Branch data Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * heapam.c
4 : : * heap access method code
5 : : *
6 : : * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
7 : : * Portions Copyright (c) 1994, Regents of the University of California
8 : : *
9 : : *
10 : : * IDENTIFICATION
11 : : * src/backend/access/heap/heapam.c
12 : : *
13 : : *
14 : : * INTERFACE ROUTINES
15 : : * heap_beginscan - begin relation scan
16 : : * heap_rescan - restart a relation scan
17 : : * heap_endscan - end relation scan
18 : : * heap_getnext - retrieve next tuple in scan
19 : : * heap_fetch - retrieve tuple with given tid
20 : : * heap_insert - insert tuple into a relation
21 : : * heap_multi_insert - insert multiple tuples into a relation
22 : : * heap_delete - delete a tuple from a relation
23 : : * heap_update - replace a tuple in a relation with another tuple
24 : : *
25 : : * NOTES
26 : : * This file contains the heap_ routines which implement
27 : : * the POSTGRES heap access method used for all POSTGRES
28 : : * relations.
29 : : *
30 : : *-------------------------------------------------------------------------
31 : : */
32 : : #include "postgres.h"
33 : :
34 : : #include "access/heapam.h"
35 : : #include "access/heaptoast.h"
36 : : #include "access/hio.h"
37 : : #include "access/multixact.h"
38 : : #include "access/subtrans.h"
39 : : #include "access/syncscan.h"
40 : : #include "access/valid.h"
41 : : #include "access/visibilitymap.h"
42 : : #include "access/xloginsert.h"
43 : : #include "catalog/pg_database.h"
44 : : #include "catalog/pg_database_d.h"
45 : : #include "commands/vacuum.h"
46 : : #include "executor/instrument_node.h"
47 : : #include "pgstat.h"
48 : : #include "port/pg_bitutils.h"
49 : : #include "storage/lmgr.h"
50 : : #include "storage/predicate.h"
51 : : #include "storage/proc.h"
52 : : #include "storage/procarray.h"
53 : : #include "utils/datum.h"
54 : : #include "utils/injection_point.h"
55 : : #include "utils/inval.h"
56 : : #include "utils/spccache.h"
57 : : #include "utils/syscache.h"
58 : :
59 : :
60 : : static HeapTuple heap_prepare_insert(Relation relation, HeapTuple tup,
61 : : TransactionId xid, CommandId cid, uint32 options);
62 : : static XLogRecPtr log_heap_update(Relation reln, Buffer oldbuf,
63 : : Buffer vmbuffer_old, Buffer newbuf,
64 : : Buffer vmbuffer_new, HeapTuple oldtup,
65 : : HeapTuple newtup, HeapTuple old_key_tuple,
66 : : bool all_visible_cleared, bool new_all_visible_cleared,
67 : : bool walLogical);
68 : : #ifdef USE_ASSERT_CHECKING
69 : : static void check_lock_if_inplace_updateable_rel(Relation relation,
70 : : const ItemPointerData *otid,
71 : : HeapTuple newtup);
72 : : static void check_inplace_rel_lock(HeapTuple oldtup);
73 : : #endif
74 : : static Bitmapset *HeapDetermineColumnsInfo(Relation relation,
75 : : Bitmapset *interesting_cols,
76 : : Bitmapset *external_cols,
77 : : HeapTuple oldtup, HeapTuple newtup,
78 : : bool *has_external);
79 : : static bool heap_acquire_tuplock(Relation relation, const ItemPointerData *tid,
80 : : LockTupleMode mode, LockWaitPolicy wait_policy,
81 : : bool *have_tuple_lock);
82 : : static inline BlockNumber heapgettup_advance_block(HeapScanDesc scan,
83 : : BlockNumber block,
84 : : ScanDirection dir);
85 : : static pg_noinline BlockNumber heapgettup_initial_block(HeapScanDesc scan,
86 : : ScanDirection dir);
87 : : static void compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
88 : : uint16 old_infomask2, TransactionId add_to_xmax,
89 : : LockTupleMode mode, bool is_update,
90 : : TransactionId *result_xmax, uint16 *result_infomask,
91 : : uint16 *result_infomask2);
92 : : static TM_Result heap_lock_updated_tuple(Relation rel,
93 : : uint16 prior_infomask,
94 : : TransactionId prior_raw_xmax,
95 : : const ItemPointerData *prior_ctid,
96 : : TransactionId xid,
97 : : LockTupleMode mode);
98 : : static void GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
99 : : uint16 *new_infomask2);
100 : : static TransactionId MultiXactIdGetUpdateXid(TransactionId xmax,
101 : : uint16 t_infomask);
102 : : static bool DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
103 : : LockTupleMode lockmode, bool *current_is_member);
104 : : static void MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
105 : : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
106 : : int *remaining);
107 : : static bool ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
108 : : uint16 infomask, Relation rel, int *remaining,
109 : : bool logLockFailure);
110 : : static void index_delete_sort(TM_IndexDeleteOp *delstate);
111 : : static int bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate);
112 : : static XLogRecPtr log_heap_new_cid(Relation relation, HeapTuple tup);
113 : : static HeapTuple ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
114 : : bool *copy);
115 : :
116 : :
117 : : /*
118 : : * This table lists the heavyweight lock mode that corresponds to each tuple
119 : : * lock mode, as well as one or two corresponding MultiXactStatus values:
120 : : * .lockstatus to merely lock tuples, and .updstatus to update them. The
121 : : * latter is set to -1 if the corresponding tuple lock mode does not allow
122 : : * updating tuples -- see get_mxact_status_for_lock().
123 : : *
124 : : * These interact with InplaceUpdateTupleLock, an alias for ExclusiveLock.
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 : : } tupleLockExtraInfo[] =
135 : :
136 : : {
137 : : [LockTupleKeyShare] = {
138 : : .hwlock = AccessShareLock,
139 : : .lockstatus = MultiXactStatusForKeyShare,
140 : : /* KeyShare does not allow updating tuples */
141 : : .updstatus = -1
142 : : },
143 : : [LockTupleShare] = {
144 : : .hwlock = RowShareLock,
145 : : .lockstatus = MultiXactStatusForShare,
146 : : /* Share does not allow updating tuples */
147 : : .updstatus = -1
148 : : },
149 : : [LockTupleNoKeyExclusive] = {
150 : : .hwlock = ExclusiveLock,
151 : : .lockstatus = MultiXactStatusForNoKeyUpdate,
152 : : .updstatus = MultiXactStatusNoKeyUpdate
153 : : },
154 : : [LockTupleExclusive] = {
155 : : .hwlock = AccessExclusiveLock,
156 : : .lockstatus = MultiXactStatusForUpdate,
157 : : .updstatus = MultiXactStatusUpdate
158 : : }
159 : : };
160 : :
161 : : /* Get the LOCKMODE for a given MultiXactStatus */
162 : : #define LOCKMODE_from_mxstatus(status) \
163 : : (tupleLockExtraInfo[TUPLOCK_from_mxstatus((status))].hwlock)
164 : :
165 : : /*
166 : : * Acquire heavyweight locks on tuples, using a LockTupleMode strength value.
167 : : * This is more readable than having every caller translate it to lock.h's
168 : : * LOCKMODE.
169 : : */
170 : : #define LockTupleTuplock(rel, tup, mode) \
171 : : LockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
172 : : #define UnlockTupleTuplock(rel, tup, mode) \
173 : : UnlockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
174 : : #define ConditionalLockTupleTuplock(rel, tup, mode, log) \
175 : : ConditionalLockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock, (log))
176 : :
177 : : #ifdef USE_PREFETCH
178 : : /*
179 : : * heap_index_delete_tuples and index_delete_prefetch_buffer use this
180 : : * structure to coordinate prefetching activity
181 : : */
182 : : typedef struct
183 : : {
184 : : BlockNumber cur_hblkno;
185 : : int next_item;
186 : : int ndeltids;
187 : : TM_IndexDelete *deltids;
188 : : } IndexDeletePrefetchState;
189 : : #endif
190 : :
191 : : /* heap_index_delete_tuples bottom-up index deletion costing constants */
192 : : #define BOTTOMUP_MAX_NBLOCKS 6
193 : : #define BOTTOMUP_TOLERANCE_NBLOCKS 3
194 : :
195 : : /*
196 : : * heap_index_delete_tuples uses this when determining which heap blocks it
197 : : * must visit to help its bottom-up index deletion caller
198 : : */
199 : : typedef struct IndexDeleteCounts
200 : : {
201 : : int16 npromisingtids; /* Number of "promising" TIDs in group */
202 : : int16 ntids; /* Number of TIDs in group */
203 : : int16 ifirsttid; /* Offset to group's first deltid */
204 : : } IndexDeleteCounts;
205 : :
206 : : /*
207 : : * This table maps tuple lock strength values for each particular
208 : : * MultiXactStatus value.
