHeaderAdvanceConflictHorizon(HeapTupleHeader tuple,
7493 : TransactionId *snapshotConflictHorizon)
7494 ECB : {
7495 CBC 1575911 : TransactionId xmin = HeapTupleHeaderGetXmin(tuple);
7496 GIC 1575911 : TransactionId xmax = HeapTupleHeaderGetUpdateXid(tuple);
7497 1575911 : TransactionId xvac = HeapTupleHeaderGetXvac(tuple);
7498 :
7499 1575911 : if (tuple->t_infomask & HEAP_MOVED)
7500 ECB : {
7501 UNC 0 : if (TransactionIdPrecedes(*snapshotConflictHorizon, xvac))
7502 0 : *snapshotConflictHorizon = xvac;
7503 : }
7504 ECB :
7505 : /*
7506 : * Ignore tuples inserted by an aborted transaction or if the tuple was
7507 : * updated/deleted by the inserting transaction.
7508 : *
7509 : * Look for a committed hint bit, or if no xmin bit is set, check clog.
7510 : */
7511 CBC 1575911 : if (HeapTupleHeaderXminCommitted(tuple) ||
7512 52461 : (!HeapTupleHeaderXminInvalid(tuple) && TransactionIdDidCommit(xmin)))
7513 : {
7514 GIC 2852914 : if (xmax != xmin &&
7515 GNC 1328708 : TransactionIdFollows(xmax, *snapshotConflictHorizon))
7516 165059 : *snapshotConflictHorizon = xmax;
7517 EUB : }
7518 GIC 1575911 : }
7519 :
7520 ECB : #ifdef USE_PREFETCH
7521 : /*
7522 : * Helper function for heap_index_delete_tuples. Issues prefetch requests for
7523 : * prefetch_count buffers. The prefetch_state keeps track of all the buffers
7524 : * we can prefetch, and which have already been prefetched; each call to this
7525 : * function picks up where the previous call left off.
7526 : *
7527 : * Note: we expect the deltids array to be sorted in an order that groups TIDs
7528 : * by heap block, with all TIDs for each block appearing together in exactly
7529 : * one group.
7530 : */
7531 : static void
7532 GIC 25415 : index_delete_prefetch_buffer(Relation rel,
7533 : IndexDeletePrefetchState *prefetch_state,
7534 : int prefetch_count)
7535 : {
7536 25415 : BlockNumber cur_hblkno = prefetch_state->cur_hblkno;
7537 25415 : int count = 0;
7538 ECB : int i;
7539 GIC 25415 : int ndeltids = prefetch_state->ndeltids;
7540 25415 : TM_IndexDelete *deltids = prefetch_state->deltids;
7541 :
7542 25415 : for (i = prefetch_state->next_item;
7543 879445 : i < ndeltids && count < prefetch_count;
7544 854030 : i++)
7545 ECB : {
7546 GIC 854030 : ItemPointer htid = &deltids[i].tid;
7547 :
7548 CBC 1699780 : if (cur_hblkno == InvalidBlockNumber ||
7549 845750 : ItemPointerGetBlockNumber(htid) != cur_hblkno)
7550 : {
7551 25166 : cur_hblkno = ItemPointerGetBlockNumber(htid);
7552 25166 : PrefetchBuffer(rel, MAIN_FORKNUM, cur_hblkno);
7553 25166 : count++;
7554 ECB : }
7555 : }
7556 :
7557 : /*
7558 : * Save the prefetch position so that next time we can continue from that
7559 : * position.
7560 : */
7561 CBC 25415 : prefetch_state->next_item = i;
7562 25415 : prefetch_state->cur_hblkno = cur_hblkno;
7563 GIC 25415 : }
7564 ECB : #endif
7565 :
7566 : /*
7567 : * Helper function for heap_index_delete_tuples. Checks for index corruption
7568 : * involving an invalid TID in index AM caller's index page.
7569 : *
7570 : * This is an ideal place for these checks. The index AM must hold a buffer
7571 : * lock on the index page containing the TIDs we examine here, so we don't
7572 : * have to worry about concurrent VACUUMs at all. We can be sure that the
7573 : * index is corrupt when htid points directly to an LP_UNUSED item or
7574 : * heap-only tuple, which is not the case during standard index scans.
7575 : */
7576 : static inline void
7577 GIC 621751 : index_delete_check_htid(TM_IndexDeleteOp *delstate,
7578 : Page page, OffsetNumber maxoff,
7579 ECB : ItemPointer htid, TM_IndexStatus *istatus)
7580 : {
7581 GIC 621751 : OffsetNumber indexpagehoffnum = ItemPointerGetOffsetNumber(htid);
7582 EUB : ItemId iid;
7583 :
7584 GIC 621751 : Assert(OffsetNumberIsValid(istatus->idxoffnum));
7585 EUB :
7586 GIC 621751 : if (unlikely(indexpagehoffnum > maxoff))
7587 UIC 0 : ereport(ERROR,
7588 : (errcode(ERRCODE_INDEX_CORRUPTED),
7589 : 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\"",
7590 : ItemPointerGetBlockNumber(htid),
7591 : indexpagehoffnum,
7592 : istatus->idxoffnum, delstate->iblknum,
7593 ECB : RelationGetRelationName(delstate->irel))));
7594 :
7595 CBC 621751 : iid = PageGetItemId(page, indexpagehoffnum);
7596 621751 : if (unlikely(!ItemIdIsUsed(iid)))
7597 LBC 0 : ereport(ERROR,
7598 : (errcode(ERRCODE_INDEX_CORRUPTED),
7599 : errmsg_internal("heap tid from index tuple (%u,%u) points to unused heap page item at offset %u of block %u in index \"%s\"",
7600 ECB : ItemPointerGetBlockNumber(htid),
7601 : indexpagehoffnum,
7602 : istatus->idxoffnum, delstate->iblknum,
7603 : RelationGetRelationName(delstate->irel))));
7604 :
7605 CBC 621751 : if (ItemIdHasStorage(iid))
7606 ECB : {
7607 : HeapTupleHeader htup;
7608 EUB :
7609 CBC 389368 : Assert(ItemIdIsNormal(iid));
7610 GBC 389368 : htup = (HeapTupleHeader) PageGetItem(page, iid);
7611 :
7612 CBC 389368 : if (unlikely(HeapTupleHeaderIsHeapOnly(htup)))
7613 LBC 0 : ereport(ERROR,
7614 : (errcode(ERRCODE_INDEX_CORRUPTED),
7615 : errmsg_internal("heap tid from index tuple (%u,%u) points to heap-only tuple at offset %u of block %u in index \"%s\"",
7616 ECB : ItemPointerGetBlockNumber(htid),
7617 : indexpagehoffnum,
7618 EUB : istatus->idxoffnum, delstate->iblknum,
7619 : RelationGetRelationName(delstate->irel))));
7620 : }
7621 GBC 621751 : }
7622 EUB :
7623 : /*
7624 : * heapam implementation of tableam's index_delete_tuples interface.
7625 : *
7626 : * This helper function is called by index AMs during index tuple deletion.
7627 : * See tableam header comments for an explanation of the interface implemented
7628 : * here and a general theory of operation. Note that each call here is either
7629 ECB : * a simple index deletion call, or a bottom-up index deletion call.
7630 : *
7631 : * It's possible for this to generate a fair amount of I/O, since we may be
7632 : * deleting hundreds of tuples from a single index block. To amortize that
7633 : * cost to some degree, this uses prefetching and combines repeat accesses to
7634 : * the same heap block.
7635 : */
7636 : TransactionId
7637 GIC 8280 : heap_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
7638 : {
7639 : /* Initial assumption is that earlier pruning took care of conflict */
7640 GNC 8280 : TransactionId snapshotConflictHorizon = InvalidTransactionId;
7641 GIC 8280 : BlockNumber blkno = InvalidBlockNumber;
7642 8280 : Buffer buf = InvalidBuffer;
7643 8280 : Page page = NULL;
7644 8280 : OffsetNumber maxoff = InvalidOffsetNumber;
7645 : TransactionId priorXmax;
7646 ECB : #ifdef USE_PREFETCH
7647 : IndexDeletePrefetchState prefetch_state;
7648 : int prefetch_distance;
7649 : #endif
7650 : SnapshotData SnapshotNonVacuumable;
7651 CBC 8280 : int finalndeltids = 0,
7652 GIC 8280 : nblocksaccessed = 0;
7653 ECB :
7654 : /* State that's only used in bottom-up index deletion case */
7655 GBC 8280 : int nblocksfavorable = 0;
7656 8280 : int curtargetfreespace = delstate->bottomupfreespace,
7657 GIC 8280 : lastfreespace = 0,
7658 8280 : actualfreespace = 0;
7659 8280 : bool bottomup_final_block = false;
7660 :
7661 8280 : InitNonVacuumableSnapshot(SnapshotNonVacuumable, GlobalVisTestFor(rel));
7662 :
7663 : /* Sort caller's deltids array by TID for further processing */
7664 8280 : index_delete_sort(delstate);
7665 ECB :
7666 : /*
7667 : * Bottom-up case: resort deltids array in an order attuned to where the
7668 : * greatest number of promising TIDs are to be found, and determine how
7669 : * many blocks from the start of sorted array should be considered
7670 : * favorable. This will also shrink the deltids array in order to
7671 : * eliminate completely unfavorable blocks up front.
7672 : */
7673 GIC 8280 : if (delstate->bottomup)
7674 3512 : nblocksfavorable = bottomup_sort_and_shrink(delstate);
7675 :
7676 : #ifdef USE_PREFETCH
7677 : /* Initialize prefetch state. */
7678 8280 : prefetch_state.cur_hblkno = InvalidBlockNumber;
7679 8280 : prefetch_state.next_item = 0;
7680 8280 : prefetch_state.ndeltids = delstate->ndeltids;
7681 8280 : prefetch_state.deltids = delstate->deltids;
7682 :
7683 : /*
7684 : * Determine the prefetch distance that we will attempt to maintain.
7685 : *
7686 ECB : * Since the caller holds a buffer lock somewhere in rel, we'd better make
7687 : * sure that isn't a catalog relation before we call code that does
7688 : * syscache lookups, to avoid risk of deadlock.
7689 : */
7690 CBC 8280 : if (IsCatalogRelation(rel))
7691 6680 : prefetch_distance = maintenance_io_concurrency;
7692 : else
7693 ECB : prefetch_distance =
7694 CBC 1600 : get_tablespace_maintenance_io_concurrency(rel->rd_rel->reltablespace);
7695 :
7696 ECB : /* Cap initial prefetch distance for bottom-up deletion caller */
7697 CBC 8280 : if (delstate->bottomup)
7698 ECB : {
7699 GIC 3512 : Assert(nblocksfavorable >= 1);
7700 CBC 3512 : Assert(nblocksfavorable <= BOTTOMUP_MAX_NBLOCKS);
7701 GIC 3512 : prefetch_distance = Min(prefetch_distance, nblocksfavorable);
7702 ECB : }
7703 :
7704 : /* Start prefetching. */
7705 CBC 8280 : index_delete_prefetch_buffer(rel, &prefetch_state, prefetch_distance);
7706 ECB : #endif
7707 :
7708 : /* Iterate over deltids, determine which to delete, check their horizon */
7709 GIC 8280 : Assert(delstate->ndeltids > 0);
7710 630031 : for (int i = 0; i < delstate->ndeltids; i++)
7711 : {
7712 625262 : TM_IndexDelete *ideltid = &delstate->deltids[i];
7713 625262 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
7714 625262 : ItemPointer htid = &ideltid->tid;
7715 ECB : OffsetNumber offnum;
7716 :
7717 : /*
7718 : * Read buffer, and perform required extra steps each time a new block
7719 : * is encountered. Avoid refetching if it's the same block as the one
7720 : * from the last htid.
7721 : */
7722 GIC 1242244 : if (blkno == InvalidBlockNumber ||
7723 616982 : ItemPointerGetBlockNumber(htid) != blkno)
7724 : {
7725 : /*
7726 : * Consider giving up early for bottom-up index deletion caller
7727 : * first. (Only prefetch next-next block afterwards, when it
7728 : * becomes clear that we're at least going to access the next
7729 : * block in line.)
7730 : *
7731 ECB : * Sometimes the first block frees so much space for bottom-up
7732 : * caller that the deletion process can end without accessing any
7733 : * more blocks. It is usually necessary to access 2 or 3 blocks
7734 : * per bottom-up deletion operation, though.
7735 : */
7736 GIC 20646 : if (delstate->bottomup)
7737 : {
7738 ECB : /*
7739 : * We often allow caller to delete a few additional items
7740 : * whose entries we reached after the point that space target
7741 EUB : * from caller was satisfied. The cost of accessing the page
7742 : * was already paid at that point, so it made sense to finish
7743 : * it off. When that happened, we finalize everything here
7744 : * (by finishing off the whole bottom-up deletion operation
7745 : * without needlessly paying the cost of accessing any more
7746 : * blocks).
7747 : */
7748 GIC 7439 : if (bottomup_final_block)
7749 CBC 107 : break;
7750 ECB :
7751 EUB : /*
7752 : * Give up when we didn't enable our caller to free any
7753 : * additional space as a result of processing the page that we
7754 : * just finished up with. This rule is the main way in which
7755 : * we keep the cost of bottom-up deletion under control.
7756 : */
7757 GIC 7332 : if (nblocksaccessed >= 1 && actualfreespace == lastfreespace)
7758 3404 : break;
7759 CBC 3928 : lastfreespace = actualfreespace; /* for next time */
7760 :
7761 : /*
7762 : * Deletion operation (which is bottom-up) will definitely
7763 ECB : * access the next block in line. Prepare for that now.
7764 : *
7765 : * Decay target free space so that we don't hang on for too
7766 : * long with a marginal case. (Space target is only truly
7767 EUB : * helpful when it allows us to recognize that we don't need
7768 : * to access more than 1 or 2 blocks to satisfy caller due to
7769 : * agreeable workload characteristics.)
7770 : *
7771 : * We are a bit more patient when we encounter contiguous
7772 : * blocks, though: these are treated as favorable blocks. The
7773 : * decay process is only applied when the next block in line
7774 : * is not a favorable/contiguous block. This is not an
7775 ECB : * exception to the general rule; we still insist on finding
7776 : * at least one deletable item per block accessed. See
7777 : * bottomup_nblocksfavorable() for full details of the theory
7778 : * behind favorable blocks and heap block locality in general.
7779 : *
7780 : * Note: The first block in line is always treated as a
7781 : * favorable block, so the earliest possible point that the
7782 : * decay can be applied is just before we access the second
7783 : * block in line. The Assert() verifies this for us.
7784 : */
7785 GIC 3928 : Assert(nblocksaccessed > 0 || nblocksfavorable > 0);
7786 3928 : if (nblocksfavorable > 0)
7787 3565 : nblocksfavorable--;
7788 : else
7789 363 : curtargetfreespace /= 2;
7790 : }
7791 ECB :
7792 : /* release old buffer */
7793 GIC 17135 : if (BufferIsValid(buf))
7794 CBC 8855 : UnlockReleaseBuffer(buf);
7795 ECB :
7796 CBC 17135 : blkno = ItemPointerGetBlockNumber(htid);
7797 17135 : buf = ReadBuffer(rel, blkno);
7798 17135 : nblocksaccessed++;
7799 GIC 17135 : Assert(!delstate->bottomup ||
7800 : nblocksaccessed <= BOTTOMUP_MAX_NBLOCKS);
7801 :
7802 : #ifdef USE_PREFETCH
7803 :
7804 : /*
7805 ECB : * To maintain the prefetch distance, prefetch one more page for
7806 : * each page we read.
7807 : */
7808 GIC 17135 : index_delete_prefetch_buffer(rel, &prefetch_state, 1);
7809 ECB : #endif
7810 :
7811 CBC 17135 : LockBuffer(buf, BUFFER_LOCK_SHARE);
7812 ECB :
7813 CBC 17135 : page = BufferGetPage(buf);
7814 GIC 17135 : maxoff = PageGetMaxOffsetNumber(page);
7815 ECB : }
7816 :
7817 : /*
7818 : * In passing, detect index corruption involving an index page with a
7819 : * TID that points to a location in the heap that couldn't possibly be
7820 : * correct. We only do this with actual TIDs from caller's index page
7821 : * (not items reached by traversing through a HOT chain).
7822 : */
7823 GIC 621751 : index_delete_check_htid(delstate, page, maxoff, htid, istatus);
7824 :
7825 621751 : if (istatus->knowndeletable)
7826 125582 : Assert(!delstate->bottomup && !istatus->promising);
7827 ECB : else
7828 : {
7829 GIC 496169 : ItemPointerData tmp = *htid;
7830 : HeapTupleData heapTuple;
7831 :
7832 ECB : /* Are any tuples from this HOT chain non-vacuumable? */
7833 CBC 496169 : if (heap_hot_search_buffer(&tmp, rel, buf, &SnapshotNonVacuumable,
7834 ECB : &heapTuple, NULL, true))
7835 CBC 359792 : continue; /* can't delete entry */
7836 :
7837 : /* Caller will delete, since whole HOT chain is vacuumable */
7838 GIC 136377 : istatus->knowndeletable = true;
7839 :
7840 : /* Maintain index free space info for bottom-up deletion case */
7841 136377 : if (delstate->bottomup)
7842 : {
7843 8387 : Assert(istatus->freespace > 0);
7844 CBC 8387 : actualfreespace += istatus->freespace;
7845 8387 : if (actualfreespace >= curtargetfreespace)
7846 GIC 3312 : bottomup_final_block = true;
7847 : }
7848 ECB : }
7849 :
7850 : /*
7851 : * Maintain snapshotConflictHorizon value for deletion operation as a
7852 : * whole by advancing current value using heap tuple headers. This is
7853 : * loosely based on the logic for pruning a HOT chain.
7854 : */
7855 CBC 261959 : offnum = ItemPointerGetOffsetNumber(htid);
7856 GIC 261959 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
7857 : for (;;)
7858 19298 : {
7859 ECB : ItemId lp;
7860 : HeapTupleHeader htup;
7861 :
7862 : /* Sanity check (pure paranoia) */
7863 CBC 281257 : if (offnum < FirstOffsetNumber)
7864 LBC 0 : break;
7865 :
7866 ECB : /*
7867 : * An offset past the end of page's line pointer array is possible
7868 : * when the array was truncated
7869 : */
7870 GIC 281257 : if (offnum > maxoff)
7871 UIC 0 : break;
7872 :
7873 GIC 281257 : lp = PageGetItemId(page, offnum);
7874 281257 : if (ItemIdIsRedirected(lp))
7875 : {
7876 CBC 8606 : offnum = ItemIdGetRedirect(lp);
7877 8606 : continue;
7878 : }
7879 :
7880 : /*
7881 : * We'll often encounter LP_DEAD line pointers (especially with an
7882 : * entry marked knowndeletable by our caller up front). No heap
7883 : * tuple headers get examined for an htid that leads us to an
7884 : * LP_DEAD item. This is okay because the earlier pruning
7885 : * operation that made the line pointer LP_DEAD in the first place
7886 : * must have considered the original tuple header as part of
7887 : * generating its own snapshotConflictHorizon value.
7888 : *
7889 : * Relying on XLOG_HEAP2_PRUNE records like this is the same
7890 ECB : * strategy that index vacuuming uses in all cases. Index VACUUM
7891 : * WAL records don't even have a snapshotConflictHorizon field of
7892 : * their own for this reason.
7893 : */
7894 GIC 272651 : if (!ItemIdIsNormal(lp))
7895 176529 : break;
7896 :
7897 96122 : htup = (HeapTupleHeader) PageGetItem(page, lp);
7898 :
7899 : /*
7900 : * Check the tuple XMIN against prior XMAX, if any
7901 : */
7902 CBC 106814 : if (TransactionIdIsValid(priorXmax) &&
7903 10692 : !TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
7904 UIC 0 : break;
7905 :
7906 GNC 96122 : HeapTupleHeaderAdvanceConflictHorizon(htup,
7907 : &snapshotConflictHorizon);
7908 :
7909 : /*
7910 : * If the tuple is not HOT-updated, then we are at the end of this
7911 : * HOT-chain. No need to visit later tuples from the same update
7912 ECB : * chain (they get their own index entries) -- just move on to
7913 : * next htid from index AM caller.
7914 : */
7915 GIC 96122 : if (!HeapTupleHeaderIsHotUpdated(htup))
7916 : break;
7917 :
7918 : /* Advance to next HOT chain member */
7919 10692 : Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blkno);
7920 10692 : offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
7921 10692 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
7922 : }
7923 :
7924 : /* Enable further/final shrinking of deltids for caller */
7925 261959 : finalndeltids = i + 1;
7926 : }
7927 :
7928 8280 : UnlockReleaseBuffer(buf);
7929 :
7930 : /*
7931 : * Shrink deltids array to exclude non-deletable entries at the end. This
7932 : * is not just a minor optimization. Final deltids array size might be
7933 : * zero for a bottom-up caller. Index AM is explicitly allowed to rely on
7934 : * ndeltids being zero in all cases with zero total deletable entries.
7935 : */
7936 8280 : Assert(finalndeltids > 0 || delstate->bottomup);
7937 8280 : delstate->ndeltids = finalndeltids;
7938 :
7939 GNC 8280 : return snapshotConflictHorizon;
7940 ECB : }
7941 :
7942 : /*
7943 : * Specialized inlineable comparison function for index_delete_sort()
7944 : */
7945 : static inline int
7946 GIC 18594991 : index_delete_sort_cmp(TM_IndexDelete *deltid1, TM_IndexDelete *deltid2)
7947 : {
7948 CBC 18594991 : ItemPointer tid1 = &deltid1->tid;
7949 18594991 : ItemPointer tid2 = &deltid2->tid;
7950 :
7951 ECB : {
7952 CBC 18594991 : BlockNumber blk1 = ItemPointerGetBlockNumber(tid1);
7953 18594991 : BlockNumber blk2 = ItemPointerGetBlockNumber(tid2);
7954 ECB :
7955 GIC 18594991 : if (blk1 != blk2)
7956 7685436 : return (blk1 < blk2) ? -1 : 1;
7957 : }
7958 : {
7959 10909555 : OffsetNumber pos1 = ItemPointerGetOffsetNumber(tid1);
7960 10909555 : OffsetNumber pos2 = ItemPointerGetOffsetNumber(tid2);
7961 :
7962 10909555 : if (pos1 != pos2)
7963 CBC 10909555 : return (pos1 < pos2) ? -1 : 1;
7964 : }
7965 :
7966 LBC 0 : Assert(false);
7967 :
7968 ECB : return 0;
7969 : }
7970 :
7971 : /*
7972 : * Sort deltids array from delstate by TID. This prepares it for further
7973 : * processing by heap_index_delete_tuples().
7974 : *
7975 : * This operation becomes a noticeable consumer of CPU cycles with some
7976 : * workloads, so we go to the trouble of specialization/micro optimization.
7977 : * We use shellsort for this because it's easy to specialize, compiles to
7978 : * relatively few instructions, and is adaptive to presorted inputs/subsets
7979 : * (which are typical here).
7980 : */
7981 : static void
7982 GIC 8280 : index_delete_sort(TM_IndexDeleteOp *delstate)
7983 : {
7984 CBC 8280 : TM_IndexDelete *deltids = delstate->deltids;
7985 GIC 8280 : int ndeltids = delstate->ndeltids;
7986 8280 : int low = 0;
7987 :
7988 ECB : /*
7989 : * Shellsort gap sequence (taken from Sedgewick-Incerpi paper).
7990 : *
7991 : * This implementation is fast with array sizes up to ~4500. This covers
7992 : * all supported BLCKSZ values.
7993 : */
7994 GIC 8280 : const int gaps[9] = {1968, 861, 336, 112, 48, 21, 7, 3, 1};
7995 :
7996 ECB : /* Think carefully before changing anything here -- keep swaps cheap */
7997 : StaticAssertDecl(sizeof(TM_IndexDelete) <= 8,
7998 : "element size exceeds 8 bytes");
7999 :
8000 CBC 82800 : for (int g = 0; g < lengthof(gaps); g++)
8001 ECB : {
8002 GIC 10647541 : for (int hi = gaps[g], i = low + hi; i < ndeltids; i++)
8003 : {
8004 10573021 : TM_IndexDelete d = deltids[i];
8005 10573021 : int j = i;
8006 :
8007 19191674 : while (j >= hi && index_delete_sort_cmp(&deltids[j - hi], &d) >= 0)
8008 : {
8009 8618653 : deltids[j] = deltids[j - hi];
8010 CBC 8618653 : j -= hi;
8011 ECB : }
8012 GIC 10573021 : deltids[j] = d;
8013 ECB : }
8014 : }
8015 GIC 8280 : }
8016 :
8017 : /*
8018 ECB : * Returns how many blocks should be considered favorable/contiguous for a
8019 EUB : * bottom-up index deletion pass. This is a number of heap blocks that starts
8020 : * from and includes the first block in line.
8021 : *
8022 : * There is always at least one favorable block during bottom-up index
8023 : * deletion. In the worst case (i.e. with totally random heap blocks) the
8024 : * first block in line (the only favorable block) can be thought of as a
8025 ECB : * degenerate array of contiguous blocks that consists of a single block.
8026 EUB : * heap_index_delete_tuples() will expect this.
8027 : *
8028 ECB : * Caller passes blockgroups, a description of the final order that deltids
8029 : * will be sorted in for heap_index_delete_tuples() bottom-up index deletion
8030 : * processing. Note that deltids need not actually be sorted just yet (caller
8031 : * only passes deltids to us so that we can interpret blockgroups).
8032 : *
8033 : * You might guess that the existence of contiguous blocks cannot matter much,
8034 : * since in general the main factor that determines which blocks we visit is
8035 : * the number of promising TIDs, which is a fixed hint from the index AM.
8036 : * We're not really targeting the general case, though -- the actual goal is
8037 : * to adapt our behavior to a wide variety of naturally occurring conditions.
8038 : * The effects of most of the heuristics we apply are only noticeable in the
8039 : * aggregate, over time and across many _related_ bottom-up index deletion
8040 : * passes.
8041 : *
8042 : * Deeming certain blocks favorable allows heapam to recognize and adapt to
8043 : * workloads where heap blocks visited during bottom-up index deletion can be
8044 : * accessed contiguously, in the sense that each newly visited block is the
8045 : * neighbor of the block that bottom-up deletion just finished processing (or
8046 : * close enough to it). It will likely be cheaper to access more favorable
8047 : * blocks sooner rather than later (e.g. in this pass, not across a series of
8048 : * related bottom-up passes). Either way it is probably only a matter of time
8049 : * (or a matter of further correlated version churn) before all blocks that
8050 : * appear together as a single large batch of favorable blocks get accessed by
8051 : * _some_ bottom-up pass. Large batches of favorable blocks tend to either
8052 : * appear almost constantly or not even once (it all depends on per-index
8053 : * workload characteristics).
8054 : *
8055 : * Note that the blockgroups sort order applies a power-of-two bucketing
8056 : * scheme that creates opportunities for contiguous groups of blocks to get
8057 : * batched together, at least with workloads that are naturally amenable to
8058 : * being driven by heap block locality. This doesn't just enhance the spatial
8059 EUB : * locality of bottom-up heap block processing in the obvious way. It also
8060 : * enables temporal locality of access, since sorting by heap block number
8061 ECB : * naturally tends to make the bottom-up processing order deterministic.
8062 : *
8063 : * Consider the following example to get a sense of how temporal locality
8064 : * might matter: There is a heap relation with several indexes, each of which
8065 : * is low to medium cardinality. It is subject to constant non-HOT updates.
8066 : * The updates are skewed (in one part of the primary key, perhaps). None of
8067 : * the indexes are logically modified by the UPDATE statements (if they were
8068 : * then bottom-up index deletion would not be triggered in the first place).
8069 : * Naturally, each new round of index tuples (for each heap tuple that gets a
8070 : * heap_update() call) will have the same heap TID in each and every index.
8071 : * Since these indexes are low cardinality and never get logically modified,
8072 : * heapam processing during bottom-up deletion passes will access heap blocks
8073 : * in approximately sequential order. Temporal locality of access occurs due
8074 : * to bottom-up deletion passes behaving very similarly across each of the
8075 : * indexes at any given moment. This keeps the number of buffer misses needed
8076 : * to visit heap blocks to a minimum.
8077 : */
8078 : static int
8079 GIC 3512 : bottomup_nblocksfavorable(IndexDeleteCounts *blockgroups, int nblockgroups,
8080 ECB : TM_IndexDelete *deltids)
8081 : {
8082 GIC 3512 : int64 lastblock = -1;
8083 CBC 3512 : int nblocksfavorable = 0;
8084 :
8085 GIC 3512 : Assert(nblockgroups >= 1);
8086 3512 : Assert(nblockgroups <= BOTTOMUP_MAX_NBLOCKS);
8087 :
8088 : /*
8089 : * We tolerate heap blocks that will be accessed only slightly out of
8090 : * physical order. Small blips occur when a pair of almost-contiguous
8091 ECB : * blocks happen to fall into different buckets (perhaps due only to a
8092 : * small difference in npromisingtids that the bucketing scheme didn't
8093 : * quite manage to ignore). We effectively ignore these blips by applying
8094 : * a small tolerance. The precise tolerance we use is a little arbitrary,
8095 : * but it works well enough in practice.
8096 : */
8097 GIC 12318 : for (int b = 0; b < nblockgroups; b++)
8098 : {
8099 11563 : IndexDeleteCounts *group = blockgroups + b;
8100 11563 : TM_IndexDelete *firstdtid = deltids + group->ifirsttid;
8101 CBC 11563 : BlockNumber block = ItemPointerGetBlockNumber(&firstdtid->tid);
8102 :
8103 11563 : if (lastblock != -1 &&
8104 8051 : ((int64) block < lastblock - BOTTOMUP_TOLERANCE_NBLOCKS ||
8105 GIC 7393 : (int64) block > lastblock + BOTTOMUP_TOLERANCE_NBLOCKS))
8106 : break;
8107 ECB :
8108 CBC 8806 : nblocksfavorable++;
8109 GIC 8806 : lastblock = block;
8110 ECB : }
8111 :
8112 : /* Always indicate that there is at least 1 favorable block */
8113 GIC 3512 : Assert(nblocksfavorable >= 1);
8114 ECB :
8115 CBC 3512 : return nblocksfavorable;
8116 : }
8117 ECB :
8118 : /*
8119 : * qsort comparison function for bottomup_sort_and_shrink()
8120 : */
8121 EUB : static int
8122 GIC 247741 : bottomup_sort_and_shrink_cmp(const void *arg1, const void *arg2)
8123 : {
8124 247741 : const IndexDeleteCounts *group1 = (const IndexDeleteCounts *) arg1;
8125 247741 : const IndexDeleteCounts *group2 = (const IndexDeleteCounts *) arg2;
8126 :
8127 : /*
8128 : * Most significant field is npromisingtids (which we invert the order of
8129 : * so as to sort in desc order).
8130 : *
8131 : * Caller should have already normalized npromisingtids fields into
8132 : * power-of-two values (buckets).
8133 : */
8134 247741 : if (group1->npromisingtids > group2->npromisingtids)
8135 15597 : return -1;
8136 232144 : if (group1->npromisingtids < group2->npromisingtids)
8137 CBC 21578 : return 1;
8138 :
8139 ECB : /*
8140 : * Tiebreak: desc ntids sort order.
8141 : *
8142 : * We cannot expect power-of-two values for ntids fields. We should
8143 : * behave as if they were already rounded up for us instead.
8144 : */
8145 GIC 210566 : if (group1->ntids != group2->ntids)
8146 : {
8147 137005 : uint32 ntids1 = pg_nextpower2_32((uint32) group1->ntids);
8148 137005 : uint32 ntids2 = pg_nextpower2_32((uint32) group2->ntids);
8149 ECB :
8150 GIC 137005 : if (ntids1 > ntids2)
8151 22329 : return -1;
8152 114676 : if (ntids1 < ntids2)
8153 28790 : return 1;
8154 : }
8155 ECB :
8156 : /*
8157 : * Tiebreak: asc offset-into-deltids-for-block (offset to first TID for
8158 : * block in deltids array) order.
8159 : *
8160 : * This is equivalent to sorting in ascending heap block number order
8161 : * (among otherwise equal subsets of the array). This approach allows us
8162 : * to avoid accessing the out-of-line TID. (We rely on the assumption
8163 : * that the deltids array was sorted in ascending heap TID order when
8164 : * these offsets to the first TID from each heap block group were formed.)
8165 : */
8166 GIC 159447 : if (group1->ifirsttid > group2->ifirsttid)
8167 CBC 78579 : return 1;
8168 GIC 80868 : if (group1->ifirsttid < group2->ifirsttid)
8169 80868 : return -1;
8170 ECB :
8171 UIC 0 : pg_unreachable();
8172 :
8173 : return 0;
8174 : }
8175 :
8176 : /*
8177 : * heap_index_delete_tuples() helper function for bottom-up deletion callers.