209 : : */
210 : : static const int MultiXactStatusLock[MaxMultiXactStatus + 1] =
211 : : {
212 : : LockTupleKeyShare, /* ForKeyShare */
213 : : LockTupleShare, /* ForShare */
214 : : LockTupleNoKeyExclusive, /* ForNoKeyUpdate */
215 : : LockTupleExclusive, /* ForUpdate */
216 : : LockTupleNoKeyExclusive, /* NoKeyUpdate */
217 : : LockTupleExclusive /* Update */
218 : : };
219 : :
220 : : /* Get the LockTupleMode for a given MultiXactStatus */
221 : : #define TUPLOCK_from_mxstatus(status) \
222 : : (MultiXactStatusLock[(status)])
223 : :
224 : : /*
225 : : * Check that we have a valid snapshot if we might need TOAST access.
226 : : */
227 : : static inline void
228 : 16665118 : AssertHasSnapshotForToast(Relation rel)
229 : : {
230 : : #ifdef USE_ASSERT_CHECKING
231 : :
232 : : /* bootstrap mode in particular breaks this rule */
233 : : if (!IsNormalProcessingMode())
234 : : return;
235 : :
236 : : /* if the relation doesn't have a TOAST table, we are good */
237 : : if (!OidIsValid(rel->rd_rel->reltoastrelid))
238 : : return;
239 : :
240 : : Assert(HaveRegisteredOrActiveSnapshot());
241 : :
242 : : #endif /* USE_ASSERT_CHECKING */
243 : 16665118 : }
244 : :
245 : : /* ----------------------------------------------------------------
246 : : * heap support routines
247 : : * ----------------------------------------------------------------
248 : : */
249 : :
250 : : /*
251 : : * Streaming read API callback for parallel sequential scans. Returns the next
252 : : * block the caller wants from the read stream or InvalidBlockNumber when done.
253 : : */
254 : : static BlockNumber
255 : 146259 : heap_scan_stream_read_next_parallel(ReadStream *stream,
256 : : void *callback_private_data,
257 : : void *per_buffer_data)
258 : : {
259 : 146259 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
260 : :
261 : : Assert(ScanDirectionIsForward(scan->rs_dir));
262 : : Assert(scan->rs_base.rs_parallel);
263 : :
264 [ + + ]: 146259 : if (unlikely(!scan->rs_inited))
265 : : {
266 : : /* parallel scan */
267 : 2781 : table_block_parallelscan_startblock_init(scan->rs_base.rs_rd,
268 : 2781 : scan->rs_parallelworkerdata,
269 : 2781 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel,
270 : : scan->rs_startblock,
271 : : scan->rs_numblocks);
272 : :
273 : : /* may return InvalidBlockNumber if there are no more blocks */
274 : 5562 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
275 : 2781 : scan->rs_parallelworkerdata,
276 : 2781 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
277 : 2781 : scan->rs_inited = true;
278 : : }
279 : : else
280 : : {
281 : 143478 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
282 : 143478 : scan->rs_parallelworkerdata, (ParallelBlockTableScanDesc)
283 : 143478 : scan->rs_base.rs_parallel);
284 : : }
285 : :
286 : 146259 : return scan->rs_prefetch_block;
287 : : }
288 : :
289 : : /*
290 : : * Streaming read API callback for serial sequential and TID range scans.
291 : : * Returns the next block the caller wants from the read stream or
292 : : * InvalidBlockNumber when done.
293 : : */
294 : : static BlockNumber
295 : 4429962 : heap_scan_stream_read_next_serial(ReadStream *stream,
296 : : void *callback_private_data,
297 : : void *per_buffer_data)
298 : : {
299 : 4429962 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
300 : :
301 [ + + ]: 4429962 : if (unlikely(!scan->rs_inited))
302 : : {
303 : 1090255 : scan->rs_prefetch_block = heapgettup_initial_block(scan, scan->rs_dir);
304 : 1090255 : scan->rs_inited = true;
305 : : }
306 : : else
307 : 3339707 : scan->rs_prefetch_block = heapgettup_advance_block(scan,
308 : : scan->rs_prefetch_block,
309 : : scan->rs_dir);
310 : :
311 : 4429962 : return scan->rs_prefetch_block;
312 : : }
313 : :
314 : : /*
315 : : * Read stream API callback for bitmap heap scans.
316 : : * Returns the next block the caller wants from the read stream or
317 : : * InvalidBlockNumber when done.
318 : : */
319 : : static BlockNumber
320 : 260507 : bitmapheap_stream_read_next(ReadStream *pgsr, void *private_data,
321 : : void *per_buffer_data)
322 : : {
323 : 260507 : TBMIterateResult *tbmres = per_buffer_data;
324 : 260507 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) private_data;
325 : 260507 : HeapScanDesc hscan = (HeapScanDesc) bscan;
326 : 260507 : TableScanDesc sscan = &hscan->rs_base;
327 : :
328 : : for (;;)
329 : : {
330 [ - + ]: 260507 : CHECK_FOR_INTERRUPTS();
331 : :
332 : : /* no more entries in the bitmap */
333 [ + + ]: 260507 : if (!tbm_iterate(&sscan->st.rs_tbmiterator, tbmres))
334 : 15847 : return InvalidBlockNumber;
335 : :
336 : : /*
337 : : * Ignore any claimed entries past what we think is the end of the
338 : : * relation. It may have been extended after the start of our scan (we
339 : : * only hold an AccessShareLock, and it could be inserts from this
340 : : * backend). We don't take this optimization in SERIALIZABLE
341 : : * isolation though, as we need to examine all invisible tuples
342 : : * reachable by the index.
343 : : */
344 [ + + ]: 244660 : if (!IsolationIsSerializable() &&
345 [ - + ]: 244551 : tbmres->blockno >= hscan->rs_nblocks)
346 : 0 : continue;
347 : :
348 : 244660 : return tbmres->blockno;
349 : : }
350 : :
351 : : /* not reachable */
352 : : Assert(false);
353 : : }
354 : :
355 : : /* ----------------
356 : : * initscan - scan code common to heap_beginscan and heap_rescan
357 : : * ----------------
358 : : */
359 : : static void
360 : 1122721 : initscan(HeapScanDesc scan, ScanKey key, bool keep_startblock)
361 : : {
362 : 1122721 : ParallelBlockTableScanDesc bpscan = NULL;
363 : : bool allow_strat;
364 : : bool allow_sync;
365 : :
366 : : /*
367 : : * Determine the number of blocks we have to scan.
368 : : *
369 : : * It is sufficient to do this once at scan start, since any tuples added
370 : : * while the scan is in progress will be invisible to my snapshot anyway.
371 : : * (That is not true when using a non-MVCC snapshot. However, we couldn't
372 : : * guarantee to return tuples added after scan start anyway, since they
373 : : * might go into pages we already scanned. To guarantee consistent
374 : : * results for a non-MVCC snapshot, the caller must hold some higher-level
375 : : * lock that ensures the interesting tuple(s) won't change.)
376 : : */
377 [ + + ]: 1122721 : if (scan->rs_base.rs_parallel != NULL)
378 : : {
379 : 4671 : bpscan = (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel;
380 : 4671 : scan->rs_nblocks = bpscan->phs_nblocks;
381 : : }
382 : : else
383 : 1118050 : scan->rs_nblocks = RelationGetNumberOfBlocks(scan->rs_base.rs_rd);
384 : :
385 : : /*
386 : : * If the table is large relative to NBuffers, use a bulk-read access
387 : : * strategy and enable synchronized scanning (see syncscan.c). Although
388 : : * the thresholds for these features could be different, we make them the
389 : : * same so that there are only two behaviors to tune rather than four.
390 : : * (However, some callers need to be able to disable one or both of these
391 : : * behaviors, independently of the size of the table; also there is a GUC
392 : : * variable that can disable synchronized scanning.)
393 : : *
394 : : * Note that table_block_parallelscan_initialize has a very similar test;
395 : : * if you change this, consider changing that one, too.
396 : : */
397 [ + + ]: 1122719 : if (!RelationUsesLocalBuffers(scan->rs_base.rs_rd) &&
398 [ + + ]: 1112723 : scan->rs_nblocks > NBuffers / 4)
399 : : {
400 : 14057 : allow_strat = (scan->rs_base.rs_flags & SO_ALLOW_STRAT) != 0;
401 : 14057 : allow_sync = (scan->rs_base.rs_flags & SO_ALLOW_SYNC) != 0;
402 : : }
403 : : else
404 : 1108662 : allow_strat = allow_sync = false;
405 : :
406 [ + + ]: 1122719 : if (allow_strat)
407 : : {
408 : : /* During a rescan, keep the previous strategy object. */
409 [ + + ]: 12713 : if (scan->rs_strategy == NULL)
410 : 12530 : scan->rs_strategy = GetAccessStrategy(BAS_BULKREAD);
411 : : }
412 : : else
413 : : {
414 [ - + ]: 1110006 : if (scan->rs_strategy != NULL)
415 : 0 : FreeAccessStrategy(scan->rs_strategy);
416 : 1110006 : scan->rs_strategy = NULL;
417 : : }
418 : :
419 [ + + ]: 1122719 : if (scan->rs_base.rs_parallel != NULL)
420 : : {
421 : : /* For parallel scan, believe whatever ParallelTableScanDesc says. */
422 [ + + ]: 4671 : if (scan->rs_base.rs_parallel->phs_syncscan)
423 : 5 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
424 : : else
425 : 4666 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
426 : :
427 : : /*
428 : : * If not rescanning, initialize the startblock. Finding the actual
429 : : * start location is done in table_block_parallelscan_startblock_init,
430 : : * based on whether an alternative start location has been set with
431 : : * heap_setscanlimits, or using the syncscan location, when syncscan
432 : : * is enabled.