8178 : *
8179 : * Sorts deltids array in the order needed for useful processing by bottom-up
8180 : * deletion. The array should already be sorted in TID order when we're
8181 : * called. The sort process groups heap TIDs from deltids into heap block
8182 : * groupings. Earlier/more-promising groups/blocks are usually those that are
8183 : * known to have the most "promising" TIDs.
8184 : *
8185 : * Sets new size of deltids array (ndeltids) in state. deltids will only have
8186 : * TIDs from the BOTTOMUP_MAX_NBLOCKS most promising heap blocks when we
8187 : * return. This often means that deltids will be shrunk to a small fraction
8188 : * of its original size (we eliminate many heap blocks from consideration for
8189 : * caller up front).
8190 : *
8191 : * Returns the number of "favorable" blocks. See bottomup_nblocksfavorable()
8192 : * for a definition and full details.
8193 : */
8194 : static int
8195 GIC 3512 : bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate)
8196 : {
8197 : IndexDeleteCounts *blockgroups;
8198 : TM_IndexDelete *reordereddeltids;
8199 3512 : BlockNumber curblock = InvalidBlockNumber;
8200 3512 : int nblockgroups = 0;
8201 3512 : int ncopied = 0;
8202 3512 : int nblocksfavorable = 0;
8203 :
8204 3512 : Assert(delstate->bottomup);
8205 3512 : Assert(delstate->ndeltids > 0);
8206 :
8207 : /* Calculate per-heap-block count of TIDs */
8208 3512 : blockgroups = palloc(sizeof(IndexDeleteCounts) * delstate->ndeltids);
8209 1484596 : for (int i = 0; i < delstate->ndeltids; i++)
8210 : {
8211 1481084 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8212 1481084 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8213 1481084 : ItemPointer htid = &ideltid->tid;
8214 1481084 : bool promising = istatus->promising;
8215 :
8216 1481084 : if (curblock != ItemPointerGetBlockNumber(htid))
8217 : {
8218 : /* New block group */
8219 57032 : nblockgroups++;
8220 :
8221 57032 : Assert(curblock < ItemPointerGetBlockNumber(htid) ||
8222 : !BlockNumberIsValid(curblock));
8223 :
8224 57032 : curblock = ItemPointerGetBlockNumber(htid);
8225 57032 : blockgroups[nblockgroups - 1].ifirsttid = i;
8226 57032 : blockgroups[nblockgroups - 1].ntids = 1;
8227 57032 : blockgroups[nblockgroups - 1].npromisingtids = 0;
8228 : }
8229 : else
8230 : {
8231 1424052 : blockgroups[nblockgroups - 1].ntids++;
8232 : }
8233 :
8234 CBC 1481084 : if (promising)
8235 GIC 246652 : blockgroups[nblockgroups - 1].npromisingtids++;
8236 : }
8237 ECB :
8238 : /*
8239 : * We're about ready to sort block groups to determine the optimal order
8240 : * for visiting heap blocks. But before we do, round the number of
8241 : * promising tuples for each block group up to the next power-of-two,
8242 : * unless it is very low (less than 4), in which case we round up to 4.
8243 : * npromisingtids is far too noisy to trust when choosing between a pair
8244 : * of block groups that both have very low values.
8245 : *
8246 : * This scheme divides heap blocks/block groups into buckets. Each bucket
8247 : * contains blocks that have _approximately_ the same number of promising
8248 : * TIDs as each other. The goal is to ignore relatively small differences
8249 : * in the total number of promising entries, so that the whole process can
8250 : * give a little weight to heapam factors (like heap block locality)
8251 : * instead. This isn't a trade-off, really -- we have nothing to lose. It
8252 : * would be foolish to interpret small differences in npromisingtids
8253 : * values as anything more than noise.
8254 : *
8255 : * We tiebreak on nhtids when sorting block group subsets that have the
8256 : * same npromisingtids, but this has the same issues as npromisingtids,
8257 : * and so nhtids is subject to the same power-of-two bucketing scheme. The
8258 : * only reason that we don't fix nhtids in the same way here too is that
8259 : * we'll need accurate nhtids values after the sort. We handle nhtids
8260 : * bucketization dynamically instead (in the sort comparator).
8261 : *
8262 : * See bottomup_nblocksfavorable() for a full explanation of when and how
8263 : * heap locality/favorable blocks can significantly influence when and how
8264 : * heap blocks are accessed.
8265 : */
8266 GIC 60544 : for (int b = 0; b < nblockgroups; b++)
8267 : {
8268 CBC 57032 : IndexDeleteCounts *group = blockgroups + b;
8269 :
8270 ECB : /* Better off falling back on nhtids with low npromisingtids */
8271 GIC 57032 : if (group->npromisingtids <= 4)
8272 45762 : group->npromisingtids = 4;
8273 : else
8274 11270 : group->npromisingtids =
8275 11270 : pg_nextpower2_32((uint32) group->npromisingtids);
8276 : }
8277 ECB :
8278 : /* Sort groups and rearrange caller's deltids array */
8279 CBC 3512 : qsort(blockgroups, nblockgroups, sizeof(IndexDeleteCounts),
8280 ECB : bottomup_sort_and_shrink_cmp);
8281 GIC 3512 : reordereddeltids = palloc(delstate->ndeltids * sizeof(TM_IndexDelete));
8282 :
8283 3512 : nblockgroups = Min(BOTTOMUP_MAX_NBLOCKS, nblockgroups);
8284 : /* Determine number of favorable blocks at the start of final deltids */
8285 3512 : nblocksfavorable = bottomup_nblocksfavorable(blockgroups, nblockgroups,
8286 : delstate->deltids);
8287 :
8288 24010 : for (int b = 0; b < nblockgroups; b++)
8289 ECB : {
8290 CBC 20498 : IndexDeleteCounts *group = blockgroups + b;
8291 20498 : TM_IndexDelete *firstdtid = delstate->deltids + group->ifirsttid;
8292 ECB :
8293 GIC 20498 : memcpy(reordereddeltids + ncopied, firstdtid,
8294 20498 : sizeof(TM_IndexDelete) * group->ntids);
8295 20498 : ncopied += group->ntids;
8296 : }
8297 :
8298 : /* Copy final grouped and sorted TIDs back into start of caller's array */
8299 3512 : memcpy(delstate->deltids, reordereddeltids,
8300 ECB : sizeof(TM_IndexDelete) * ncopied);
8301 GIC 3512 : delstate->ndeltids = ncopied;
8302 ECB :
8303 CBC 3512 : pfree(reordereddeltids);
8304 GIC 3512 : pfree(blockgroups);
8305 ECB :
8306 CBC 3512 : return nblocksfavorable;
8307 ECB : }
8308 :
8309 : /*
8310 : * Perform XLogInsert for a heap-visible operation. 'block' is the block
8311 : * being marked all-visible, and vm_buffer is the buffer containing the
8312 : * corresponding visibility map block. Both should have already been modified
8313 : * and dirtied.
8314 : *
8315 : * snapshotConflictHorizon comes from the largest xmin on the page being
8316 : * marked all-visible. REDO routine uses it to generate recovery conflicts.
8317 : *
8318 : * If checksums or wal_log_hints are enabled, we may also generate a full-page
8319 : * image of heap_buffer. Otherwise, we optimize away the FPI (by specifying
8320 : * REGBUF_NO_IMAGE for the heap buffer), in which case the caller should *not*
8321 : * update the heap page's LSN.
8322 : */
8323 : XLogRecPtr
8324 GNC 154036 : log_heap_visible(Relation rel, Buffer heap_buffer, Buffer vm_buffer,
8325 : TransactionId snapshotConflictHorizon, uint8 vmflags)
8326 ECB : {
8327 : xl_heap_visible xlrec;
8328 : XLogRecPtr recptr;
8329 : uint8 flags;
8330 :
8331 GIC 154036 : Assert(BufferIsValid(heap_buffer));
8332 CBC 154036 : Assert(BufferIsValid(vm_buffer));
8333 ECB :
8334 GNC 154036 : xlrec.snapshotConflictHorizon = snapshotConflictHorizon;
8335 CBC 154036 : xlrec.flags = vmflags;
8336 GNC 154036 : if (RelationIsAccessibleInLogicalDecoding(rel))
8337 25 : xlrec.flags |= VISIBILITYMAP_XLOG_CATALOG_REL;
8338 CBC 154036 : XLogBeginInsert();
8339 154036 : XLogRegisterData((char *) &xlrec, SizeOfHeapVisible);
8340 ECB :
8341 GIC 154036 : XLogRegisterBuffer(0, vm_buffer, 0);
8342 ECB :
8343 GIC 154036 : flags = REGBUF_STANDARD;
8344 154036 : if (!XLogHintBitIsNeeded())
8345 CBC 145698 : flags |= REGBUF_NO_IMAGE;
8346 GIC 154036 : XLogRegisterBuffer(1, heap_buffer, flags);
8347 ECB :
8348 GIC 154036 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_VISIBLE);
8349 :
8350 CBC 154036 : return recptr;
8351 ECB : }
8352 :
8353 : /*
8354 : * Perform XLogInsert for a heap-update operation. Caller must already
8355 : * have modified the buffer(s) and marked them dirty.
8356 : */
8357 : static XLogRecPtr
8358 GIC 407557 : log_heap_update(Relation reln, Buffer oldbuf,
8359 : Buffer newbuf, HeapTuple oldtup, HeapTuple newtup,
8360 ECB : HeapTuple old_key_tuple,
8361 : bool all_visible_cleared, bool new_all_visible_cleared)
8362 : {
8363 : xl_heap_update xlrec;
8364 : xl_heap_header xlhdr;
8365 : xl_heap_header xlhdr_idx;
8366 : uint8 info;
8367 : uint16 prefix_suffix[2];
8368 GIC 407557 : uint16 prefixlen = 0,
8369 407557 : suffixlen = 0;
8370 : XLogRecPtr recptr;
8371 407557 : Page page = BufferGetPage(newbuf);
8372 407557 : bool need_tuple_data = RelationIsLogicallyLogged(reln);
8373 : bool init;
8374 : int bufflags;
8375 :
8376 : /* Caller should not call me on a non-WAL-logged relation */
8377 407557 : Assert(RelationNeedsWAL(reln));
8378 :
8379 407557 : XLogBeginInsert();
8380 :
8381 407557 : if (HeapTupleIsHeapOnly(newtup))
8382 212632 : info = XLOG_HEAP_HOT_UPDATE;
8383 : else
8384 194925 : info = XLOG_HEAP_UPDATE;
8385 :
8386 : /*
8387 : * If the old and new tuple are on the same page, we only need to log the
8388 : * parts of the new tuple that were changed. That saves on the amount of
8389 : * WAL we need to write. Currently, we just count any unchanged bytes in
8390 : * the beginning and end of the tuple. That's quick to check, and
8391 : * perfectly covers the common case that only one field is updated.
8392 ECB : *
8393 : * We could do this even if the old and new tuple are on different pages,
8394 : * but only if we don't make a full-page image of the old page, which is
8395 : * difficult to know in advance. Also, if the old tuple is corrupt for
8396 : * some reason, it would allow the corruption to propagate the new page,
8397 : * so it seems best to avoid. Under the general assumption that most
8398 : * updates tend to create the new tuple version on the same page, there
8399 : * isn't much to be gained by doing this across pages anyway.
8400 : *
8401 : * Skip this if we're taking a full-page image of the new page, as we
8402 : * don't include the new tuple in the WAL record in that case. Also
8403 : * disable if wal_level='logical', as logical decoding needs to be able to
8404 : * read the new tuple in whole from the WAL record alone.
8405 : */
8406 GIC 407557 : if (oldbuf == newbuf && !need_tuple_data &&
8407 CBC 187993 : !XLogCheckBufferNeedsBackup(newbuf))
8408 : {
8409 187185 : char *oldp = (char *) oldtup->t_data + oldtup->t_data->t_hoff;
8410 GIC 187185 : char *newp = (char *) newtup->t_data + newtup->t_data->t_hoff;
8411 CBC 187185 : int oldlen = oldtup->t_len - oldtup->t_data->t_hoff;
8412 GIC 187185 : int newlen = newtup->t_len - newtup->t_data->t_hoff;
8413 :
8414 ECB : /* Check for common prefix between old and new tuple */
8415 GIC 18651419 : for (prefixlen = 0; prefixlen < Min(oldlen, newlen); prefixlen++)
8416 ECB : {
8417 CBC 18603122 : if (newp[prefixlen] != oldp[prefixlen])
8418 GIC 138888 : break;
8419 ECB : }
8420 :
8421 : /*
8422 : * Storing the length of the prefix takes 2 bytes, so we need to save
8423 : * at least 3 bytes or there's no point.
8424 : */
8425 CBC 187185 : if (prefixlen < 3)
8426 GIC 21858 : prefixlen = 0;
8427 ECB :
8428 : /* Same for suffix */
8429 CBC 6371228 : for (suffixlen = 0; suffixlen < Min(oldlen, newlen) - prefixlen; suffixlen++)
8430 ECB : {
8431 GIC 6322739 : if (newp[newlen - suffixlen - 1] != oldp[oldlen - suffixlen - 1])
8432 CBC 138696 : break;
8433 : }
8434 GIC 187185 : if (suffixlen < 3)
8435 56199 : suffixlen = 0;
8436 : }
8437 :
8438 : /* Prepare main WAL data chain */
8439 407557 : xlrec.flags = 0;
8440 407557 : if (all_visible_cleared)
8441 960 : xlrec.flags |= XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED;
8442 407557 : if (new_all_visible_cleared)
8443 362 : xlrec.flags |= XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED;
8444 407557 : if (prefixlen > 0)
8445 165327 : xlrec.flags |= XLH_UPDATE_PREFIX_FROM_OLD;
8446 407557 : if (suffixlen > 0)
8447 130986 : xlrec.flags |= XLH_UPDATE_SUFFIX_FROM_OLD;
8448 407557 : if (need_tuple_data)
8449 : {
8450 CBC 84736 : xlrec.flags |= XLH_UPDATE_CONTAINS_NEW_TUPLE;
8451 GIC 84736 : if (old_key_tuple)
8452 : {
8453 214 : if (reln->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
8454 76 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_TUPLE;
8455 : else
8456 138 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_KEY;
8457 ECB : }
8458 : }
8459 :
8460 : /* If new tuple is the single and first tuple on page... */
8461 CBC 414488 : if (ItemPointerGetOffsetNumber(&(newtup->t_self)) == FirstOffsetNumber &&
8462 6931 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
8463 ECB : {
8464 CBC 6864 : info |= XLOG_HEAP_INIT_PAGE;
8465 6864 : init = true;
8466 : }
8467 ECB : else
8468 GIC 400693 : init = false;
8469 ECB :
8470 : /* Prepare WAL data for the old page */
8471 CBC 407557 : xlrec.old_offnum = ItemPointerGetOffsetNumber(&oldtup->t_self);
8472 407557 : xlrec.old_xmax = HeapTupleHeaderGetRawXmax(oldtup->t_data);
8473 GIC 815114 : xlrec.old_infobits_set = compute_infobits(oldtup->t_data->t_infomask,
8474 CBC 407557 : oldtup->t_data->t_infomask2);
8475 :
8476 ECB : /* Prepare WAL data for the new page */
8477 GIC 407557 : xlrec.new_offnum = ItemPointerGetOffsetNumber(&newtup->t_self);
8478 407557 : xlrec.new_xmax = HeapTupleHeaderGetRawXmax(newtup->t_data);
8479 :
8480 407557 : bufflags = REGBUF_STANDARD;
8481 407557 : if (init)
8482 6864 : bufflags |= REGBUF_WILL_INIT;
8483 407557 : if (need_tuple_data)
8484 CBC 84736 : bufflags |= REGBUF_KEEP_DATA;
8485 :
8486 GIC 407557 : XLogRegisterBuffer(0, newbuf, bufflags);
8487 407557 : if (oldbuf != newbuf)
8488 186055 : XLogRegisterBuffer(1, oldbuf, REGBUF_STANDARD);
8489 :
8490 407557 : XLogRegisterData((char *) &xlrec, SizeOfHeapUpdate);
8491 :
8492 : /*
8493 : * Prepare WAL data for the new tuple.