433 : : */
434 [ + + ]: 4671 : if (!keep_startblock)
435 : 4519 : scan->rs_startblock = InvalidBlockNumber;
436 : : }
437 : : else
438 : : {
439 [ + + ]: 1118048 : if (keep_startblock)
440 : : {
441 : : /*
442 : : * When rescanning, we want to keep the previous startblock
443 : : * setting, so that rewinding a cursor doesn't generate surprising
444 : : * results. Reset the active syncscan setting, though.
445 : : */
446 [ + + + + ]: 638419 : if (allow_sync && synchronize_seqscans)
447 : 50 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
448 : : else
449 : 638369 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
450 : : }
451 [ + + + + ]: 479629 : else if (allow_sync && synchronize_seqscans)
452 : : {
453 : 98 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
454 : 98 : scan->rs_startblock = ss_get_location(scan->rs_base.rs_rd, scan->rs_nblocks);
455 : : }
456 : : else
457 : : {
458 : 479531 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
459 : 479531 : scan->rs_startblock = 0;
460 : : }
461 : : }
462 : :
463 : 1122719 : scan->rs_numblocks = InvalidBlockNumber;
464 : 1122719 : scan->rs_inited = false;
465 : 1122719 : scan->rs_ctup.t_data = NULL;
466 : 1122719 : ItemPointerSetInvalid(&scan->rs_ctup.t_self);
467 : 1122719 : scan->rs_cbuf = InvalidBuffer;
468 : 1122719 : scan->rs_cblock = InvalidBlockNumber;
469 : 1122719 : scan->rs_ntuples = 0;
470 : 1122719 : scan->rs_cindex = 0;
471 : :
472 : : /*
473 : : * Initialize to ForwardScanDirection because it is most common and
474 : : * because heap scans go forward before going backward (e.g. CURSORs).
475 : : */
476 : 1122719 : scan->rs_dir = ForwardScanDirection;
477 : 1122719 : scan->rs_prefetch_block = InvalidBlockNumber;
478 : :
479 : : /* page-at-a-time fields are always invalid when not rs_inited */
480 : :
481 : : /*
482 : : * copy the scan key, if appropriate
483 : : */
484 [ + + + + ]: 1122719 : if (key != NULL && scan->rs_base.rs_nkeys > 0)
485 : 269900 : memcpy(scan->rs_base.rs_key, key, scan->rs_base.rs_nkeys * sizeof(ScanKeyData));
486 : :
487 : : /*
488 : : * Currently, we only have a stats counter for sequential heap scans (but
489 : : * e.g for bitmap scans the underlying bitmap index scans will be counted,
490 : : * and for sample scans we update stats for tuple fetches).
491 : : */
492 [ + + ]: 1122719 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN)
493 [ + + + + : 1093594 : pgstat_count_heap_scan(scan->rs_base.rs_rd);
+ + ]
494 : 1122719 : }
495 : :
496 : : /*
497 : : * heap_setscanlimits - restrict range of a heapscan
498 : : *
499 : : * startBlk is the page to start at
500 : : * numBlks is number of pages to scan (InvalidBlockNumber means "all")
501 : : */
502 : : void
503 : 3285 : heap_setscanlimits(TableScanDesc sscan, BlockNumber startBlk, BlockNumber numBlks)
504 : : {
505 : 3285 : HeapScanDesc scan = (HeapScanDesc) sscan;
506 : :
507 : : Assert(!scan->rs_inited); /* else too late to change */
508 : : /* else rs_startblock is significant */
509 : : Assert(!(scan->rs_base.rs_flags & SO_ALLOW_SYNC));
510 : :
511 : : /* Check startBlk is valid (but allow case of zero blocks...) */
512 : : Assert(startBlk == 0 || startBlk < scan->rs_nblocks);
513 : :
514 : 3285 : scan->rs_startblock = startBlk;
515 : 3285 : scan->rs_numblocks = numBlks;
516 : 3285 : }
517 : :
518 : : /*
519 : : * Per-tuple loop for heap_prepare_pagescan(). Pulled out so it can be called
520 : : * multiple times, with constant arguments for all_visible,
521 : : * check_serializable.
522 : : */
523 : : pg_always_inline
524 : : static int
525 : 3350759 : page_collect_tuples(HeapScanDesc scan, Snapshot snapshot,
526 : : Page page, Buffer buffer,
527 : : BlockNumber block, int lines,
528 : : bool all_visible, bool check_serializable)
529 : : {
530 : 3350759 : Oid relid = RelationGetRelid(scan->rs_base.rs_rd);
531 : 3350759 : int ntup = 0;
532 : 3350759 : int nvis = 0;
533 : : BatchMVCCState batchmvcc;
534 : :
535 : : /* page at a time should have been disabled otherwise */
536 : : Assert(IsMVCCSnapshot(snapshot));
537 : :
538 : : /* first find all tuples on the page */
539 [ + + ]: 182738075 : for (OffsetNumber lineoff = FirstOffsetNumber; lineoff <= lines; lineoff++)
540 : : {
541 : 179387316 : ItemId lpp = PageGetItemId(page, lineoff);
542 : : HeapTuple tup;
543 : :
544 [ + + ]: 179387316 : if (unlikely(!ItemIdIsNormal(lpp)))
545 : 40999248 : continue;
546 : :
547 : : /*
548 : : * If the page is not all-visible or we need to check serializability,
549 : : * maintain enough state to be able to refind the tuple efficiently,
550 : : * without again first needing to fetch the item and then via that the
551 : : * tuple.
552 : : */
553 [ + + - + ]: 138388068 : if (!all_visible || check_serializable)
554 : : {
555 : 77920746 : tup = &batchmvcc.tuples[ntup];
556 : :
557 : 77920746 : tup->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
558 : 77920746 : tup->t_len = ItemIdGetLength(lpp);
559 : 77920746 : tup->t_tableOid = relid;
560 : 77920746 : ItemPointerSet(&(tup->t_self), block, lineoff);
561 : : }
562 : :
563 : : /*
564 : : * If the page is all visible, these fields otherwise won't be
565 : : * populated in loop below.
566 : : */
567 [ + + ]: 138388068 : if (all_visible)
568 : : {
569 [ - + ]: 60467322 : if (check_serializable)
570 : : {
571 : 0 : batchmvcc.visible[ntup] = true;
572 : : }
573 : 60467322 : scan->rs_vistuples[ntup] = lineoff;
574 : : }
575 : :
576 : 138388068 : ntup++;
577 : : }
578 : :
579 : : Assert(ntup <= MaxHeapTuplesPerPage);
580 : :
581 : : /*
582 : : * Unless the page is all visible, test visibility for all tuples one go.
583 : : * That is considerably more efficient than calling
584 : : * HeapTupleSatisfiesMVCC() one-by-one.
585 : : */
586 [ + + ]: 3350759 : if (all_visible)
587 : 1293833 : nvis = ntup;
588 : : else
589 : 2056926 : nvis = HeapTupleSatisfiesMVCCBatch(snapshot, buffer,
590 : : ntup,
591 : : &batchmvcc,
592 : 2056926 : scan->rs_vistuples);
593 : :
594 : : /*
595 : : * So far we don't have batch API for testing serializabilty, so do so
596 : : * one-by-one.
597 : : */
598 [ + + ]: 3350759 : if (check_serializable)
599 : : {
600 [ + + ]: 2087 : for (int i = 0; i < ntup; i++)
601 : : {
602 : 1464 : HeapCheckForSerializableConflictOut(batchmvcc.visible[i],
603 : : scan->rs_base.rs_rd,
604 : : &batchmvcc.tuples[i],
605 : : buffer, snapshot);
606 : : }
607 : : }
608 : :
609 : 3350751 : return nvis;
610 : : }
611 : :
612 : : /*
613 : : * heap_prepare_pagescan - Prepare current scan page to be scanned in pagemode
614 : : *
615 : : * Preparation currently consists of 1. prune the scan's rs_cbuf page, and 2.
616 : : * fill the rs_vistuples[] array with the OffsetNumbers of visible tuples.