8494 ECB : */
8495 CBC 407557 : if (prefixlen > 0 || suffixlen > 0)
8496 : {
8497 186819 : if (prefixlen > 0 && suffixlen > 0)
8498 ECB : {
8499 GIC 109494 : prefix_suffix[0] = prefixlen;
8500 109494 : prefix_suffix[1] = suffixlen;
8501 109494 : XLogRegisterBufData(0, (char *) &prefix_suffix, sizeof(uint16) * 2);
8502 : }
8503 CBC 77325 : else if (prefixlen > 0)
8504 : {
8505 55833 : XLogRegisterBufData(0, (char *) &prefixlen, sizeof(uint16));
8506 : }
8507 ECB : else
8508 : {
8509 GIC 21492 : XLogRegisterBufData(0, (char *) &suffixlen, sizeof(uint16));
8510 ECB : }
8511 : }
8512 :
8513 GIC 407557 : xlhdr.t_infomask2 = newtup->t_data->t_infomask2;
8514 407557 : xlhdr.t_infomask = newtup->t_data->t_infomask;
8515 407557 : xlhdr.t_hoff = newtup->t_data->t_hoff;
8516 407557 : Assert(SizeofHeapTupleHeader + prefixlen + suffixlen <= newtup->t_len);
8517 :
8518 : /*
8519 : * PG73FORMAT: write bitmap [+ padding] [+ oid] + data
8520 : *
8521 : * The 'data' doesn't include the common prefix or suffix.
8522 : */
8523 407557 : XLogRegisterBufData(0, (char *) &xlhdr, SizeOfHeapHeader);
8524 407557 : if (prefixlen == 0)
8525 : {
8526 242230 : XLogRegisterBufData(0,
8527 242230 : ((char *) newtup->t_data) + SizeofHeapTupleHeader,
8528 242230 : newtup->t_len - SizeofHeapTupleHeader - suffixlen);
8529 : }
8530 : else
8531 : {
8532 ECB : /*
8533 : * Have to write the null bitmap and data after the common prefix as
8534 : * two separate rdata entries.
8535 : */
8536 : /* bitmap [+ padding] [+ oid] */
8537 CBC 165327 : if (newtup->t_data->t_hoff - SizeofHeapTupleHeader > 0)
8538 ECB : {
8539 GIC 165327 : XLogRegisterBufData(0,
8540 165327 : ((char *) newtup->t_data) + SizeofHeapTupleHeader,
8541 CBC 165327 : newtup->t_data->t_hoff - SizeofHeapTupleHeader);
8542 : }
8543 ECB :
8544 : /* data after common prefix */
8545 GIC 165327 : XLogRegisterBufData(0,
8546 165327 : ((char *) newtup->t_data) + newtup->t_data->t_hoff + prefixlen,
8547 165327 : newtup->t_len - newtup->t_data->t_hoff - prefixlen - suffixlen);
8548 : }
8549 :
8550 : /* We need to log a tuple identity */
8551 CBC 407557 : if (need_tuple_data && old_key_tuple)
8552 ECB : {
8553 : /* don't really need this, but its more comfy to decode */
8554 GIC 214 : xlhdr_idx.t_infomask2 = old_key_tuple->t_data->t_infomask2;
8555 CBC 214 : xlhdr_idx.t_infomask = old_key_tuple->t_data->t_infomask;
8556 GIC 214 : xlhdr_idx.t_hoff = old_key_tuple->t_data->t_hoff;
8557 ECB :
8558 CBC 214 : XLogRegisterData((char *) &xlhdr_idx, SizeOfHeapHeader);
8559 :
8560 ECB : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
8561 CBC 214 : XLogRegisterData((char *) old_key_tuple->t_data + SizeofHeapTupleHeader,
8562 GIC 214 : old_key_tuple->t_len - SizeofHeapTupleHeader);
8563 : }
8564 :
8565 ECB : /* filtering by origin on a row level is much more efficient */
8566 CBC 407557 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
8567 ECB :
8568 CBC 407557 : recptr = XLogInsert(RM_HEAP_ID, info);
8569 ECB :
8570 CBC 407557 : return recptr;
8571 ECB : }
8572 :
8573 : /*
8574 : * Perform XLogInsert of an XLOG_HEAP2_NEW_CID record
8575 : *
8576 : * This is only used in wal_level >= WAL_LEVEL_LOGICAL, and only for catalog
8577 : * tuples.
8578 : */
8579 : static XLogRecPtr
8580 CBC 26222 : log_heap_new_cid(Relation relation, HeapTuple tup)
8581 : {
8582 ECB : xl_heap_new_cid xlrec;
8583 :
8584 : XLogRecPtr recptr;
8585 GIC 26222 : HeapTupleHeader hdr = tup->t_data;
8586 :
8587 CBC 26222 : Assert(ItemPointerIsValid(&tup->t_self));
8588 26222 : Assert(tup->t_tableOid != InvalidOid);
8589 :
8590 26222 : xlrec.top_xid = GetTopTransactionId();
8591 GNC 26222 : xlrec.target_locator = relation->rd_locator;
8592 GIC 26222 : xlrec.target_tid = tup->t_self;
8593 :
8594 ECB : /*
8595 : * If the tuple got inserted & deleted in the same TX we definitely have a
8596 : * combo CID, set cmin and cmax.
8597 : */
8598 CBC 26222 : if (hdr->t_infomask & HEAP_COMBOCID)
8599 ECB : {
8600 CBC 2091 : Assert(!(hdr->t_infomask & HEAP_XMAX_INVALID));
8601 GIC 2091 : Assert(!HeapTupleHeaderXminInvalid(hdr));
8602 2091 : xlrec.cmin = HeapTupleHeaderGetCmin(hdr);
8603 CBC 2091 : xlrec.cmax = HeapTupleHeaderGetCmax(hdr);
8604 2091 : xlrec.combocid = HeapTupleHeaderGetRawCommandId(hdr);
8605 : }
8606 ECB : /* No combo CID, so only cmin or cmax can be set by this TX */
8607 : else
8608 : {
8609 : /*
8610 : * Tuple inserted.
8611 : *
8612 : * We need to check for LOCK ONLY because multixacts might be
8613 : * transferred to the new tuple in case of FOR KEY SHARE updates in
8614 : * which case there will be an xmax, although the tuple just got
8615 : * inserted.
8616 : */
8617 GIC 24131 : if (hdr->t_infomask & HEAP_XMAX_INVALID ||
8618 5788 : HEAP_XMAX_IS_LOCKED_ONLY(hdr->t_infomask))
8619 : {
8620 18344 : xlrec.cmin = HeapTupleHeaderGetRawCommandId(hdr);
8621 CBC 18344 : xlrec.cmax = InvalidCommandId;
8622 : }
8623 ECB : /* Tuple from a different tx updated or deleted. */
8624 : else
8625 : {
8626 CBC 5787 : xlrec.cmin = InvalidCommandId;
8627 5787 : xlrec.cmax = HeapTupleHeaderGetRawCommandId(hdr);
8628 : }
8629 24131 : xlrec.combocid = InvalidCommandId;
8630 : }
8631 ECB :
8632 : /*
8633 : * Note that we don't need to register the buffer here, because this
8634 : * operation does not modify the page. The insert/update/delete that
8635 : * called us certainly did, but that's WAL-logged separately.
8636 : */
8637 GIC 26222 : XLogBeginInsert();
8638 26222 : XLogRegisterData((char *) &xlrec, SizeOfHeapNewCid);
8639 ECB :
8640 : /* will be looked at irrespective of origin */
8641 :
8642 CBC 26222 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_NEW_CID);
8643 :
8644 GIC 26222 : return recptr;
8645 : }
8646 :
8647 : /*
8648 : * Build a heap tuple representing the configured REPLICA IDENTITY to represent
8649 ECB : * the old tuple in an UPDATE or DELETE.
8650 : *
8651 : * Returns NULL if there's no need to log an identity or if there's no suitable
8652 : * key defined.
8653 : *
8654 : * Pass key_required true if any replica identity columns changed value, or if
8655 : * any of them have any external data. Delete must always pass true.
8656 : *
8657 : * *copy is set to true if the returned tuple is a modified copy rather than
8658 : * the same tuple that was passed in.
8659 : */
8660 : static HeapTuple
8661 GIC 1838966 : ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
8662 : bool *copy)
8663 ECB : {
8664 GIC 1838966 : TupleDesc desc = RelationGetDescr(relation);
8665 CBC 1838966 : char replident = relation->rd_rel->relreplident;
8666 ECB : Bitmapset *idattrs;
8667 : HeapTuple key_tuple;
8668 : bool nulls[MaxHeapAttributeNumber];
8669 : Datum values[MaxHeapAttributeNumber];
8670 :
8671 CBC 1838966 : *copy = false;
8672 ECB :
8673 CBC 1838966 : if (!RelationIsLogicallyLogged(relation))
8674 GIC 1665648 : return NULL;
8675 :
8676 173318 : if (replident == REPLICA_IDENTITY_NOTHING)
8677 CBC 234 : return NULL;
8678 :
8679 GIC 173084 : if (replident == REPLICA_IDENTITY_FULL)
8680 ECB : {
8681 : /*
8682 : * When logging the entire old tuple, it very well could contain
8683 : * toasted columns. If so, force them to be inlined.
8684 : */
8685 GIC 297 : if (HeapTupleHasExternal(tp))
8686 : {
8687 CBC 4 : *copy = true;
8688 4 : tp = toast_flatten_tuple(tp, desc);
8689 : }
8690 GIC 297 : return tp;
8691 : }
8692 ECB :
8693 : /* if the key isn't required and we're only logging the key, we're done */
8694 CBC 172787 : if (!key_required)
8695 GIC 84522 : return NULL;
8696 ECB :
8697 : /* find out the replica identity columns */
8698 GIC 88265 : idattrs = RelationGetIndexAttrBitmap(relation,
8699 : INDEX_ATTR_BITMAP_IDENTITY_KEY);
8700 :
8701 : /*
8702 : * If there's no defined replica identity columns, treat as !key_required.
8703 : * (This case should not be reachable from heap_update, since that should
8704 : * calculate key_required accurately. But heap_delete just passes
8705 : * constant true for key_required, so we can hit this case in deletes.)
8706 ECB : */
8707 GIC 88265 : if (bms_is_empty(idattrs))
8708 6028 : return NULL;
8709 :
8710 : /*
8711 ECB : * Construct a new tuple containing only the replica identity columns,
8712 : * with nulls elsewhere. While we're at it, assert that the replica
8713 : * identity columns aren't null.
8714 : */
8715 GIC 82237 : heap_deform_tuple(tp, desc, values, nulls);
8716 ECB :
8717 CBC 326811 : for (int i = 0; i < desc->natts; i++)
8718 ECB : {
8719 GIC 244574 : if (bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
8720 : idattrs))
8721 82246 : Assert(!nulls[i]);
8722 : else
8723 162328 : nulls[i] = true;
8724 ECB : }
8725 :
8726 CBC 82237 : key_tuple = heap_form_tuple(desc, values, nulls);
8727 82237 : *copy = true;
8728 ECB :
8729 CBC 82237 : bms_free(idattrs);
8730 ECB :
8731 : /*
8732 : * If the tuple, which by here only contains indexed columns, still has
8733 : * toasted columns, force them to be inlined. This is somewhat unlikely
8734 : * since there's limits on the size of indexed columns, so we don't
8735 : * duplicate toast_flatten_tuple()s functionality in the above loop over
8736 : * the indexed columns, even if it would be more efficient.
8737 : */
8738 GIC 82237 : if (HeapTupleHasExternal(key_tuple))
8739 : {
8740 4 : HeapTuple oldtup = key_tuple;
8741 :
8742 4 : key_tuple = toast_flatten_tuple(oldtup, desc);
8743 CBC 4 : heap_freetuple(oldtup);
8744 ECB : }
8745 :
8746 CBC 82237 : return key_tuple;
8747 ECB : }
8748 :
8749 : /*
8750 : * Handles XLOG_HEAP2_PRUNE record type.
8751 : *
8752 : * Acquires a full cleanup lock.
8753 : */
8754 : static void
8755 CBC 6391 : heap_xlog_prune(XLogReaderState *record)
8756 : {
8757 GIC 6391 : XLogRecPtr lsn = record->EndRecPtr;
8758 6391 : xl_heap_prune *xlrec = (xl_heap_prune *) XLogRecGetData(record);
8759 : Buffer buffer;
8760 : RelFileLocator rlocator;
8761 : BlockNumber blkno;
8762 : XLogRedoAction action;
8763 ECB :
8764 GNC 6391 : XLogRecGetBlockTag(record, 0, &rlocator, NULL, &blkno);
8765 :
8766 : /*
8767 : * We're about to remove tuples. In Hot Standby mode, ensure that there's
8768 ECB : * no queries running for which the removed tuples are still visible.
8769 : */
8770 CBC 6391 : if (InHotStandby)
8771 GNC 6059 : ResolveRecoveryConflictWithSnapshot(xlrec->snapshotConflictHorizon,
8772 6059 : xlrec->isCatalogRel,
8773 : rlocator);
8774 :
8775 : /*
8776 : * If we have a full-page image, restore it (using a cleanup lock) and
8777 : * we're done.
8778 : */
8779 GIC 6391 : action = XLogReadBufferForRedoExtended(record, 0, RBM_NORMAL, true,
8780 : &buffer);
8781 6391 : if (action == BLK_NEEDS_REDO)
8782 : {
8783 5725 : Page page = (Page) BufferGetPage(buffer);
8784 : OffsetNumber *end;
8785 : OffsetNumber *redirected;
8786 : OffsetNumber *nowdead;
8787 : OffsetNumber *nowunused;
8788 : int nredirected;
8789 ECB : int ndead;
8790 : int nunused;
8791 : Size datalen;
8792 :
8793 CBC 5725 : redirected = (OffsetNumber *) XLogRecGetBlockData(record, 0, &datalen);
8794 :
8795 GIC 5725 : nredirected = xlrec->nredirected;
8796 5725 : ndead = xlrec->ndead;
8797 5725 : end = (OffsetNumber *) ((char *) redirected + datalen);
8798 5725 : nowdead = redirected + (nredirected * 2);
8799 CBC 5725 : nowunused = nowdead + ndead;
8800 GIC 5725 : nunused = (end - nowunused);
8801 CBC 5725 : Assert(nunused >= 0);
8802 ECB :
8803 : /* Update all line pointers per the record, and repair fragmentation */
8804 CBC 5725 : heap_page_prune_execute(buffer,
8805 ECB : redirected, nredirected,
8806 : nowdead, ndead,
8807 : nowunused, nunused);
8808 :
8809 : /*
8810 : * Note: we don't worry about updating the page's prunability hints.
8811 : * At worst this will cause an extra prune cycle to occur soon.
8812 : */
8813 :
8814 GIC 5725 : PageSetLSN(page, lsn);
8815 CBC 5725 : MarkBufferDirty(buffer);
8816 ECB : }
8817 :
8818 CBC 6391 : if (BufferIsValid(buffer))
8819 : {
8820 GIC 6391 : Size freespace = PageGetHeapFreeSpace(BufferGetPage(buffer));
8821 :
8822 CBC 6391 : UnlockReleaseBuffer(buffer);
8823 ECB :
8824 : /*
8825 : * After pruning records from a page, it's useful to update the FSM
8826 : * about it, as it may cause the page become target for insertions
8827 : * later even if vacuum decides not to visit it (which is possible if
8828 : * gets marked all-visible.)
8829 : *
8830 : * Do this regardless of a full-page image being applied, since the
8831 : * FSM data is not in the page anyway.
8832 : */
8833 GNC 6391 : XLogRecordPageWithFreeSpace(rlocator, blkno, freespace);
8834 : }
8835 CBC 6391 : }
8836 ECB :
8837 : /*
8838 : * Handles XLOG_HEAP2_VACUUM record type.
8839 : *
8840 : * Acquires an ordinary exclusive lock only.
8841 : */
8842 : static void
8843 CBC 1348 : heap_xlog_vacuum(XLogReaderState *record)
8844 : {
8845 1348 : XLogRecPtr lsn = record->EndRecPtr;
8846 GIC 1348 : xl_heap_vacuum *xlrec = (xl_heap_vacuum *) XLogRecGetData(record);
8847 ECB : Buffer buffer;
8848 : BlockNumber blkno;
8849 : XLogRedoAction action;
8850 :
8851 : /*
8852 : * If we have a full-page image, restore it (without using a cleanup lock)
8853 : * and we're done.