617 : : */
618 : : void
619 : 3350759 : heap_prepare_pagescan(TableScanDesc sscan)
620 : : {
621 : 3350759 : HeapScanDesc scan = (HeapScanDesc) sscan;
622 : 3350759 : Buffer buffer = scan->rs_cbuf;
623 : 3350759 : BlockNumber block = scan->rs_cblock;
624 : : Snapshot snapshot;
625 : : Page page;
626 : : int lines;
627 : : bool all_visible;
628 : : bool check_serializable;
629 : :
630 : : Assert(BufferGetBlockNumber(buffer) == block);
631 : :
632 : : /* ensure we're not accidentally being used when not in pagemode */
633 : : Assert(scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE);
634 : 3350759 : snapshot = scan->rs_base.rs_snapshot;
635 : :
636 : : /*
637 : : * Prune and repair fragmentation for the whole page, if possible.
638 : : */
639 : 3350759 : heap_page_prune_opt(scan->rs_base.rs_rd, buffer, &scan->rs_vmbuffer,
640 : 3350759 : sscan->rs_flags & SO_HINT_REL_READ_ONLY);
641 : :
642 : : /*
643 : : * We must hold share lock on the buffer content while examining tuple
644 : : * visibility. Afterwards, however, the tuples we have found to be
645 : : * visible are guaranteed good as long as we hold the buffer pin.
646 : : */
647 : 3350759 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
648 : :
649 : 3350759 : page = BufferGetPage(buffer);
650 : 3350759 : lines = PageGetMaxOffsetNumber(page);
651 : :
652 : : /*
653 : : * If the all-visible flag indicates that all tuples on the page are
654 : : * visible to everyone, we can skip the per-tuple visibility tests.
655 : : *
656 : : * Note: In hot standby, a tuple that's already visible to all
657 : : * transactions on the primary might still be invisible to a read-only
658 : : * transaction in the standby. We partly handle this problem by tracking
659 : : * the minimum xmin of visible tuples as the cut-off XID while marking a
660 : : * page all-visible on the primary and WAL log that along with the
661 : : * visibility map SET operation. In hot standby, we wait for (or abort)
662 : : * all transactions that can potentially may not see one or more tuples on
663 : : * the page. That's how index-only scans work fine in hot standby. A
664 : : * crucial difference between index-only scans and heap scans is that the
665 : : * index-only scan completely relies on the visibility map where as heap
666 : : * scan looks at the page-level PD_ALL_VISIBLE flag. We are not sure if
667 : : * the page-level flag can be trusted in the same way, because it might
668 : : * get propagated somehow without being explicitly WAL-logged, e.g. via a
669 : : * full page write. Until we can prove that beyond doubt, let's check each
670 : : * tuple for visibility the hard way.
671 : : */
672 [ + + + + ]: 3350759 : all_visible = PageIsAllVisible(page) && !snapshot->takenDuringRecovery;
673 : : check_serializable =
674 : 3350759 : CheckForSerializableConflictOutNeeded(scan->rs_base.rs_rd, snapshot);
675 : :
676 : : /*
677 : : * We call page_collect_tuples() with constant arguments, to get the
678 : : * compiler to constant fold the constant arguments. Separate calls with
679 : : * constant arguments, rather than variables, are needed on several
680 : : * compilers to actually perform constant folding.
681 : : */
682 [ + + ]: 3350759 : if (likely(all_visible))
683 : : {
684 [ + - ]: 1293833 : if (likely(!check_serializable))
685 : 1293833 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
686 : : block, lines, true, false);
687 : : else
688 : 0 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
689 : : block, lines, true, true);
690 : : }
691 : : else
692 : : {
693 [ + + ]: 2056926 : if (likely(!check_serializable))
694 : 2056295 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
695 : : block, lines, false, false);
696 : : else
697 : 631 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
698 : : block, lines, false, true);
699 : : }
700 : :
701 : 3350751 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
702 : 3350751 : }
703 : :
704 : : /*
705 : : * heap_fetch_next_buffer - read and pin the next block from MAIN_FORKNUM.
706 : : *
707 : : * Read the next block of the scan relation from the read stream and save it
708 : : * in the scan descriptor. It is already pinned.
709 : : */
710 : : static inline void
711 : 4360149 : heap_fetch_next_buffer(HeapScanDesc scan, ScanDirection dir)
712 : : {
713 : : Assert(scan->rs_read_stream);
714 : :
715 : : /* release previous scan buffer, if any */
716 [ + + ]: 4360149 : if (BufferIsValid(scan->rs_cbuf))
717 : : {
718 : 3267112 : ReleaseBuffer(scan->rs_cbuf);
719 : 3267112 : scan->rs_cbuf = InvalidBuffer;
720 : : }
721 : :
722 : : /*
723 : : * Be sure to check for interrupts at least once per page. Checks at
724 : : * higher code levels won't be able to stop a seqscan that encounters many
725 : : * pages' worth of consecutive dead tuples.
726 : : */
727 [ + + ]: 4360149 : CHECK_FOR_INTERRUPTS();
728 : :
729 : : /*
730 : : * If the scan direction is changing, reset the prefetch block to the
731 : : * current block. Otherwise, we will incorrectly prefetch the blocks
732 : : * between the prefetch block and the current block again before
733 : : * prefetching blocks in the new, correct scan direction.
734 : : */
735 [ + + ]: 4360147 : if (unlikely(scan->rs_dir != dir))
736 : : {
737 : 102 : scan->rs_prefetch_block = scan->rs_cblock;
738 : 102 : read_stream_reset(scan->rs_read_stream);
739 : : }
740 : :
741 : 4360147 : scan->rs_dir = dir;
742 : :
743 : 4360147 : scan->rs_cbuf = read_stream_next_buffer(scan->rs_read_stream, NULL);
744 [ + + ]: 4360120 : if (BufferIsValid(scan->rs_cbuf))
745 : 3462015 : scan->rs_cblock = BufferGetBlockNumber(scan->rs_cbuf);
746 : 4360120 : }
747 : :
748 : : /*
749 : : * heapgettup_initial_block - return the first BlockNumber to scan
750 : : *
751 : : * Returns InvalidBlockNumber when there are no blocks to scan. This can
752 : : * occur with empty tables and in parallel scans when parallel workers get all
753 : : * of the pages before we can get a chance to get our first page.
754 : : */
755 : : static pg_noinline BlockNumber
756 : 1090255 : heapgettup_initial_block(HeapScanDesc scan, ScanDirection dir)
757 : : {
758 : : Assert(!scan->rs_inited);
759 : : Assert(scan->rs_base.rs_parallel == NULL);
760 : :
761 : : /* When there are no pages to scan, return InvalidBlockNumber */
762 [ + + + + ]: 1090255 : if (scan->rs_nblocks == 0 || scan->rs_numblocks == 0)
763 : 524927 : return InvalidBlockNumber;
764 : :
765 [ + + ]: 565328 : if (ScanDirectionIsForward(dir))
766 : : {
767 : 565286 : return scan->rs_startblock;
768 : : }
769 : : else
770 : : {
771 : : /*
772 : : * Disable reporting to syncscan logic in a backwards scan; it's not
773 : : * very likely anyone else is doing the same thing at the same time,
774 : : * and much more likely that we'll just bollix things for forward
775 : : * scanners.
776 : : */
777 : 42 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
778 : :
779 : : /*
780 : : * Start from last page of the scan. Ensure we take into account
781 : : * rs_numblocks if it's been adjusted by heap_setscanlimits().
782 : : */
783 [ + + ]: 42 : if (scan->rs_numblocks != InvalidBlockNumber)
784 : 4 : return (scan->rs_startblock + scan->rs_numblocks - 1) % scan->rs_nblocks;
785 : :
786 [ - + ]: 38 : if (scan->rs_startblock > 0)
787 : 0 : return scan->rs_startblock - 1;
788 : :
789 : 38 : return scan->rs_nblocks - 1;
790 : : }
791 : : }
792 : :
793 : :
794 : : /*
795 : : * heapgettup_start_page - helper function for heapgettup()
796 : : *
797 : : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
798 : : * to the number of tuples on this page. Also set *lineoff to the first
799 : : * offset to scan with forward scans getting the first offset and backward
800 : : * getting the final offset on the page.
801 : : */
802 : : static Page
803 : 116943 : heapgettup_start_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
804 : : OffsetNumber *lineoff)
805 : : {
806 : : Page page;
807 : :
808 : : Assert(scan->rs_inited);
809 : : Assert(BufferIsValid(scan->rs_cbuf));
810 : :
811 : : /* Caller is responsible for ensuring buffer is locked if needed */
812 : 116943 : page = BufferGetPage(scan->rs_cbuf);
813 : :
814 : 116943 : *linesleft = PageGetMaxOffsetNumber(page) - FirstOffsetNumber + 1;
815 : :
816 [ + - ]: 116943 : if (ScanDirectionIsForward(dir))
817 : 116943 : *lineoff = FirstOffsetNumber;
818 : : else
819 : 0 : *lineoff = (OffsetNumber) (*linesleft);
820 : :
821 : : /* lineoff now references the physically previous or next tid */
822 : 116943 : return page;
823 : : }
824 : :
825 : :
826 : : /*
827 : : * heapgettup_continue_page - helper function for heapgettup()
828 : : *
829 : : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
830 : : * to the number of tuples left to scan on this page. Also set *lineoff to
831 : : * the next offset to scan according to the ScanDirection in 'dir'.