8854 : */
8855 CBC 1348 : action = XLogReadBufferForRedoExtended(record, 0, RBM_NORMAL, false,
8856 : &buffer);
8857 1348 : if (action == BLK_NEEDS_REDO)
8858 : {
8859 GIC 1274 : Page page = (Page) BufferGetPage(buffer);
8860 : OffsetNumber *nowunused;
8861 : Size datalen;
8862 : OffsetNumber *offnum;
8863 :
8864 1274 : nowunused = (OffsetNumber *) XLogRecGetBlockData(record, 0, &datalen);
8865 :
8866 ECB : /* Shouldn't be a record unless there's something to do */
8867 GIC 1274 : Assert(xlrec->nunused > 0);
8868 ECB :
8869 : /* Update all now-unused line pointers */
8870 CBC 1274 : offnum = nowunused;
8871 132840 : for (int i = 0; i < xlrec->nunused; i++)
8872 : {
8873 GIC 131566 : OffsetNumber off = *offnum++;
8874 CBC 131566 : ItemId lp = PageGetItemId(page, off);
8875 :
8876 GIC 131566 : Assert(ItemIdIsDead(lp) && !ItemIdHasStorage(lp));
8877 131566 : ItemIdSetUnused(lp);
8878 : }
8879 :
8880 : /* Attempt to truncate line pointer array now */
8881 1274 : PageTruncateLinePointerArray(page);
8882 :
8883 CBC 1274 : PageSetLSN(page, lsn);
8884 GIC 1274 : MarkBufferDirty(buffer);
8885 ECB : }
8886 :
8887 GIC 1348 : if (BufferIsValid(buffer))
8888 : {
8889 1348 : Size freespace = PageGetHeapFreeSpace(BufferGetPage(buffer));
8890 : RelFileLocator rlocator;
8891 :
8892 GNC 1348 : XLogRecGetBlockTag(record, 0, &rlocator, NULL, &blkno);
8893 :
8894 GIC 1348 : UnlockReleaseBuffer(buffer);
8895 :
8896 : /*
8897 : * After vacuuming LP_DEAD items from a page, it's useful to update
8898 ECB : * the FSM about it, as it may cause the page become target for
8899 : * insertions later even if vacuum decides not to visit it (which is
8900 : * possible if gets marked all-visible.)
8901 : *
8902 : * Do this regardless of a full-page image being applied, since the
8903 : * FSM data is not in the page anyway.
8904 : */
8905 GNC 1348 : XLogRecordPageWithFreeSpace(rlocator, blkno, freespace);
8906 : }
8907 CBC 1348 : }
8908 :
8909 ECB : /*
8910 : * Replay XLOG_HEAP2_VISIBLE record.
8911 : *
8912 : * The critical integrity requirement here is that we must never end up with
8913 : * a situation where the visibility map bit is set, and the page-level
8914 : * PD_ALL_VISIBLE bit is clear. If that were to occur, then a subsequent
8915 : * page modification would fail to clear the visibility map bit.
8916 : */
8917 : static void
8918 GIC 3638 : heap_xlog_visible(XLogReaderState *record)
8919 : {
8920 3638 : XLogRecPtr lsn = record->EndRecPtr;
8921 CBC 3638 : xl_heap_visible *xlrec = (xl_heap_visible *) XLogRecGetData(record);
8922 GIC 3638 : Buffer vmbuffer = InvalidBuffer;
8923 ECB : Buffer buffer;
8924 : Page page;
8925 : RelFileLocator rlocator;
8926 : BlockNumber blkno;
8927 : XLogRedoAction action;
8928 :
8929 GNC 3638 : Assert((xlrec->flags & VISIBILITYMAP_XLOG_VALID_BITS) == xlrec->flags);
8930 :
8931 3638 : XLogRecGetBlockTag(record, 1, &rlocator, NULL, &blkno);
8932 :
8933 : /*
8934 ECB : * If there are any Hot Standby transactions running that have an xmin
8935 : * horizon old enough that this page isn't all-visible for them, they
8936 : * might incorrectly decide that an index-only scan can skip a heap fetch.
8937 : *
8938 : * NB: It might be better to throw some kind of "soft" conflict here that
8939 : * forces any index-only scan that is in flight to perform heap fetches,
8940 : * rather than killing the transaction outright.
8941 : */
8942 GIC 3638 : if (InHotStandby)
8943 GNC 3509 : ResolveRecoveryConflictWithSnapshot(xlrec->snapshotConflictHorizon,
8944 3509 : xlrec->flags & VISIBILITYMAP_XLOG_CATALOG_REL,
8945 : rlocator);
8946 ECB :
8947 : /*
8948 : * Read the heap page, if it still exists. If the heap file has dropped or
8949 : * truncated later in recovery, we don't need to update the page, but we'd
8950 : * better still update the visibility map.
8951 : */
8952 CBC 3638 : action = XLogReadBufferForRedo(record, 1, &buffer);
8953 GIC 3638 : if (action == BLK_NEEDS_REDO)
8954 ECB : {
8955 : /*
8956 : * We don't bump the LSN of the heap page when setting the visibility
8957 : * map bit (unless checksums or wal_hint_bits is enabled, in which
8958 : * case we must). This exposes us to torn page hazards, but since
8959 : * we're not inspecting the existing page contents in any way, we
8960 : * don't care.
8961 : */
8962 GIC 2615 : page = BufferGetPage(buffer);
8963 :
8964 2615 : PageSetAllVisible(page);
8965 :
8966 2615 : if (XLogHintBitIsNeeded())
8967 2570 : PageSetLSN(page, lsn);
8968 :
8969 CBC 2615 : MarkBufferDirty(buffer);
8970 : }
8971 ECB : else if (action == BLK_RESTORED)
8972 : {
8973 : /*
8974 : * If heap block was backed up, we already restored it and there's
8975 : * nothing more to do. (This can only happen with checksums or
8976 : * wal_log_hints enabled.)
8977 : */
8978 : }
8979 :
8980 GIC 3638 : if (BufferIsValid(buffer))
8981 ECB : {
8982 GIC 3593 : Size space = PageGetFreeSpace(BufferGetPage(buffer));
8983 ECB :
8984 GIC 3593 : UnlockReleaseBuffer(buffer);
8985 ECB :
8986 : /*
8987 : * Since FSM is not WAL-logged and only updated heuristically, it
8988 : * easily becomes stale in standbys. If the standby is later promoted
8989 : * and runs VACUUM, it will skip updating individual free space
8990 : * figures for pages that became all-visible (or all-frozen, depending
8991 : * on the vacuum mode,) which is troublesome when FreeSpaceMapVacuum
8992 : * propagates too optimistic free space values to upper FSM layers;
8993 : * later inserters try to use such pages only to find out that they
8994 : * are unusable. This can cause long stalls when there are many such
8995 : * pages.
8996 : *
8997 : * Forestall those problems by updating FSM's idea about a page that
8998 : * is becoming all-visible or all-frozen.
8999 : *
9000 : * Do this regardless of a full-page image being applied, since the
9001 : * FSM data is not in the page anyway.
9002 : */
9003 CBC 3593 : if (xlrec->flags & VISIBILITYMAP_VALID_BITS)
9004 GNC 3593 : XLogRecordPageWithFreeSpace(rlocator, blkno, space);
9005 : }
9006 :
9007 ECB : /*
9008 : * Even if we skipped the heap page update due to the LSN interlock, it's
9009 : * still safe to update the visibility map. Any WAL record that clears
9010 : * the visibility map bit does so before checking the page LSN, so any
9011 : * bits that need to be cleared will still be cleared.
9012 : */
9013 CBC 3638 : if (XLogReadBufferForRedoExtended(record, 0, RBM_ZERO_ON_ERROR, false,
9014 : &vmbuffer) == BLK_NEEDS_REDO)
9015 ECB : {
9016 GIC 3443 : Page vmpage = BufferGetPage(vmbuffer);
9017 : Relation reln;
9018 : uint8 vmbits;
9019 ECB :
9020 : /* initialize the page if it was read as zeros */
9021 CBC 3443 : if (PageIsNew(vmpage))
9022 UIC 0 : PageInit(vmpage, BLCKSZ, 0);
9023 :
9024 : /* remove VISIBILITYMAP_XLOG_* */
9025 GNC 3443 : vmbits = xlrec->flags & VISIBILITYMAP_VALID_BITS;
9026 :
9027 : /*
9028 : * XLogReadBufferForRedoExtended locked the buffer. But
9029 : * visibilitymap_set will handle locking itself.
9030 : */
9031 GIC 3443 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
9032 :
9033 GNC 3443 : reln = CreateFakeRelcacheEntry(rlocator);
9034 GIC 3443 : visibilitymap_pin(reln, blkno, &vmbuffer);
9035 ECB :
9036 GNC 3443 : visibilitymap_set(reln, blkno, InvalidBuffer, lsn, vmbuffer,
9037 : xlrec->snapshotConflictHorizon, vmbits);
9038 ECB :
9039 CBC 3443 : ReleaseBuffer(vmbuffer);
9040 3443 : FreeFakeRelcacheEntry(reln);
9041 : }
9042 GIC 195 : else if (BufferIsValid(vmbuffer))
9043 195 : UnlockReleaseBuffer(vmbuffer);
9044 3638 : }
9045 :
9046 : /*
9047 ECB : * Replay XLOG_HEAP2_FREEZE_PAGE records
9048 : */
9049 : static void
9050 GIC 90 : heap_xlog_freeze_page(XLogReaderState *record)
9051 : {
9052 90 : XLogRecPtr lsn = record->EndRecPtr;
9053 90 : xl_heap_freeze_page *xlrec = (xl_heap_freeze_page *) XLogRecGetData(record);
9054 : Buffer buffer;
9055 :
9056 : /*
9057 : * In Hot Standby mode, ensure that there's no queries running which still
9058 ECB : * consider the frozen xids as running.
9059 : */
9060 CBC 90 : if (InHotStandby)
9061 : {
9062 : RelFileLocator rlocator;
9063 :
9064 GNC 90 : XLogRecGetBlockTag(record, 0, &rlocator, NULL, NULL);
9065 90 : ResolveRecoveryConflictWithSnapshot(xlrec->snapshotConflictHorizon,
9066 90 : xlrec->isCatalogRel,
9067 : rlocator);
9068 ECB : }
9069 :
9070 GIC 90 : if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
9071 : {
9072 89 : Page page = BufferGetPage(buffer);
9073 : xl_heap_freeze_plan *plans;
9074 : OffsetNumber *offsets;
9075 GNC 89 : int curoff = 0;
9076 :
9077 89 : plans = (xl_heap_freeze_plan *) XLogRecGetBlockData(record, 0, NULL);
9078 89 : offsets = (OffsetNumber *) ((char *) plans +
9079 89 : (xlrec->nplans *
9080 : sizeof(xl_heap_freeze_plan)));
9081 196 : for (int p = 0; p < xlrec->nplans; p++)
9082 ECB : {
9083 : HeapTupleFreeze frz;
9084 :
9085 : /*
9086 : * Convert freeze plan representation from WAL record into
9087 : * per-tuple format used by heap_execute_freeze_tuple
9088 : */
9089 GNC 107 : frz.xmax = plans[p].xmax;
9090 107 : frz.t_infomask2 = plans[p].t_infomask2;
9091 107 : frz.t_infomask = plans[p].t_infomask;
9092 107 : frz.frzflags = plans[p].frzflags;
9093 107 : frz.offset = InvalidOffsetNumber; /* unused, but be tidy */
9094 :
9095 2496 : for (int i = 0; i < plans[p].ntuples; i++)
9096 : {
9097 2389 : OffsetNumber offset = offsets[curoff++];
9098 : ItemId lp;
9099 : HeapTupleHeader tuple;
9100 :
9101 2389 : lp = PageGetItemId(page, offset);
9102 2389 : tuple = (HeapTupleHeader) PageGetItem(page, lp);
9103 2389 : heap_execute_freeze_tuple(tuple, &frz);
9104 : }
9105 : }
9106 :
9107 GIC 89 : PageSetLSN(page, lsn);
9108 89 : MarkBufferDirty(buffer);
9109 : }
9110 90 : if (BufferIsValid(buffer))
9111 CBC 90 : UnlockReleaseBuffer(buffer);
9112 GIC 90 : }
9113 ECB :
9114 : /*
9115 : * Given an "infobits" field from an XLog record, set the correct bits in the
9116 : * given infomask and infomask2 for the tuple touched by the record.
9117 : *
9118 : * (This is the reverse of compute_infobits).
9119 : */
9120 : static void
9121 GIC 418292 : fix_infomask_from_infobits(uint8 infobits, uint16 *infomask, uint16 *infomask2)
9122 : {
9123 418292 : *infomask &= ~(HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY |
9124 : HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_EXCL_LOCK);
9125 418292 : *infomask2 &= ~HEAP_KEYS_UPDATED;
9126 :
9127 418292 : if (infobits & XLHL_XMAX_IS_MULTI)
9128 2 : *infomask |= HEAP_XMAX_IS_MULTI;
9129 418292 : if (infobits & XLHL_XMAX_LOCK_ONLY)
9130 54090 : *infomask |= HEAP_XMAX_LOCK_ONLY;
9131 418292 : if (infobits & XLHL_XMAX_EXCL_LOCK)
9132 53796 : *infomask |= HEAP_XMAX_EXCL_LOCK;
9133 : /* note HEAP_XMAX_SHR_LOCK isn't considered here */
9134 CBC 418292 : if (infobits & XLHL_XMAX_KEYSHR_LOCK)
9135 305 : *infomask |= HEAP_XMAX_KEYSHR_LOCK;
9136 :
9137 GIC 418292 : if (infobits & XLHL_KEYS_UPDATED)
9138 278893 : *infomask2 |= HEAP_KEYS_UPDATED;
9139 418292 : }
9140 :
9141 : static void
9142 277802 : heap_xlog_delete(XLogReaderState *record)
9143 : {
9144 CBC 277802 : XLogRecPtr lsn = record->EndRecPtr;
9145 GIC 277802 : xl_heap_delete *xlrec = (xl_heap_delete *) XLogRecGetData(record);
9146 : Buffer buffer;
9147 ECB : Page page;
9148 GIC 277802 : ItemId lp = NULL;
9149 : HeapTupleHeader htup;
9150 : BlockNumber blkno;
9151 : RelFileLocator target_locator;
9152 ECB : ItemPointerData target_tid;
9153 EUB :
9154 GNC 277802 : XLogRecGetBlockTag(record, 0, &target_locator, NULL, &blkno);
9155 GIC 277802 : ItemPointerSetBlockNumber(&target_tid, blkno);
9156 CBC 277802 : ItemPointerSetOffsetNumber(&target_tid, xlrec->offnum);
9157 :
9158 : /*
9159 : * The visibility map may need to be fixed even if the heap page is
9160 : * already up-to-date.
9161 : */
9162 277802 : if (xlrec->flags & XLH_DELETE_ALL_VISIBLE_CLEARED)
9163 : {
9164 GNC 4 : Relation reln = CreateFakeRelcacheEntry(target_locator);
9165 CBC 4 : Buffer vmbuffer = InvalidBuffer;
9166 :
9167 4 : visibilitymap_pin(reln, blkno, &vmbuffer);
9168 GIC 4 : visibilitymap_clear(reln, blkno, vmbuffer, VISIBILITYMAP_VALID_BITS);
9169 4 : ReleaseBuffer(vmbuffer);
9170 CBC 4 : FreeFakeRelcacheEntry(reln);
9171 ECB : }
9172 :
9173 CBC 277802 : if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
9174 ECB : {
9175 CBC 277729 : page = BufferGetPage(buffer);
9176 :
9177 GIC 277729 : if (PageGetMaxOffsetNumber(page) >= xlrec->offnum)
9178 277729 : lp = PageGetItemId(page, xlrec->offnum);
9179 :
9180 277729 : if (PageGetMaxOffsetNumber(page) < xlrec->offnum || !ItemIdIsNormal(lp))
9181 LBC 0 : elog(PANIC, "invalid lp");
9182 :
9183 CBC 277729 : htup = (HeapTupleHeader) PageGetItem(page, lp);
9184 ECB :
9185 GIC 277729 : htup->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
9186 277729 : htup->t_infomask2 &= ~HEAP_KEYS_UPDATED;
9187 277729 : HeapTupleHeaderClearHotUpdated(htup);
9188 277729 : fix_infomask_from_infobits(xlrec->infobits_set,
9189 : &htup->t_infomask, &htup->t_infomask2);
9190 277729 : if (!(xlrec->flags & XLH_DELETE_IS_SUPER))
9191 CBC 277729 : HeapTupleHeaderSetXmax(htup, xlrec->xmax);
9192 : else
9193 UIC 0 : HeapTupleHeaderSetXmin(htup, InvalidTransactionId);
9194 GIC 277729 : HeapTupleHeaderSetCmax(htup, FirstCommandId, false);
9195 ECB :
9196 : /* Mark the page as a candidate for pruning */
9197 CBC 277729 : PageSetPrunable(page, XLogRecGetXid(record));
9198 :
9199 GIC 277729 : if (xlrec->flags & XLH_DELETE_ALL_VISIBLE_CLEARED)
9200 3 : PageClearAllVisible(page);
9201 ECB :
9202 : /* Make sure t_ctid is set correctly */
9203 CBC 277729 : if (xlrec->flags & XLH_DELETE_IS_PARTITION_MOVE)
9204 GIC 119 : HeapTupleHeaderSetMovedPartitions(htup);
9205 : else
9206 CBC 277610 : htup->t_ctid = target_tid;
9207 GIC 277729 : PageSetLSN(page, lsn);
9208 CBC 277729 : MarkBufferDirty(buffer);
9209 ECB : }
9210 CBC 277802 : if (BufferIsValid(buffer))
9211 GIC 277802 : UnlockReleaseBuffer(buffer);
9212 CBC 277802 : }
9213 :
9214 : static void
9215 GIC 1197422 : heap_xlog_insert(XLogReaderState *record)
9216 : {
9217 1197422 : XLogRecPtr lsn = record->EndRecPtr;
9218 1197422 : xl_heap_insert *xlrec = (xl_heap_insert *) XLogRecGetData(record);
9219 : Buffer buffer;
9220 ECB : Page page;
9221 : union
9222 : {
9223 : HeapTupleHeaderData hdr;
9224 : char data[MaxHeapTupleSize];
9225 : } tbuf;
9226 : HeapTupleHeader htup;
9227 : xl_heap_header xlhdr;
9228 : uint32 newlen;
9229 GIC 1197422 : Size freespace = 0;
9230 : RelFileLocator target_locator;
9231 : BlockNumber blkno;
9232 ECB : ItemPointerData target_tid;
9233 : XLogRedoAction action;
9234 :
9235 GNC 1197422 : XLogRecGetBlockTag(record, 0, &target_locator, NULL, &blkno);
9236 GIC 1197422 : ItemPointerSetBlockNumber(&target_tid, blkno);
9237 1197422 : ItemPointerSetOffsetNumber(&target_tid, xlrec->offnum);
9238 ECB :
9239 : /*
9240 : * The visibility map may need to be fixed even if the heap page is
9241 : * already up-to-date.