832 : : */
833 : : static inline Page
834 : 9413567 : heapgettup_continue_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
835 : : OffsetNumber *lineoff)
836 : : {
837 : : Page page;
838 : :
839 : : Assert(scan->rs_inited);
840 : : Assert(BufferIsValid(scan->rs_cbuf));
841 : :
842 : : /* Caller is responsible for ensuring buffer is locked if needed */
843 : 9413567 : page = BufferGetPage(scan->rs_cbuf);
844 : :
845 [ + - ]: 9413567 : if (ScanDirectionIsForward(dir))
846 : : {
847 : 9413567 : *lineoff = OffsetNumberNext(scan->rs_coffset);
848 : 9413567 : *linesleft = PageGetMaxOffsetNumber(page) - (*lineoff) + 1;
849 : : }
850 : : else
851 : : {
852 : : /*
853 : : * The previous returned tuple may have been vacuumed since the
854 : : * previous scan when we use a non-MVCC snapshot, so we must
855 : : * re-establish the lineoff <= PageGetMaxOffsetNumber(page) invariant
856 : : */
857 [ # # ]: 0 : *lineoff = Min(PageGetMaxOffsetNumber(page), OffsetNumberPrev(scan->rs_coffset));
858 : 0 : *linesleft = *lineoff;
859 : : }
860 : :
861 : : /* lineoff now references the physically previous or next tid */
862 : 9413567 : return page;
863 : : }
864 : :
865 : : /*
866 : : * heapgettup_advance_block - helper for heap_fetch_next_buffer()
867 : : *
868 : : * Given the current block number, the scan direction, and various information
869 : : * contained in the scan descriptor, calculate the BlockNumber to scan next
870 : : * and return it. If there are no further blocks to scan, return
871 : : * InvalidBlockNumber to indicate this fact to the caller.
872 : : *
873 : : * This should not be called to determine the initial block number -- only for
874 : : * subsequent blocks.
875 : : *
876 : : * This also adjusts rs_numblocks when a limit has been imposed by
877 : : * heap_setscanlimits().
878 : : */
879 : : static inline BlockNumber
880 : 3339707 : heapgettup_advance_block(HeapScanDesc scan, BlockNumber block, ScanDirection dir)
881 : : {
882 : : Assert(scan->rs_base.rs_parallel == NULL);
883 : :
884 [ + + ]: 3339707 : if (likely(ScanDirectionIsForward(dir)))
885 : : {
886 : 3339629 : block++;
887 : :
888 : : /* wrap back to the start of the heap */
889 [ + + ]: 3339629 : if (block >= scan->rs_nblocks)
890 : 445243 : block = 0;
891 : :
892 : : /*
893 : : * Report our new scan position for synchronization purposes. We don't
894 : : * do that when moving backwards, however. That would just mess up any
895 : : * other forward-moving scanners.
896 : : *
897 : : * Note: we do this before checking for end of scan so that the final
898 : : * state of the position hint is back at the start of the rel. That's
899 : : * not strictly necessary, but otherwise when you run the same query
900 : : * multiple times the starting position would shift a little bit
901 : : * backwards on every invocation, which is confusing. We don't
902 : : * guarantee any specific ordering in general, though.
903 : : */
904 [ + + ]: 3339629 : if (scan->rs_base.rs_flags & SO_ALLOW_SYNC)
905 : 39647 : ss_report_location(scan->rs_base.rs_rd, block);
906 : :
907 : : /* we're done if we're back at where we started */
908 [ + + ]: 3339629 : if (block == scan->rs_startblock)
909 : 445193 : return InvalidBlockNumber;
910 : :
911 : : /* check if the limit imposed by heap_setscanlimits() is met */
912 [ + + ]: 2894436 : if (scan->rs_numblocks != InvalidBlockNumber)
913 : : {
914 [ + + ]: 2818 : if (--scan->rs_numblocks == 0)
915 : 1590 : return InvalidBlockNumber;
916 : : }
917 : :
918 : 2892846 : return block;
919 : : }
920 : : else
921 : : {
922 : : /* we're done if the last block is the start position */
923 [ + - ]: 78 : if (block == scan->rs_startblock)
924 : 78 : return InvalidBlockNumber;
925 : :
926 : : /* check if the limit imposed by heap_setscanlimits() is met */
927 [ # # ]: 0 : if (scan->rs_numblocks != InvalidBlockNumber)
928 : : {
929 [ # # ]: 0 : if (--scan->rs_numblocks == 0)
930 : 0 : return InvalidBlockNumber;
931 : : }
932 : :
933 : : /* wrap to the end of the heap when the last page was page 0 */
934 [ # # ]: 0 : if (block == 0)
935 : 0 : block = scan->rs_nblocks;
936 : :
937 : 0 : block--;
938 : :
939 : 0 : return block;
940 : : }
941 : : }
942 : :
943 : : /* ----------------
944 : : * heapgettup - fetch next heap tuple
945 : : *
946 : : * Initialize the scan if not already done; then advance to the next
947 : : * tuple as indicated by "dir"; return the next tuple in scan->rs_ctup,
948 : : * or set scan->rs_ctup.t_data = NULL if no more tuples.
949 : : *
950 : : * Note: the reason nkeys/key are passed separately, even though they are
951 : : * kept in the scan descriptor, is that the caller may not want us to check
952 : : * the scankeys.
953 : : *
954 : : * Note: when we fall off the end of the scan in either direction, we
955 : : * reset rs_inited. This means that a further request with the same
956 : : * scan direction will restart the scan, which is a bit odd, but a
957 : : * request with the opposite scan direction will start a fresh scan
958 : : * in the proper direction. The latter is required behavior for cursors,
959 : : * while the former case is generally undefined behavior in Postgres
960 : : * so we don't care too much.
961 : : * ----------------
962 : : */
963 : : static void
964 : 9441101 : heapgettup(HeapScanDesc scan,
965 : : ScanDirection dir,
966 : : int nkeys,
967 : : ScanKey key)
968 : : {
969 : 9441101 : HeapTuple tuple = &(scan->rs_ctup);
970 : : Page page;
971 : : OffsetNumber lineoff;
972 : : int linesleft;
973 : :
974 [ + + ]: 9441101 : if (likely(scan->rs_inited))
975 : : {
976 : : /* continue from previously returned page/tuple */
977 : 9413567 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
978 : 9413567 : page = heapgettup_continue_page(scan, dir, &linesleft, &lineoff);
979 : 9413567 : goto continue_page;
980 : : }
981 : :
982 : : /*
983 : : * advance the scan until we find a qualifying tuple or run out of stuff
984 : : * to scan
985 : : */
986 : : while (true)
987 : : {
988 : 143587 : heap_fetch_next_buffer(scan, dir);
989 : :
990 : : /* did we run out of blocks to scan? */
991 [ + + ]: 143587 : if (!BufferIsValid(scan->rs_cbuf))
992 : 26644 : break;
993 : :
994 : : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
995 : :
996 : 116943 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
997 : 116943 : page = heapgettup_start_page(scan, dir, &linesleft, &lineoff);
998 : 9530510 : continue_page:
999 : :
1000 : : /*
1001 : : * Only continue scanning the page while we have lines left.
1002 : : *
1003 : : * Note that this protects us from accessing line pointers past
1004 : : * PageGetMaxOffsetNumber(); both for forward scans when we resume the
1005 : : * table scan, and for when we start scanning a new page.
1006 : : */
1007 [ + + ]: 9600466 : for (; linesleft > 0; linesleft--, lineoff += dir)
1008 : : {
1009 : : bool visible;
1010 : 9484413 : ItemId lpp = PageGetItemId(page, lineoff);
1011 : :
1012 [ + + ]: 9484413 : if (!ItemIdIsNormal(lpp))
1013 : 47880 : continue;
1014 : :
1015 : 9436533 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1016 : 9436533 : tuple->t_len = ItemIdGetLength(lpp);
1017 : 9436533 : ItemPointerSet(&(tuple->t_self), scan->rs_cblock, lineoff);
1018 : :
1019 : 9436533 : visible = HeapTupleSatisfiesVisibility(tuple,
1020 : : scan->rs_base.rs_snapshot,
1021 : : scan->rs_cbuf);
1022 : :
1023 : 9436533 : HeapCheckForSerializableConflictOut(visible, scan->rs_base.rs_rd,
1024 : : tuple, scan->rs_cbuf,
1025 : : scan->rs_base.rs_snapshot);
1026 : :
1027 : : /* skip tuples not visible to this snapshot */
1028 [ + + ]: 9436533 : if (!visible)
1029 : 7582 : continue;
1030 : :
1031 : : /* skip any tuples that don't match the scan key */
1032 [ + + ]: 9428951 : if (key != NULL &&
1033 [ + + ]: 15232 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1034 : : nkeys, key))
1035 : 14494 : continue;
1036 : :
1037 : 9414457 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1038 : 9414457 : scan->rs_coffset = lineoff;
1039 : 9414457 : return;
1040 : : }
1041 : :
1042 : : /*
1043 : : * if we get here, it means we've exhausted the items on this page and
1044 : : * it's time to move to the next.
1045 : : */
1046 : 116053 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
1047 : : }
1048 : :
1049 : : /* end of scan */
1050 [ - + ]: 26644 : if (BufferIsValid(scan->rs_cbuf))
1051 : 0 : ReleaseBuffer(scan->rs_cbuf);
1052 : :
1053 : 26644 : scan->rs_cbuf = InvalidBuffer;
1054 : 26644 : scan->rs_cblock = InvalidBlockNumber;
1055 : 26644 : scan->rs_prefetch_block = InvalidBlockNumber;
1056 : 26644 : tuple->t_data = NULL;
1057 : 26644 : scan->rs_inited = false;
1058 : : }
1059 : :
1060 : : /* ----------------
1061 : : * heapgettup_pagemode - fetch next heap tuple in page-at-a-time mode
1062 : : *
1063 : : * Same API as heapgettup, but used in page-at-a-time mode
1064 : : *
1065 : : * The internal logic is much the same as heapgettup's too, but there are some
1066 : : * differences: we do not take the buffer content lock (that only needs to
1067 : : * happen inside heap_prepare_pagescan), and we iterate through just the
1068 : : * tuples listed in rs_vistuples[] rather than all tuples on the page. Notice
1069 : : * that lineindex is 0-based, where the corresponding loop variable lineoff in
1070 : : * heapgettup is 1-based.
1071 : : * ----------------
1072 : : */
1073 : : static void
1074 : 64453132 : heapgettup_pagemode(HeapScanDesc scan,
1075 : : ScanDirection dir,
1076 : : int nkeys,
1077 : : ScanKey key)
1078 : : {
1079 : 64453132 : HeapTuple tuple = &(scan->rs_ctup);
1080 : : Page page;
1081 : : uint32 lineindex;
1082 : : uint32 linesleft;
1083 : :
1084 [ + + ]: 64453132 : if (likely(scan->rs_inited))
1085 : : {
1086 : : /* continue from previously returned page/tuple */
1087 : 63387629 : page = BufferGetPage(scan->rs_cbuf);
1088 : :
1089 : 63387629 : lineindex = scan->rs_cindex + dir;
1090 [ + + ]: 63387629 : if (ScanDirectionIsForward(dir))
1091 : 63387191 : linesleft = scan->rs_ntuples - lineindex;
1092 : : else
1093 : 438 : linesleft = scan->rs_cindex;
1094 : : /* lineindex now references the next or previous visible tid */
1095 : :
1096 : 63387629 : goto continue_page;
1097 : : }
1098 : :
1099 : : /*
1100 : : * advance the scan until we find a qualifying tuple or run out of stuff
1101 : : * to scan
1102 : : */
1103 : : while (true)
1104 : : {
1105 : 4216562 : heap_fetch_next_buffer(scan, dir);
1106 : :
1107 : : /* did we run out of blocks to scan? */
1108 [ + + ]: 4216533 : if (!BufferIsValid(scan->rs_cbuf))
1109 : 871461 : break;
1110 : :
1111 : : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
1112 : :
1113 : : /* prune the page and determine visible tuple offsets */
1114 : 3345072 : heap_prepare_pagescan((TableScanDesc) scan);
1115 : 3345064 : page = BufferGetPage(scan->rs_cbuf);
1116 : 3345064 : linesleft = scan->rs_ntuples;
1117 [ + + ]: 3345064 : lineindex = ScanDirectionIsForward(dir) ? 0 : linesleft - 1;
1118 : :
1119 : : /* block is the same for all tuples, set it once outside the loop */
1120 : 3345064 : ItemPointerSetBlockNumber(&tuple->t_self, scan->rs_cblock);
1121 : :
1122 : : /* lineindex now references the next or previous visible tid */
1123 : 66732693 : continue_page:
1124 : :
1125 [ + + ]: 128403162 : for (; linesleft > 0; linesleft--, lineindex += dir)
1126 : : {
1127 : : ItemId lpp;
1128 : : OffsetNumber lineoff;
1129 : :
1130 : : Assert(lineindex < scan->rs_ntuples);
1131 : 125252103 : lineoff = scan->rs_vistuples[lineindex];
1132 : 125252103 : lpp = PageGetItemId(page, lineoff);
1133 : : Assert(ItemIdIsNormal(lpp));
1134 : :
1135 : 125252103 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
1136 : 125252103 : tuple->t_len = ItemIdGetLength(lpp);
1137 : 125252103 : ItemPointerSetOffsetNumber(&tuple->t_self, lineoff);
1138 : :
1139 : : /* skip any tuples that don't match the scan key */
1140 [ + + ]: 125252103 : if (key != NULL &&
1141 [ + + ]: 62111732 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1142 : : nkeys, key))
1143 : 61670469 : continue;
1144 : :
1145 : 63581634 : scan->rs_cindex = lineindex;
1146 : 63581634 : return;
1147 : : }
1148 : : }
1149 : :
1150 : : /* end of scan */
1151 [ - + ]: 871461 : if (BufferIsValid(scan->rs_cbuf))
1152 : 0 : ReleaseBuffer(scan->rs_cbuf);
1153 : 871461 : scan->rs_cbuf = InvalidBuffer;
1154 : 871461 : scan->rs_cblock = InvalidBlockNumber;
1155 : 871461 : scan->rs_prefetch_block = InvalidBlockNumber;
1156 : 871461 : tuple->t_data = NULL;
1157 : 871461 : scan->rs_inited = false;
1158 : : }
1159 : :
1160 : :
1161 : : /* ----------------------------------------------------------------
1162 : : * heap access method interface
1163 : : * ----------------------------------------------------------------
1164 : : */
1165 : :
1166 : :
1167 : : TableScanDesc
1168 : 484150 : heap_beginscan(Relation relation, Snapshot snapshot,
1169 : : int nkeys, ScanKey key,
1170 : : ParallelTableScanDesc parallel_scan,
1171 : : uint32 flags)
1172 : : {
1173 : : HeapScanDesc scan;
1174 : :
1175 : : /*
1176 : : * increment relation ref count while scanning relation
1177 : : *
1178 : : * This is just to make really sure the relcache entry won't go away while
1179 : : * the scan has a pointer to it. Caller should be holding the rel open
1180 : : * anyway, so this is redundant in all normal scenarios...
1181 : : */
1182 : 484150 : RelationIncrementReferenceCount(relation);
1183 : :
1184 : : /*
1185 : : * allocate and initialize scan descriptor
1186 : : */
1187 [ + + ]: 484150 : if (flags & SO_TYPE_BITMAPSCAN)
1188 : : {
1189 : 13546 : BitmapHeapScanDesc bscan = palloc_object(BitmapHeapScanDescData);
1190 : :
1191 : : /*
1192 : : * Bitmap Heap scans do not have any fields that a normal Heap Scan
1193 : : * does not have, so no special initializations required here.
1194 : : */
1195 : 13546 : scan = (HeapScanDesc) bscan;
1196 : : }
1197 : : else
1198 : 470604 : scan = (HeapScanDesc) palloc_object(HeapScanDescData);
1199 : :
1200 : 484150 : scan->rs_base.rs_rd = relation;
1201 : 484150 : scan->rs_base.rs_snapshot = snapshot;
1202 : 484150 : scan->rs_base.rs_nkeys = nkeys;
1203 : 484150 : scan->rs_base.rs_flags = flags;
1204 : 484150 : scan->rs_base.rs_parallel = parallel_scan;
1205 : 484150 : scan->rs_base.rs_instrument = NULL;
1206 : 484150 : scan->rs_strategy = NULL; /* set in initscan */
1207 : 484150 : scan->rs_cbuf = InvalidBuffer;
1208 : :
1209 : : /*
1210 : : * Disable page-at-a-time mode if it's not a MVCC-safe snapshot.