9242 : */
9243 CBC 1197422 : if (xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED)
9244 : {
9245 GNC 509 : Relation reln = CreateFakeRelcacheEntry(target_locator);
9246 GIC 509 : Buffer vmbuffer = InvalidBuffer;
9247 :
9248 509 : visibilitymap_pin(reln, blkno, &vmbuffer);
9249 509 : visibilitymap_clear(reln, blkno, vmbuffer, VISIBILITYMAP_VALID_BITS);
9250 509 : ReleaseBuffer(vmbuffer);
9251 509 : FreeFakeRelcacheEntry(reln);
9252 ECB : }
9253 :
9254 : /*
9255 : * If we inserted the first and only tuple on the page, re-initialize the
9256 : * page from scratch.
9257 : */
9258 CBC 1197422 : if (XLogRecGetInfo(record) & XLOG_HEAP_INIT_PAGE)
9259 ECB : {
9260 CBC 15692 : buffer = XLogInitBufferForRedo(record, 0);
9261 15692 : page = BufferGetPage(buffer);
9262 15692 : PageInit(page, BufferGetPageSize(buffer), 0);
9263 15692 : action = BLK_NEEDS_REDO;
9264 : }
9265 ECB : else
9266 CBC 1181730 : action = XLogReadBufferForRedo(record, 0, &buffer);
9267 GIC 1197422 : if (action == BLK_NEEDS_REDO)
9268 ECB : {
9269 : Size datalen;
9270 : char *data;
9271 :
9272 GIC 1196841 : page = BufferGetPage(buffer);
9273 ECB :
9274 GIC 1196841 : if (PageGetMaxOffsetNumber(page) + 1 < xlrec->offnum)
9275 LBC 0 : elog(PANIC, "invalid max offset number");
9276 ECB :
9277 GIC 1196841 : data = XLogRecGetBlockData(record, 0, &datalen);
9278 :
9279 CBC 1196841 : newlen = datalen - SizeOfHeapHeader;
9280 GIC 1196841 : Assert(datalen > SizeOfHeapHeader && newlen <= MaxHeapTupleSize);
9281 1196841 : memcpy((char *) &xlhdr, data, SizeOfHeapHeader);
9282 1196841 : data += SizeOfHeapHeader;
9283 :
9284 1196841 : htup = &tbuf.hdr;
9285 CBC 1196841 : MemSet((char *) htup, 0, SizeofHeapTupleHeader);
9286 ECB : /* PG73FORMAT: get bitmap [+ padding] [+ oid] + data */
9287 CBC 1196841 : memcpy((char *) htup + SizeofHeapTupleHeader,
9288 : data,
9289 : newlen);
9290 GIC 1196841 : newlen += SizeofHeapTupleHeader;
9291 1196841 : htup->t_infomask2 = xlhdr.t_infomask2;
9292 1196841 : htup->t_infomask = xlhdr.t_infomask;
9293 CBC 1196841 : htup->t_hoff = xlhdr.t_hoff;
9294 GIC 1196841 : HeapTupleHeaderSetXmin(htup, XLogRecGetXid(record));
9295 CBC 1196841 : HeapTupleHeaderSetCmin(htup, FirstCommandId);
9296 1196841 : htup->t_ctid = target_tid;
9297 :
9298 1196841 : if (PageAddItem(page, (Item) htup, newlen, xlrec->offnum,
9299 ECB : true, true) == InvalidOffsetNumber)
9300 LBC 0 : elog(PANIC, "failed to add tuple");
9301 ECB :
9302 GIC 1196841 : freespace = PageGetHeapFreeSpace(page); /* needed to update FSM below */
9303 :
9304 CBC 1196841 : PageSetLSN(page, lsn);
9305 :
9306 1196841 : if (xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED)
9307 GIC 305 : PageClearAllVisible(page);
9308 ECB :
9309 : /* XLH_INSERT_ALL_FROZEN_SET implies that all tuples are visible */
9310 GIC 1196841 : if (xlrec->flags & XLH_INSERT_ALL_FROZEN_SET)
9311 LBC 0 : PageSetAllVisible(page);
9312 EUB :
9313 GIC 1196841 : MarkBufferDirty(buffer);
9314 ECB : }
9315 GIC 1197422 : if (BufferIsValid(buffer))
9316 CBC 1197422 : UnlockReleaseBuffer(buffer);
9317 ECB :
9318 : /*
9319 : * If the page is running low on free space, update the FSM as well.
9320 : * Arbitrarily, our definition of "low" is less than 20%. We can't do much
9321 : * better than that without knowing the fill-factor for the table.
9322 : *
9323 : * XXX: Don't do this if the page was restored from full page image. We
9324 EUB : * don't bother to update the FSM in that case, it doesn't need to be
9325 ECB : * totally accurate anyway.
9326 : */
9327 GIC 1197422 : if (action == BLK_NEEDS_REDO && freespace < BLCKSZ / 5)
9328 GNC 237217 : XLogRecordPageWithFreeSpace(target_locator, blkno, freespace);
9329 GIC 1197422 : }
9330 ECB :
9331 : /*
9332 : * Handles MULTI_INSERT record type.
9333 : */
9334 : static void
9335 CBC 46851 : heap_xlog_multi_insert(XLogReaderState *record)
9336 : {
9337 46851 : XLogRecPtr lsn = record->EndRecPtr;
9338 ECB : xl_heap_multi_insert *xlrec;
9339 : RelFileLocator rlocator;
9340 : BlockNumber blkno;
9341 : Buffer buffer;
9342 : Page page;
9343 : union
9344 : {
9345 : HeapTupleHeaderData hdr;
9346 : char data[MaxHeapTupleSize];
9347 : } tbuf;
9348 : HeapTupleHeader htup;
9349 : uint32 newlen;
9350 GIC 46851 : Size freespace = 0;
9351 : int i;
9352 46851 : bool isinit = (XLogRecGetInfo(record) & XLOG_HEAP_INIT_PAGE) != 0;
9353 : XLogRedoAction action;
9354 :
9355 : /*
9356 : * Insertion doesn't overwrite MVCC data, so no conflict processing is
9357 : * required.
9358 : */
9359 46851 : xlrec = (xl_heap_multi_insert *) XLogRecGetData(record);
9360 ECB :
9361 GNC 46851 : XLogRecGetBlockTag(record, 0, &rlocator, NULL, &blkno);
9362 :
9363 : /* check that the mutually exclusive flags are not both set */
9364 GIC 46851 : Assert(!((xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED) &&
9365 : (xlrec->flags & XLH_INSERT_ALL_FROZEN_SET)));
9366 ECB :
9367 : /*
9368 : * The visibility map may need to be fixed even if the heap page is
9369 : * already up-to-date.
9370 : */
9371 GIC 46851 : if (xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED)
9372 : {
9373 GNC 477 : Relation reln = CreateFakeRelcacheEntry(rlocator);
9374 CBC 477 : Buffer vmbuffer = InvalidBuffer;
9375 :
9376 477 : visibilitymap_pin(reln, blkno, &vmbuffer);
9377 477 : visibilitymap_clear(reln, blkno, vmbuffer, VISIBILITYMAP_VALID_BITS);
9378 GIC 477 : ReleaseBuffer(vmbuffer);
9379 CBC 477 : FreeFakeRelcacheEntry(reln);
9380 ECB : }
9381 :
9382 CBC 46851 : if (isinit)
9383 : {
9384 GIC 1979 : buffer = XLogInitBufferForRedo(record, 0);
9385 1979 : page = BufferGetPage(buffer);
9386 1979 : PageInit(page, BufferGetPageSize(buffer), 0);
9387 1979 : action = BLK_NEEDS_REDO;
9388 : }
9389 ECB : else
9390 GIC 44872 : action = XLogReadBufferForRedo(record, 0, &buffer);
9391 CBC 46851 : if (action == BLK_NEEDS_REDO)
9392 ECB : {
9393 : char *tupdata;
9394 : char *endptr;
9395 : Size len;
9396 :
9397 : /* Tuples are stored as block data */
9398 CBC 46406 : tupdata = XLogRecGetBlockData(record, 0, &len);
9399 GIC 46406 : endptr = tupdata + len;
9400 :
9401 46406 : page = (Page) BufferGetPage(buffer);
9402 :
9403 CBC 236311 : for (i = 0; i < xlrec->ntuples; i++)
9404 : {
9405 ECB : OffsetNumber offnum;
9406 EUB : xl_multi_insert_tuple *xlhdr;
9407 :
9408 ECB : /*
9409 : * If we're reinitializing the page, the tuples are stored in
9410 : * order from FirstOffsetNumber. Otherwise there's an array of
9411 : * offsets in the WAL record, and the tuples come after that.
9412 : */
9413 CBC 189905 : if (isinit)
9414 GIC 99841 : offnum = FirstOffsetNumber + i;
9415 ECB : else
9416 CBC 90064 : offnum = xlrec->offsets[i];
9417 GIC 189905 : if (PageGetMaxOffsetNumber(page) + 1 < offnum)
9418 LBC 0 : elog(PANIC, "invalid max offset number");
9419 :
9420 GIC 189905 : xlhdr = (xl_multi_insert_tuple *) SHORTALIGN(tupdata);
9421 CBC 189905 : tupdata = ((char *) xlhdr) + SizeOfMultiInsertTuple;
9422 ECB :
9423 CBC 189905 : newlen = xlhdr->datalen;
9424 189905 : Assert(newlen <= MaxHeapTupleSize);
9425 189905 : htup = &tbuf.hdr;
9426 189905 : MemSet((char *) htup, 0, SizeofHeapTupleHeader);
9427 ECB : /* PG73FORMAT: get bitmap [+ padding] [+ oid] + data */
9428 GIC 189905 : memcpy((char *) htup + SizeofHeapTupleHeader,
9429 ECB : (char *) tupdata,
9430 : newlen);
9431 GBC 189905 : tupdata += newlen;
9432 :
9433 CBC 189905 : newlen += SizeofHeapTupleHeader;
9434 GIC 189905 : htup->t_infomask2 = xlhdr->t_infomask2;
9435 CBC 189905 : htup->t_infomask = xlhdr->t_infomask;
9436 GIC 189905 : htup->t_hoff = xlhdr->t_hoff;
9437 CBC 189905 : HeapTupleHeaderSetXmin(htup, XLogRecGetXid(record));
9438 189905 : HeapTupleHeaderSetCmin(htup, FirstCommandId);
9439 GIC 189905 : ItemPointerSetBlockNumber(&htup->t_ctid, blkno);
9440 189905 : ItemPointerSetOffsetNumber(&htup->t_ctid, offnum);
9441 ECB :
9442 GBC 189905 : offnum = PageAddItem(page, (Item) htup, newlen, offnum, true, true);
9443 GIC 189905 : if (offnum == InvalidOffsetNumber)
9444 LBC 0 : elog(PANIC, "failed to add tuple");
9445 : }
9446 CBC 46406 : if (tupdata != endptr)
9447 LBC 0 : elog(PANIC, "total tuple length mismatch");
9448 :
9449 GIC 46406 : freespace = PageGetHeapFreeSpace(page); /* needed to update FSM below */
9450 :
9451 46406 : PageSetLSN(page, lsn);
9452 :
9453 46406 : if (xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED)
9454 181 : PageClearAllVisible(page);
9455 :
9456 : /* XLH_INSERT_ALL_FROZEN_SET implies that all tuples are visible */
9457 46406 : if (xlrec->flags & XLH_INSERT_ALL_FROZEN_SET)
9458 CBC 4 : PageSetAllVisible(page);
9459 ECB :
9460 CBC 46406 : MarkBufferDirty(buffer);
9461 : }
9462 GIC 46851 : if (BufferIsValid(buffer))
9463 46851 : UnlockReleaseBuffer(buffer);
9464 :
9465 : /*
9466 ECB : * If the page is running low on free space, update the FSM as well.
9467 : * Arbitrarily, our definition of "low" is less than 20%. We can't do much
9468 : * better than that without knowing the fill-factor for the table.
9469 : *
9470 : * XXX: Don't do this if the page was restored from full page image. We
9471 : * don't bother to update the FSM in that case, it doesn't need to be
9472 : * totally accurate anyway.
9473 : */
9474 GIC 46851 : if (action == BLK_NEEDS_REDO && freespace < BLCKSZ / 5)
9475 GNC 11673 : XLogRecordPageWithFreeSpace(rlocator, blkno, freespace);
9476 GIC 46851 : }
9477 :
9478 : /*
9479 : * Handles UPDATE and HOT_UPDATE
9480 : */
9481 ECB : static void
9482 GIC 86497 : heap_xlog_update(XLogReaderState *record, bool hot_update)
9483 ECB : {
9484 GIC 86497 : XLogRecPtr lsn = record->EndRecPtr;
9485 86497 : xl_heap_update *xlrec = (xl_heap_update *) XLogRecGetData(record);
9486 : RelFileLocator rlocator;
9487 : BlockNumber oldblk;
9488 : BlockNumber newblk;
9489 : ItemPointerData newtid;
9490 ECB : Buffer obuffer,
9491 : nbuffer;
9492 : Page page;
9493 : OffsetNumber offnum;
9494 GIC 86497 : ItemId lp = NULL;
9495 ECB : HeapTupleData oldtup;
9496 : HeapTupleHeader htup;
9497 GIC 86497 : uint16 prefixlen = 0,
9498 86497 : suffixlen = 0;
9499 : char *newp;
9500 : union
9501 : {
9502 ECB : HeapTupleHeaderData hdr;
9503 : char data[MaxHeapTupleSize];
9504 : } tbuf;
9505 : xl_heap_header xlhdr;
9506 : uint32 newlen;
9507 CBC 86497 : Size freespace = 0;
9508 ECB : XLogRedoAction oldaction;
9509 : XLogRedoAction newaction;
9510 :
9511 : /* initialize to keep the compiler quiet */
9512 GIC 86497 : oldtup.t_data = NULL;
9513 CBC 86497 : oldtup.t_len = 0;
9514 :
9515 GNC 86497 : XLogRecGetBlockTag(record, 0, &rlocator, NULL, &newblk);
9516 CBC 86497 : if (XLogRecGetBlockTagExtended(record, 1, NULL, NULL, &oldblk, NULL))
9517 ECB : {
9518 : /* HOT updates are never done across pages */
9519 GIC 53381 : Assert(!hot_update);
9520 : }
9521 ECB : else
9522 CBC 33116 : oldblk = newblk;
9523 :
9524 GIC 86497 : ItemPointerSet(&newtid, newblk, xlrec->new_offnum);
9525 :
9526 : /*
9527 : * The visibility map may need to be fixed even if the heap page is
9528 : * already up-to-date.