1211 : : */
1212 [ + + + + ]: 484150 : if (!(snapshot && IsMVCCSnapshot(snapshot)))
1213 : 38739 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1214 : :
1215 : : /* Check that a historic snapshot is not used for non-catalog tables */
1216 [ + + ]: 484150 : if (snapshot &&
1217 [ + + ]: 472945 : IsHistoricMVCCSnapshot(snapshot) &&
1218 [ - + - - : 738 : !RelationIsAccessibleInLogicalDecoding(relation))
+ - - + -
- - - - +
- - - - -
- - - -
- ]
1219 : : {
1220 [ # # ]: 0 : ereport(ERROR,
1221 : : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
1222 : : errmsg("cannot query non-catalog table \"%s\" during logical decoding",
1223 : : RelationGetRelationName(relation))));
1224 : : }
1225 : :
1226 : : /*
1227 : : * For seqscan and sample scans in a serializable transaction, acquire a
1228 : : * predicate lock on the entire relation. This is required not only to
1229 : : * lock all the matching tuples, but also to conflict with new insertions
1230 : : * into the table. In an indexscan, we take page locks on the index pages
1231 : : * covering the range specified in the scan qual, but in a heap scan there
1232 : : * is nothing more fine-grained to lock. A bitmap scan is a different
1233 : : * story, there we have already scanned the index and locked the index
1234 : : * pages covering the predicate. But in that case we still have to lock
1235 : : * any matching heap tuples. For sample scan we could optimize the locking
1236 : : * to be at least page-level granularity, but we'd need to add per-tuple
1237 : : * locking for that.
1238 : : */
1239 [ + + ]: 484150 : if (scan->rs_base.rs_flags & (SO_TYPE_SEQSCAN | SO_TYPE_SAMPLESCAN))
1240 : : {
1241 : : /*
1242 : : * Ensure a missing snapshot is noticed reliably, even if the
1243 : : * isolation mode means predicate locking isn't performed (and
1244 : : * therefore the snapshot isn't used here).
1245 : : */
1246 : : Assert(snapshot);
1247 : 457602 : PredicateLockRelation(relation, snapshot);
1248 : : }
1249 : :
1250 : : /* we only need to set this up once */
1251 : 484150 : scan->rs_ctup.t_tableOid = RelationGetRelid(relation);
1252 : :
1253 : : /*
1254 : : * Allocate memory to keep track of page allocation for parallel workers
1255 : : * when doing a parallel scan.
1256 : : */
1257 [ + + ]: 484150 : if (parallel_scan != NULL)
1258 : 4519 : scan->rs_parallelworkerdata = palloc_object(ParallelBlockTableScanWorkerData);
1259 : : else
1260 : 479631 : scan->rs_parallelworkerdata = NULL;
1261 : :
1262 : : /*
1263 : : * we do this here instead of in initscan() because heap_rescan also calls
1264 : : * initscan() and we don't want to allocate memory again
1265 : : */
1266 [ + + ]: 484150 : if (nkeys > 0)
1267 : 269900 : scan->rs_base.rs_key = palloc_array(ScanKeyData, nkeys);
1268 : : else
1269 : 214250 : scan->rs_base.rs_key = NULL;
1270 : :
1271 : 484150 : initscan(scan, key, false);
1272 : :
1273 : 484148 : scan->rs_read_stream = NULL;
1274 : :
1275 : : /*
1276 : : * Set up a read stream for sequential scans and TID range scans. This
1277 : : * should be done after initscan() because initscan() allocates the
1278 : : * BufferAccessStrategy object passed to the read stream API.
1279 : : */
1280 [ + + ]: 484148 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN ||
1281 [ + + ]: 26642 : scan->rs_base.rs_flags & SO_TYPE_TIDRANGESCAN)
1282 : 458815 : {
1283 : : ReadStreamBlockNumberCB cb;
1284 : :
1285 [ + + ]: 458820 : if (scan->rs_base.rs_parallel)
1286 : 4519 : cb = heap_scan_stream_read_next_parallel;
1287 : : else
1288 : 454301 : cb = heap_scan_stream_read_next_serial;
1289 : :
1290 : : /* ---
1291 : : * It is safe to use batchmode as the only locks taken by `cb`
1292 : : * are never taken while waiting for IO:
1293 : : * - SyncScanLock is used in the non-parallel case
1294 : : * - in the parallel case, only spinlocks and atomics are used
1295 : : * ---
1296 : : */
1297 : 458820 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_SEQUENTIAL |
1298 : : READ_STREAM_USE_BATCHING,
1299 : : scan->rs_strategy,
1300 : : scan->rs_base.rs_rd,
1301 : : MAIN_FORKNUM,
1302 : : cb,
1303 : : scan,
1304 : : 0);
1305 : : }
1306 [ + + ]: 25328 : else if (scan->rs_base.rs_flags & SO_TYPE_BITMAPSCAN)
1307 : : {
1308 : 13546 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_DEFAULT |
1309 : : READ_STREAM_USE_BATCHING,
1310 : : scan->rs_strategy,
1311 : : scan->rs_base.rs_rd,
1312 : : MAIN_FORKNUM,
1313 : : bitmapheap_stream_read_next,
1314 : : scan,
1315 : : sizeof(TBMIterateResult));
1316 : : }
1317 : :
1318 : : /* enable read stream instrumentation */
1319 [ + + + - ]: 484143 : if ((flags & SO_SCAN_INSTRUMENT) && (scan->rs_read_stream != NULL))
1320 : : {
1321 : 8 : scan->rs_base.rs_instrument = palloc0_object(TableScanInstrumentation);
1322 : 8 : read_stream_enable_stats(scan->rs_read_stream,
1323 : 8 : &scan->rs_base.rs_instrument->io);
1324 : : }
1325 : :
1326 : 484143 : scan->rs_vmbuffer = InvalidBuffer;
1327 : :
1328 : 484143 : return (TableScanDesc) scan;
1329 : : }
1330 : :
1331 : : void
1332 : 638571 : heap_rescan(TableScanDesc sscan, ScanKey key, bool set_params,
1333 : : bool allow_strat, bool allow_sync, bool allow_pagemode)
1334 : : {
1335 : 638571 : HeapScanDesc scan = (HeapScanDesc) sscan;
1336 : :
1337 [ + + ]: 638571 : if (set_params)
1338 : : {
1339 [ + - ]: 19 : if (allow_strat)
1340 : 19 : scan->rs_base.rs_flags |= SO_ALLOW_STRAT;
1341 : : else
1342 : 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_STRAT;
1343 : :
1344 [ + + ]: 19 : if (allow_sync)
1345 : 8 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
1346 : : else
1347 : 11 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
1348 : :
1349 [ + - + - ]: 19 : if (allow_pagemode && scan->rs_base.rs_snapshot &&
1350 [ + - ]: 19 : IsMVCCSnapshot(scan->rs_base.rs_snapshot))
1351 : 19 : scan->rs_base.rs_flags |= SO_ALLOW_PAGEMODE;
1352 : : else
1353 : 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1354 : : }
1355 : :
1356 : : /*
1357 : : * unpin scan buffers
1358 : : */
1359 [ + + ]: 638571 : if (BufferIsValid(scan->rs_cbuf))
1360 : : {
1361 : 3038 : ReleaseBuffer(scan->rs_cbuf);
1362 : 3038 : scan->rs_cbuf = InvalidBuffer;
1363 : : }
1364 : :
1365 [ + + ]: 638571 : if (BufferIsValid(scan->rs_vmbuffer))
1366 : : {
1367 : 14 : ReleaseBuffer(scan->rs_vmbuffer);
1368 : 14 : scan->rs_vmbuffer = InvalidBuffer;
1369 : : }
1370 : :
1371 : : /*
1372 : : * SO_TYPE_BITMAPSCAN would be cleaned up here, but it does not hold any
1373 : : * additional data vs a normal HeapScan
1374 : : */
1375 : :
1376 : : /*
1377 : : * The read stream is reset on rescan. This must be done before
1378 : : * initscan(), as some state referred to by read_stream_reset() is reset
1379 : : * in initscan().
1380 : : */
1381 [ + + ]: 638571 : if (scan->rs_read_stream)
1382 : 638548 : read_stream_reset(scan->rs_read_stream);
1383 : :
1384 : : /*
1385 : : * reinitialize scan descriptor
1386 : : */
1387 : 638571 : initscan(scan, key, true);
1388 : 638571 : }
1389 : :
1390 : : void
1391 : 480843 : heap_endscan(TableScanDesc sscan)
1392 : : {
1393 : 480843 : HeapScanDesc scan = (HeapScanDesc) sscan;
1394 : :
1395 : : /* Note: no locking manipulations needed */
1396 : :
1397 : : /*
1398 : : * unpin scan buffers
1399 : : */
1400 [ + + ]: 480843 : if (BufferIsValid(scan->rs_cbuf))
1401 : 189614 : ReleaseBuffer(scan->rs_cbuf);
1402 : :
1403 [ + + ]: 480843 : if (BufferIsValid(scan->rs_vmbuffer))
1404 : 2705 : ReleaseBuffer(scan->rs_vmbuffer);
1405 : :
1406 : : /*
1407 : : * Must free the read stream before freeing the BufferAccessStrategy.