9529 ECB : */
9530 CBC 86497 : if (xlrec->flags & XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED)
9531 : {
9532 GNC 245 : Relation reln = CreateFakeRelcacheEntry(rlocator);
9533 GIC 245 : Buffer vmbuffer = InvalidBuffer;
9534 ECB :
9535 GIC 245 : visibilitymap_pin(reln, oldblk, &vmbuffer);
9536 245 : visibilitymap_clear(reln, oldblk, vmbuffer, VISIBILITYMAP_VALID_BITS);
9537 245 : ReleaseBuffer(vmbuffer);
9538 245 : FreeFakeRelcacheEntry(reln);
9539 : }
9540 :
9541 : /*
9542 : * In normal operation, it is important to lock the two pages in
9543 : * page-number order, to avoid possible deadlocks against other update
9544 ECB : * operations going the other way. However, during WAL replay there can
9545 : * be no other update happening, so we don't need to worry about that. But
9546 : * we *do* need to worry that we don't expose an inconsistent state to Hot
9547 : * Standby queries --- so the original page can't be unlocked before we've
9548 : * added the new tuple to the new page.
9549 EUB : */
9550 :
9551 ECB : /* Deal with old tuple version */
9552 CBC 86497 : oldaction = XLogReadBufferForRedo(record, (oldblk == newblk) ? 0 : 1,
9553 : &obuffer);
9554 86497 : if (oldaction == BLK_NEEDS_REDO)
9555 ECB : {
9556 CBC 86473 : page = BufferGetPage(obuffer);
9557 86473 : offnum = xlrec->old_offnum;
9558 GIC 86473 : if (PageGetMaxOffsetNumber(page) >= offnum)
9559 CBC 86473 : lp = PageGetItemId(page, offnum);
9560 :
9561 GIC 86473 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
9562 LBC 0 : elog(PANIC, "invalid lp");
9563 :
9564 CBC 86473 : htup = (HeapTupleHeader) PageGetItem(page, lp);
9565 ECB :
9566 CBC 86473 : oldtup.t_data = htup;
9567 86473 : oldtup.t_len = ItemIdGetLength(lp);
9568 ECB :
9569 CBC 86473 : htup->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
9570 86473 : htup->t_infomask2 &= ~HEAP_KEYS_UPDATED;
9571 86473 : if (hot_update)
9572 GIC 30897 : HeapTupleHeaderSetHotUpdated(htup);
9573 ECB : else
9574 CBC 55576 : HeapTupleHeaderClearHotUpdated(htup);
9575 GBC 86473 : fix_infomask_from_infobits(xlrec->old_infobits_set, &htup->t_infomask,
9576 : &htup->t_infomask2);
9577 CBC 86473 : HeapTupleHeaderSetXmax(htup, xlrec->old_xmax);
9578 GBC 86473 : HeapTupleHeaderSetCmax(htup, FirstCommandId, false);
9579 : /* Set forward chain link in t_ctid */
9580 CBC 86473 : htup->t_ctid = newtid;
9581 :
9582 ECB : /* Mark the page as a candidate for pruning */
9583 GIC 86473 : PageSetPrunable(page, XLogRecGetXid(record));
9584 ECB :
9585 CBC 86473 : if (xlrec->flags & XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED)
9586 GIC 243 : PageClearAllVisible(page);
9587 :
9588 CBC 86473 : PageSetLSN(page, lsn);
9589 86473 : MarkBufferDirty(obuffer);
9590 : }
9591 ECB :
9592 : /*
9593 : * Read the page the new tuple goes into, if different from old.
9594 : */
9595 GIC 86497 : if (oldblk == newblk)
9596 : {
9597 33116 : nbuffer = obuffer;
9598 33116 : newaction = oldaction;
9599 : }
9600 53381 : else if (XLogRecGetInfo(record) & XLOG_HEAP_INIT_PAGE)
9601 : {
9602 594 : nbuffer = XLogInitBufferForRedo(record, 0);
9603 594 : page = (Page) BufferGetPage(nbuffer);
9604 594 : PageInit(page, BufferGetPageSize(nbuffer), 0);
9605 CBC 594 : newaction = BLK_NEEDS_REDO;
9606 ECB : }
9607 : else
9608 GIC 52787 : newaction = XLogReadBufferForRedo(record, 0, &nbuffer);
9609 :
9610 : /*
9611 : * The visibility map may need to be fixed even if the heap page is
9612 : * already up-to-date.
9613 ECB : */
9614 GIC 86497 : if (xlrec->flags & XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED)
9615 ECB : {
9616 GNC 36 : Relation reln = CreateFakeRelcacheEntry(rlocator);
9617 GIC 36 : Buffer vmbuffer = InvalidBuffer;
9618 :
9619 36 : visibilitymap_pin(reln, newblk, &vmbuffer);
9620 36 : visibilitymap_clear(reln, newblk, vmbuffer, VISIBILITYMAP_VALID_BITS);
9621 36 : ReleaseBuffer(vmbuffer);
9622 36 : FreeFakeRelcacheEntry(reln);
9623 : }
9624 :
9625 ECB : /* Deal with new tuple */
9626 GIC 86497 : if (newaction == BLK_NEEDS_REDO)
9627 : {
9628 ECB : char *recdata;
9629 : char *recdata_end;
9630 : Size datalen;
9631 : Size tuplen;
9632 :
9633 GIC 86462 : recdata = XLogRecGetBlockData(record, 0, &datalen);
9634 86462 : recdata_end = recdata + datalen;
9635 :
9636 86462 : page = BufferGetPage(nbuffer);
9637 :
9638 CBC 86462 : offnum = xlrec->new_offnum;
9639 GIC 86462 : if (PageGetMaxOffsetNumber(page) + 1 < offnum)
9640 UIC 0 : elog(PANIC, "invalid max offset number");
9641 :
9642 GIC 86462 : if (xlrec->flags & XLH_UPDATE_PREFIX_FROM_OLD)
9643 ECB : {
9644 CBC 12522 : Assert(newblk == oldblk);
9645 GIC 12522 : memcpy(&prefixlen, recdata, sizeof(uint16));
9646 CBC 12522 : recdata += sizeof(uint16);
9647 ECB : }
9648 GIC 86462 : if (xlrec->flags & XLH_UPDATE_SUFFIX_FROM_OLD)
9649 : {
9650 CBC 29908 : Assert(newblk == oldblk);
9651 GIC 29908 : memcpy(&suffixlen, recdata, sizeof(uint16));
9652 29908 : recdata += sizeof(uint16);
9653 ECB : }
9654 :
9655 CBC 86462 : memcpy((char *) &xlhdr, recdata, SizeOfHeapHeader);
9656 GIC 86462 : recdata += SizeOfHeapHeader;
9657 :
9658 86462 : tuplen = recdata_end - recdata;
9659 86462 : Assert(tuplen <= MaxHeapTupleSize);
9660 :
9661 CBC 86462 : htup = &tbuf.hdr;
9662 GIC 86462 : MemSet((char *) htup, 0, SizeofHeapTupleHeader);
9663 ECB :
9664 : /*
9665 : * Reconstruct the new tuple using the prefix and/or suffix from the
9666 : * old tuple, and the data stored in the WAL record.
9667 : */
9668 CBC 86462 : newp = (char *) htup + SizeofHeapTupleHeader;
9669 86462 : if (prefixlen > 0)
9670 : {
9671 : int len;
9672 :
9673 : /* copy bitmap [+ padding] [+ oid] from WAL record */
9674 GIC 12522 : len = xlhdr.t_hoff - SizeofHeapTupleHeader;
9675 12522 : memcpy(newp, recdata, len);
9676 12522 : recdata += len;
9677 12522 : newp += len;
9678 :
9679 : /* copy prefix from old tuple */
9680 12522 : memcpy(newp, (char *) oldtup.t_data + oldtup.t_data->t_hoff, prefixlen);
9681 12522 : newp += prefixlen;
9682 :
9683 ECB : /* copy new tuple data from WAL record */
9684 GIC 12522 : len = tuplen - (xlhdr.t_hoff - SizeofHeapTupleHeader);
9685 CBC 12522 : memcpy(newp, recdata, len);
9686 GIC 12522 : recdata += len;
9687 CBC 12522 : newp += len;
9688 ECB : }
9689 : else
9690 : {
9691 : /*
9692 : * copy bitmap [+ padding] [+ oid] + data from record, all in one
9693 EUB : * go
9694 : */
9695 CBC 73940 : memcpy(newp, recdata, tuplen);
9696 GIC 73940 : recdata += tuplen;
9697 CBC 73940 : newp += tuplen;
9698 ECB : }
9699 GIC 86462 : Assert(recdata == recdata_end);
9700 ECB :
9701 : /* copy suffix from old tuple */
9702 CBC 86462 : if (suffixlen > 0)
9703 29908 : memcpy(newp, (char *) oldtup.t_data + oldtup.t_len - suffixlen, suffixlen);
9704 :
9705 86462 : newlen = SizeofHeapTupleHeader + tuplen + prefixlen + suffixlen;
9706 86462 : htup->t_infomask2 = xlhdr.t_infomask2;
9707 GIC 86462 : htup->t_infomask = xlhdr.t_infomask;
9708 CBC 86462 : htup->t_hoff = xlhdr.t_hoff;
9709 ECB :
9710 GIC 86462 : HeapTupleHeaderSetXmin(htup, XLogRecGetXid(record));
9711 CBC 86462 : HeapTupleHeaderSetCmin(htup, FirstCommandId);
9712 GIC 86462 : HeapTupleHeaderSetXmax(htup, xlrec->new_xmax);
9713 : /* Make sure there is no forward chain link in t_ctid */
9714 CBC 86462 : htup->t_ctid = newtid;
9715 :
9716 86462 : offnum = PageAddItem(page, (Item) htup, newlen, offnum, true, true);
9717 86462 : if (offnum == InvalidOffsetNumber)
9718 UIC 0 : elog(PANIC, "failed to add tuple");
9719 ECB :
9720 CBC 86462 : if (xlrec->flags & XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED)
9721 GIC 28 : PageClearAllVisible(page);
9722 :
9723 86462 : freespace = PageGetHeapFreeSpace(page); /* needed to update FSM below */
9724 :
9725 86462 : PageSetLSN(page, lsn);
9726 CBC 86462 : MarkBufferDirty(nbuffer);
9727 : }
9728 ECB :
9729 CBC 86497 : if (BufferIsValid(nbuffer) && nbuffer != obuffer)
9730 GIC 53381 : UnlockReleaseBuffer(nbuffer);
9731 CBC 86497 : if (BufferIsValid(obuffer))
9732 GIC 86497 : UnlockReleaseBuffer(obuffer);
9733 ECB :
9734 : /*
9735 : * If the new page is running low on free space, update the FSM as well.
9736 : * Arbitrarily, our definition of "low" is less than 20%. We can't do much
9737 : * better than that without knowing the fill-factor for the table.
9738 : *
9739 : * However, don't update the FSM on HOT updates, because after crash
9740 : * recovery, either the old or the new tuple will certainly be dead and
9741 : * prunable. After pruning, the page will have roughly as much free space
9742 : * as it did before the update, assuming the new tuple is about the same
9743 : * size as the old one.
9744 : *
9745 : * XXX: Don't do this if the page was restored from full page image. We
9746 : * don't bother to update the FSM in that case, it doesn't need to be
9747 : * totally accurate anyway.
9748 : */
9749 GIC 86497 : if (newaction == BLK_NEEDS_REDO && !hot_update && freespace < BLCKSZ / 5)
9750 GNC 11226 : XLogRecordPageWithFreeSpace(rlocator, newblk, freespace);
9751 CBC 86497 : }
9752 ECB :
9753 : static void
9754 GIC 61 : heap_xlog_confirm(XLogReaderState *record)
9755 : {
9756 61 : XLogRecPtr lsn = record->EndRecPtr;
9757 CBC 61 : xl_heap_confirm *xlrec = (xl_heap_confirm *) XLogRecGetData(record);
9758 : Buffer buffer;
9759 : Page page;
9760 : OffsetNumber offnum;
9761 GIC 61 : ItemId lp = NULL;
9762 : HeapTupleHeader htup;
9763 :
9764 CBC 61 : if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
9765 ECB : {
9766 GIC 61 : page = BufferGetPage(buffer);
9767 ECB :
9768 GIC 61 : offnum = xlrec->offnum;
9769 CBC 61 : if (PageGetMaxOffsetNumber(page) >= offnum)
9770 61 : lp = PageGetItemId(page, offnum);
9771 EUB :
9772 GIC 61 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
9773 LBC 0 : elog(PANIC, "invalid lp");
9774 :
9775 CBC 61 : htup = (HeapTupleHeader) PageGetItem(page, lp);
9776 ECB :
9777 : /*
9778 : * Confirm tuple as actually inserted
9779 : */
9780 GIC 61 : ItemPointerSet(&htup->t_ctid, BufferGetBlockNumber(buffer), offnum);
9781 ECB :
9782 CBC 61 : PageSetLSN(page, lsn);
9783 61 : MarkBufferDirty(buffer);
9784 : }
9785 GIC 61 : if (BufferIsValid(buffer))
9786 CBC 61 : UnlockReleaseBuffer(buffer);
9787 61 : }
9788 :
9789 ECB : static void
9790 CBC 54149 : heap_xlog_lock(XLogReaderState *record)
9791 : {
9792 54149 : XLogRecPtr lsn = record->EndRecPtr;
9793 54149 : xl_heap_lock *xlrec = (xl_heap_lock *) XLogRecGetData(record);
9794 : Buffer buffer;
9795 : Page page;
9796 : OffsetNumber offnum;
9797 GIC 54149 : ItemId lp = NULL;
9798 : HeapTupleHeader htup;
9799 ECB :
9800 : /*
9801 : * The visibility map may need to be fixed even if the heap page is
9802 : * already up-to-date.
9803 : */
9804 GIC 54149 : if (xlrec->flags & XLH_LOCK_ALL_FROZEN_CLEARED)
9805 ECB : {
9806 : RelFileLocator rlocator;
9807 CBC 17 : Buffer vmbuffer = InvalidBuffer;
9808 ECB : BlockNumber block;
9809 : Relation reln;
9810 :
9811 GNC 17 : XLogRecGetBlockTag(record, 0, &rlocator, NULL, &block);
9812 17 : reln = CreateFakeRelcacheEntry(rlocator);
9813 :
9814 GIC 17 : visibilitymap_pin(reln, block, &vmbuffer);
9815 CBC 17 : visibilitymap_clear(reln, block, vmbuffer, VISIBILITYMAP_ALL_FROZEN);
9816 ECB :
9817 CBC 17 : ReleaseBuffer(vmbuffer);
9818 17 : FreeFakeRelcacheEntry(reln);
9819 : }
9820 :
9821 GIC 54149 : if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
9822 : {
9823 54090 : page = (Page) BufferGetPage(buffer);
9824 :
9825 54090 : offnum = xlrec->offnum;
9826 CBC 54090 : if (PageGetMaxOffsetNumber(page) >= offnum)
9827 54090 : lp = PageGetItemId(page, offnum);
9828 ECB :
9829 GIC 54090 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
9830 LBC 0 : elog(PANIC, "invalid lp");
9831 :
9832 GIC 54090 : htup = (HeapTupleHeader) PageGetItem(page, lp);
9833 ECB :
9834 CBC 54090 : htup->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
9835 GIC 54090 : htup->t_infomask2 &= ~HEAP_KEYS_UPDATED;
9836 CBC 54090 : fix_infomask_from_infobits(xlrec->infobits_set, &htup->t_infomask,
9837 ECB : &htup->t_infomask2);
9838 :
9839 : /*
9840 : * Clear relevant update flags, but only if the modified infomask says
9841 : * there's no update.
9842 : */
9843 CBC 54090 : if (HEAP_XMAX_IS_LOCKED_ONLY(htup->t_infomask))
9844 : {
9845 54090 : HeapTupleHeaderClearHotUpdated(htup);
9846 : /* Make sure there is no forward chain link in t_ctid */
9847 54090 : ItemPointerSet(&htup->t_ctid,
9848 ECB : BufferGetBlockNumber(buffer),
9849 EUB : offnum);
9850 : }
9851 CBC 54090 : HeapTupleHeaderSetXmax(htup, xlrec->locking_xid);
9852 54090 : HeapTupleHeaderSetCmax(htup, FirstCommandId, false);
9853 GIC 54090 : PageSetLSN(page, lsn);
9854 CBC 54090 : MarkBufferDirty(buffer);
9855 : }
9856 54149 : if (BufferIsValid(buffer))
9857 54149 : UnlockReleaseBuffer(buffer);
9858 GIC 54149 : }
9859 :
9860 ECB : static void
9861 LBC 0 : heap_xlog_lock_updated(XLogReaderState *record)
9862 ECB : {
9863 LBC 0 : XLogRecPtr lsn = record->EndRecPtr;
9864 : xl_heap_lock_updated *xlrec;
9865 : Buffer buffer;
9866 : Page page;
9867 : OffsetNumber offnum;
9868 UIC 0 : ItemId lp = NULL;
9869 : HeapTupleHeader htup;
9870 :
9871 0 : xlrec = (xl_heap_lock_updated *) XLogRecGetData(record);
9872 :
9873 : /*
9874 : * The visibility map may need to be fixed even if the heap page is
9875 : * already up-to-date.
9876 : */
9877 0 : if (xlrec->flags & XLH_LOCK_ALL_FROZEN_CLEARED)
9878 : {
9879 : RelFileLocator rlocator;
9880 LBC 0 : Buffer vmbuffer = InvalidBuffer;
9881 ECB : BlockNumber block;
9882 : Relation reln;
9883 :
9884 UNC 0 : XLogRecGetBlockTag(record, 0, &rlocator, NULL, &block);
9885 0 : reln = CreateFakeRelcacheEntry(rlocator);
9886 :
9887 LBC 0 : visibilitymap_pin(reln, block, &vmbuffer);
9888 0 : visibilitymap_clear(reln, block, vmbuffer, VISIBILITYMAP_ALL_FROZEN);
9889 :
9890 UIC 0 : ReleaseBuffer(vmbuffer);
9891 0 : FreeFakeRelcacheEntry(reln);
9892 ECB : }
9893 :
9894 UIC 0 : if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
9895 ECB : {
9896 UIC 0 : page = BufferGetPage(buffer);
9897 ECB :
9898 UIC 0 : offnum = xlrec->offnum;
9899 LBC 0 : if (PageGetMaxOffsetNumber(page) >= offnum)
9900 0 : lp = PageGetItemId(page, offnum);
9901 ECB :
9902 UIC 0 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
9903 LBC 0 : elog(PANIC, "invalid lp");
9904 EUB :
9905 UIC 0 : htup = (HeapTupleHeader) PageGetItem(page, lp);
9906 ECB :
9907 UIC 0 : htup->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
9908 0 : htup->t_infomask2 &= ~HEAP_KEYS_UPDATED;
9909 0 : fix_infomask_from_infobits(xlrec->infobits_set, &htup->t_infomask,
9910 : &htup->t_infomask2);
9911 LBC 0 : HeapTupleHeaderSetXmax(htup, xlrec->xmax);
9912 :
9913 0 : PageSetLSN(page, lsn);
9914 0 : MarkBufferDirty(buffer);
9915 : }
9916 0 : if (BufferIsValid(buffer))
9917 0 : UnlockReleaseBuffer(buffer);
9918 0 : }
9919 :
9920 : static void
9921 CBC 6111 : heap_xlog_inplace(XLogReaderState *record)
9922 : {
9923 6111 : XLogRecPtr lsn = record->EndRecPtr;
9924 6111 : xl_heap_inplace *xlrec = (xl_heap_inplace *) XLogRecGetData(record);
9925 : Buffer buffer;
9926 : Page page;
9927 : OffsetNumber offnum;
9928 6111 : ItemId lp = NULL;
9929 : HeapTupleHeader htup;
9930 : uint32 oldlen;
9931 : Size newlen;
9932 :
9933 GIC 6111 : if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
9934 : {
9935 CBC 6072 : char *newtup = XLogRecGetBlockData(record, 0, &newlen);
9936 :
9937 GIC 6072 : page = BufferGetPage(buffer);
9938 ECB :
9939 GIC 6072 : offnum = xlrec->offnum;
9940 6072 : if (PageGetMaxOffsetNumber(page) >= offnum)
9941 6072 : lp = PageGetItemId(page, offnum);
9942 ECB :
9943 CBC 6072 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
9944 UIC 0 : elog(PANIC, "invalid lp");
9945 ECB :
9946 CBC 6072 : htup = (HeapTupleHeader) PageGetItem(page, lp);
9947 :
9948 6072 : oldlen = ItemIdGetLength(lp) - htup->t_hoff;
9949 6072 : if (oldlen != newlen)
9950 UIC 0 : elog(PANIC, "wrong tuple length");
9951 :
9952 CBC 6072 : memcpy((char *) htup + htup->t_hoff, newtup, newlen);
9953 :
9954 6072 : PageSetLSN(page, lsn);
9955 GIC 6072 : MarkBufferDirty(buffer);
9956 ECB : }
9957 CBC 6111 : if (BufferIsValid(buffer))
9958 6111 : UnlockReleaseBuffer(buffer);
9959 GIC 6111 : }
9960 ECB :
9961 EUB : void
9962 GIC 1622042 : heap_redo(XLogReaderState *record)
9963 ECB : {
9964 GIC 1622042 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
9965 ECB :
9966 : /*
9967 : * These operations don't overwrite MVCC data so no conflict processing is
9968 : * required. The ones in heap2 rmgr do.
9969 : */
9970 :
9971 GIC 1622042 : switch (info & XLOG_HEAP_OPMASK)
9972 : {
9973 1197422 : case XLOG_HEAP_INSERT:
9974 CBC 1197422 : heap_xlog_insert(record);
9975 GIC 1197422 : break;
9976 CBC 277802 : case XLOG_HEAP_DELETE:
9977 GIC 277802 : heap_xlog_delete(record);
9978 CBC 277802 : break;
9979 GIC 55585 : case XLOG_HEAP_UPDATE:
9980 55585 : heap_xlog_update(record, false);
9981 55585 : break;
9982 LBC 0 : case XLOG_HEAP_TRUNCATE:
9983 ECB :
9984 : /*
9985 : * TRUNCATE is a no-op because the actions are already logged as
9986 : * SMGR WAL records. TRUNCATE WAL record only exists for logical
9987 : * decoding.
9988 : */
9989 LBC 0 : break;
9990 GIC 30912 : case XLOG_HEAP_HOT_UPDATE:
9991 30912 : heap_xlog_update(record, true);
9992 GBC 30912 : break;
9993 GIC 61 : case XLOG_HEAP_CONFIRM:
9994 GBC 61 : heap_xlog_confirm(record);
9995 GIC 61 : break;
9996 54149 : case XLOG_HEAP_LOCK:
9997 54149 : heap_xlog_lock(record);
9998 54149 : break;
9999 GBC 6111 : case XLOG_HEAP_INPLACE:
10000 GIC 6111 : heap_xlog_inplace(record);
10001 6111 : break;
10002 UBC 0 : default:
10003 UIC 0 : elog(PANIC, "heap_redo: unknown op code %u", info);
10004 : }
10005 GIC 1622042 : }
10006 :
10007 : void
10008 GBC 59400 : heap2_redo(XLogReaderState *record)
10009 : {
10010 GIC 59400 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
10011 EUB :
10012 GIC 59400 : switch (info & XLOG_HEAP_OPMASK)
10013 : {
10014 6391 : case XLOG_HEAP2_PRUNE:
10015 GBC 6391 : heap_xlog_prune(record);
10016 6391 : break;
10017 GIC 1348 : case XLOG_HEAP2_VACUUM:
10018 GBC 1348 : heap_xlog_vacuum(record);
10019 1348 : break;
10020 GIC 90 : case XLOG_HEAP2_FREEZE_PAGE:
10021 GBC 90 : heap_xlog_freeze_page(record);
10022 90 : break;
10023 GIC 3638 : case XLOG_HEAP2_VISIBLE:
10024 3638 : heap_xlog_visible(record);
10025 GBC 3638 : break;
10026 GIC 46851 : case XLOG_HEAP2_MULTI_INSERT:
10027 GBC 46851 : heap_xlog_multi_insert(record);
10028 GIC 46851 : break;
10029 UBC 0 : case XLOG_HEAP2_LOCK_UPDATED:
10030 0 : heap_xlog_lock_updated(record);
10031 0 : break;
10032 GIC 1082 : case XLOG_HEAP2_NEW_CID:
10033 EUB :
10034 : /*
10035 : * Nothing to do on a real replay, only used during logical
10036 : * decoding.
10037 : */
10038 GBC 1082 : break;
10039 UBC 0 : case XLOG_HEAP2_REWRITE:
10040 0 : heap_xlog_logical_rewrite(record);
10041 UIC 0 : break;
10042 UBC 0 : default:
10043 UIC 0 : elog(PANIC, "heap2_redo: unknown op code %u", info);
10044 EUB : }
10045 GBC 59400 : }
10046 :
10047 EUB : /*
10048 : * Mask a heap page before performing consistency checks on it.
10049 : */
10050 : void
10051 UIC 0 : heap_mask(char *pagedata, BlockNumber blkno)
10052 ECB : {
10053 UIC 0 : Page page = (Page) pagedata;
10054 ECB : OffsetNumber off;
10055 :
10056 UIC 0 : mask_page_lsn_and_checksum(page);
10057 :
10058 0 : mask_page_hint_bits(page);
10059 LBC 0 : mask_unused_space(page);
10060 :
10061 UIC 0 : for (off = 1; off <= PageGetMaxOffsetNumber(page); off++)
10062 : {
10063 0 : ItemId iid = PageGetItemId(page, off);
10064 ECB : char *page_item;
10065 :
10066 LBC 0 : page_item = (char *) (page + ItemIdGetOffset(iid));
10067 :
10068 0 : if (ItemIdIsNormal(iid))
10069 : {
10070 0 : HeapTupleHeader page_htup = (HeapTupleHeader) page_item;
10071 ECB :
10072 : /*
10073 : * If xmin of a tuple is not yet frozen, we should ignore
10074 : * differences in hint bits, since they can be set without
10075 EUB : * emitting WAL.
10076 : */
10077 LBC 0 : if (!HeapTupleHeaderXminFrozen(page_htup))
10078 UIC 0 : page_htup->t_infomask &= ~HEAP_XACT_MASK;
10079 ECB : else
10080 : {
10081 EUB : /* Still we need to mask xmax hint bits. */
10082 UIC 0 : page_htup->t_infomask &= ~HEAP_XMAX_INVALID;
10083 LBC 0 : page_htup->t_infomask &= ~HEAP_XMAX_COMMITTED;
10084 : }
10085 ECB :
10086 : /*
10087 : * During replay, we set Command Id to FirstCommandId. Hence, mask
10088 : * it. See heap_xlog_insert() for details.
10089 : */
10090 LBC 0 : page_htup->t_choice.t_heap.t_field3.t_cid = MASK_MARKER;
10091 :
10092 : /*
10093 ECB : * For a speculative tuple, heap_insert() does not set ctid in the
10094 : * caller-passed heap tuple itself, leaving the ctid field to
10095 : * contain a speculative token value - a per-backend monotonically
10096 : * increasing identifier. Besides, it does not WAL-log ctid under
10097 : * any circumstances.
10098 : *
10099 : * During redo, heap_xlog_insert() sets t_ctid to current block
10100 : * number and self offset number. It doesn't care about any
10101 : * speculative insertions on the primary. Hence, we set t_ctid to
10102 : * current block number and self offset number to ignore any
10103 : * inconsistency.
10104 : */
10105 LBC 0 : if (HeapTupleHeaderIsSpeculative(page_htup))
10106 0 : ItemPointerSet(&page_htup->t_ctid, blkno, off);
10107 ECB :
10108 : /*
10109 : * NB: Not ignoring ctid changes due to the tuple having moved
10110 : * (i.e. HeapTupleHeaderIndicatesMovedPartitions), because that's
10111 : * important information that needs to be in-sync between primary
10112 : * and standby, and thus is WAL logged.
10113 EUB : */
10114 : }
10115 :
10116 : /*
10117 : * Ignore any padding bytes after the tuple, when the length of the
10118 : * item is not MAXALIGNed.
10119 : */
10120 UBC 0 : if (ItemIdHasStorage(iid))
10121 ECB : {
10122 LBC 0 : int len = ItemIdGetLength(iid);
10123 0 : int padlen = MAXALIGN(len) - len;
10124 ECB :
10125 LBC 0 : if (padlen > 0)
10126 0 : memset(page_item + len, MASK_MARKER, padlen);
10127 ECB : }
10128 : }
10129 LBC 0 : }
10130 ECB :
10131 : /*
10132 : * HeapCheckForSerializableConflictOut
10133 EUB : * We are reading a tuple. If it's not visible, there may be a
10134 : * rw-conflict out with the inserter. Otherwise, if it is visible to us
10135 : * but has been deleted, there may be a rw-conflict out with the deleter.
10136 ECB : *
10137 : * We will determine the top level xid of the writing transaction with which
10138 : * we may be in conflict, and ask CheckForSerializableConflictOut() to check
10139 : * for overlap with our own transaction.
10140 : *
10141 : * This function should be called just about anywhere in heapam.c where a
10142 : * tuple has been read. The caller must hold at least a shared lock on the
10143 : * buffer, because this function might set hint bits on the tuple. There is
10144 : * currently no known reason to call this function from an index AM.
10145 : */
10146 : void
10147 CBC 241099374 : HeapCheckForSerializableConflictOut(bool visible, Relation relation,
10148 ECB : HeapTuple tuple, Buffer buffer,
10149 : Snapshot snapshot)
10150 : {
10151 : TransactionId xid;
10152 : HTSV_Result htsvResult;
10153 :
10154 CBC 241099374 : if (!CheckForSerializableConflictOutNeeded(relation, snapshot))
10155 241074039 : return;
10156 ECB :
10157 : /*
10158 : * Check to see whether the tuple has been written to by a concurrent
10159 : * transaction, either to create it not visible to us, or to delete it
10160 EUB : * while it is visible to us. The "visible" bool indicates whether the
10161 : * tuple is visible to us, while HeapTupleSatisfiesVacuum checks what else
10162 : * is going on with it.
10163 ECB : *
10164 : * In the event of a concurrently inserted tuple that also happens to have
10165 : * been concurrently updated (by a separate transaction), the xmin of the
10166 : * tuple will be used -- not the updater's xid.
10167 : */
10168 GIC 25335 : htsvResult = HeapTupleSatisfiesVacuum(tuple, TransactionXmin, buffer);
10169 CBC 25335 : switch (htsvResult)
10170 EUB : {
10171 GBC 24533 : case HEAPTUPLE_LIVE:
10172 24533 : if (visible)
10173 24520 : return;
10174 13 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
10175 GIC 13 : break;
10176 CBC 352 : case HEAPTUPLE_RECENTLY_DEAD:
10177 : case HEAPTUPLE_DELETE_IN_PROGRESS:
10178 GIC 352 : if (visible)
10179 281 : xid = HeapTupleHeaderGetUpdateXid(tuple->t_data);
10180 : else
10181 71 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
10182 EUB :
10183 GIC 352 : if (TransactionIdPrecedes(xid, TransactionXmin))
10184 EUB : {
10185 : /* This is like the HEAPTUPLE_DEAD case */
10186 GIC 62 : Assert(!visible);
10187 GBC 62 : return;
10188 : }
10189 290 : break;
10190 326 : case HEAPTUPLE_INSERT_IN_PROGRESS:
10191 GIC 326 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
10192 GBC 326 : break;
10193 GIC 124 : case HEAPTUPLE_DEAD:
10194 GBC 124 : Assert(!visible);
10195 GIC 124 : return;
10196 UIC 0 : default:
10197 EUB :
10198 : /*
10199 : * The only way to get to this default clause is if a new value is
10200 : * added to the enum type without adding it to this switch
10201 : * statement. That's a bug, so elog.
10202 : */
10203 UIC 0 : elog(ERROR, "unrecognized return value from HeapTupleSatisfiesVacuum: %u", htsvResult);
10204 :
10205 : /*
10206 : * In spite of having all enum values covered and calling elog on
10207 : * this default, some compilers think this is a code path which
10208 EUB : * allows xid to be used below without initialization. Silence
10209 : * that warning.
10210 : */
10211 : xid = InvalidTransactionId;
10212 : }
10213 :
10214 GBC 629 : Assert(TransactionIdIsValid(xid));
10215 GIC 629 : Assert(TransactionIdFollowsOrEquals(xid, TransactionXmin));
10216 :
10217 : /*
10218 : * Find top level xid. Bail out if xid is too early to be a conflict, or
10219 : * if it's our own xid.
10220 : */
10221 GBC 629 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
10222 GIC 62 : return;
10223 567 : xid = SubTransGetTopmostTransaction(xid);
10224 567 : if (TransactionIdPrecedes(xid, TransactionXmin))
10225 UIC 0 : return;
10226 :
10227 GIC 567 : CheckForSerializableConflictOut(relation, xid, snapshot);
10228 : }
|