1408 : : */
1409 [ + + ]: 480843 : if (scan->rs_read_stream)
1410 : 469128 : read_stream_end(scan->rs_read_stream);
1411 : :
1412 : : /*
1413 : : * decrement relation reference count and free scan descriptor storage
1414 : : */
1415 : 480843 : RelationDecrementReferenceCount(scan->rs_base.rs_rd);
1416 : :
1417 [ + + ]: 480843 : if (scan->rs_base.rs_key)
1418 : 269855 : pfree(scan->rs_base.rs_key);
1419 : :
1420 [ + + ]: 480843 : if (scan->rs_strategy != NULL)
1421 : 12521 : FreeAccessStrategy(scan->rs_strategy);
1422 : :
1423 [ + + ]: 480843 : if (scan->rs_parallelworkerdata != NULL)
1424 : 4519 : pfree(scan->rs_parallelworkerdata);
1425 : :
1426 [ + + ]: 480843 : if (scan->rs_base.rs_flags & SO_TEMP_SNAPSHOT)
1427 : 51920 : UnregisterSnapshot(scan->rs_base.rs_snapshot);
1428 : :
1429 [ + + ]: 480843 : if (scan->rs_base.rs_instrument)
1430 : 8 : pfree(scan->rs_base.rs_instrument);
1431 : :
1432 : 480843 : pfree(scan);
1433 : 480843 : }
1434 : :
1435 : : HeapTuple
1436 : 12518180 : heap_getnext(TableScanDesc sscan, ScanDirection direction)
1437 : : {
1438 : 12518180 : HeapScanDesc scan = (HeapScanDesc) sscan;
1439 : :
1440 : : /*
1441 : : * This is still widely used directly, without going through table AM, so
1442 : : * add a safety check. It's possible we should, at a later point,
1443 : : * downgrade this to an assert. The reason for checking the AM routine,
1444 : : * rather than the AM oid, is that this allows to write regression tests
1445 : : * that create another AM reusing the heap handler.
1446 : : */
1447 [ - + ]: 12518180 : if (unlikely(sscan->rs_rd->rd_tableam != GetHeapamTableAmRoutine()))
1448 [ # # ]: 0 : ereport(ERROR,
1449 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1450 : : errmsg_internal("only heap AM is supported")));
1451 : :
1452 : : /* Note: no locking manipulations needed */
1453 : :
1454 [ + + ]: 12518180 : if (scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
1455 : 3661917 : heapgettup_pagemode(scan, direction,
1456 : 3661917 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1457 : : else
1458 : 8856263 : heapgettup(scan, direction,
1459 : 8856263 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1460 : :
1461 [ + + ]: 12518180 : if (scan->rs_ctup.t_data == NULL)
1462 : 83769 : return NULL;
1463 : :
1464 : : /*
1465 : : * if we get here it means we have a new current scan tuple, so point to
1466 : : * the proper return buffer and return the tuple.
1467 : : */
1468 : :
1469 [ - + - - : 12434411 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
+ - ]
1470 : :
1471 : 12434411 : return &scan->rs_ctup;
1472 : : }
1473 : :
1474 : : bool
1475 : 61369417 : heap_getnextslot(TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
1476 : : {
1477 : 61369417 : HeapScanDesc scan = (HeapScanDesc) sscan;
1478 : :
1479 : : /* Note: no locking manipulations needed */
1480 : :
1481 [ + + ]: 61369417 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1482 : 60784579 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1483 : : else
1484 : 584838 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1485 : :
1486 [ + + ]: 61369390 : if (scan->rs_ctup.t_data == NULL)
1487 : : {
1488 : 814198 : ExecClearTuple(slot);
1489 : 814198 : return false;
1490 : : }
1491 : :
1492 : : /*
1493 : : * if we get here it means we have a new current scan tuple, so point to
1494 : : * the proper return buffer and return the tuple.
1495 : : */
1496 : :
1497 [ + + - + : 60555192 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
+ + ]
1498 : :
1499 : 60555192 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
1500 : : scan->rs_cbuf);
1501 : 60555192 : return true;
1502 : : }
1503 : :
1504 : : void
1505 : 1374 : heap_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
1506 : : ItemPointer maxtid)
1507 : : {
1508 : 1374 : HeapScanDesc scan = (HeapScanDesc) sscan;
1509 : : BlockNumber startBlk;
1510 : : BlockNumber numBlks;
1511 : : ItemPointerData highestItem;
1512 : : ItemPointerData lowestItem;
1513 : :
1514 : : /*
1515 : : * For relations without any pages, we can simply leave the TID range
1516 : : * unset. There will be no tuples to scan, therefore no tuples outside
1517 : : * the given TID range.
1518 : : */
1519 [ + + ]: 1374 : if (scan->rs_nblocks == 0)
1520 : 32 : return;
1521 : :
1522 : : /*
1523 : : * Set up some ItemPointers which point to the first and last possible
1524 : : * tuples in the heap.
1525 : : */
1526 : 1366 : ItemPointerSet(&highestItem, scan->rs_nblocks - 1, MaxOffsetNumber);
1527 : 1366 : ItemPointerSet(&lowestItem, 0, FirstOffsetNumber);
1528 : :
1529 : : /*
1530 : : * If the given maximum TID is below the highest possible TID in the
1531 : : * relation, then restrict the range to that, otherwise we scan to the end
1532 : : * of the relation.
1533 : : */
1534 [ + + ]: 1366 : if (ItemPointerCompare(maxtid, &highestItem) < 0)
1535 : 172 : ItemPointerCopy(maxtid, &highestItem);
1536 : :
1537 : : /*
1538 : : * If the given minimum TID is above the lowest possible TID in the
1539 : : * relation, then restrict the range to only scan for TIDs above that.
1540 : : */
1541 [ + + ]: 1366 : if (ItemPointerCompare(mintid, &lowestItem) > 0)
1542 : 1210 : ItemPointerCopy(mintid, &lowestItem);
1543 : :
1544 : : /*
1545 : : * Check for an empty range and protect from would be negative results
1546 : : * from the numBlks calculation below.
1547 : : */
1548 [ + + ]: 1366 : if (ItemPointerCompare(&highestItem, &lowestItem) < 0)
1549 : : {
1550 : : /* Set an empty range of blocks to scan */
1551 : 24 : heap_setscanlimits(sscan, 0, 0);
1552 : 24 : return;
1553 : : }
1554 : :
1555 : : /*
1556 : : * Calculate the first block and the number of blocks we must scan. We
1557 : : * could be more aggressive here and perform some more validation to try
1558 : : * and further narrow the scope of blocks to scan by checking if the
1559 : : * lowestItem has an offset above MaxOffsetNumber. In this case, we could
1560 : : * advance startBlk by one. Likewise, if highestItem has an offset of 0
1561 : : * we could scan one fewer blocks. However, such an optimization does not
1562 : : * seem worth troubling over, currently.
1563 : : */
1564 : 1342 : startBlk = ItemPointerGetBlockNumberNoCheck(&lowestItem);
1565 : :
1566 : 1342 : numBlks = ItemPointerGetBlockNumberNoCheck(&highestItem) -
1567 : 1342 : ItemPointerGetBlockNumberNoCheck(&lowestItem) + 1;
1568 : :
1569 : : /* Set the start block and number of blocks to scan */
1570 : 1342 : heap_setscanlimits(sscan, startBlk, numBlks);
1571 : :
1572 : : /* Finally, set the TID range in sscan */
1573 : 1342 : ItemPointerCopy(&lowestItem, &sscan->st.tidrange.rs_mintid);
1574 : 1342 : ItemPointerCopy(&highestItem, &sscan->st.tidrange.rs_maxtid);
1575 : : }
1576 : :
1577 : : bool
1578 : 6512 : heap_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
1579 : : TupleTableSlot *slot)
1580 : : {
1581 : 6512 : HeapScanDesc scan = (HeapScanDesc) sscan;
1582 : 6512 : ItemPointer mintid = &sscan->st.tidrange.rs_mintid;
1583 : 6512 : ItemPointer maxtid = &sscan->st.tidrange.rs_maxtid;
1584 : :
1585 : : /* Note: no locking manipulations needed */
1586 : : for (;;)
1587 : : {
1588 [ + - ]: 6636 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1589 : 6636 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1590 : : else
1591 : 0 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1592 : :
1593 [ + + ]: 6626 : if (scan->rs_ctup.t_data == NULL)
1594 : : {
1595 : 138 : ExecClearTuple(slot);
1596 : 138 : return false;
1597 : : }
1598 : :
1599 : : /*
1600 : : * heap_set_tidrange will have used heap_setscanlimits to limit the
1601 : : * range of pages we scan to only ones that can contain the TID range
1602 : : * we're scanning for. Here we must filter out any tuples from these
1603 : : * pages that are outside of that range.
1604 : : */
1605 [ + + ]: 6488 : if (ItemPointerCompare(&scan->rs_ctup.t_self, mintid) < 0)
1606 : : {
1607 : 124 : ExecClearTuple(slot);
1608 : :
1609 : : /*
1610 : : * When scanning backwards, the TIDs will be in descending order.
1611 : : * Future tuples in this direction will be lower still, so we can
1612 : : * just return false to indicate there will be no more tuples.
1613 : : */
1614 [ - + ]: 124 : if (ScanDirectionIsBackward(direction))
1615 : 0 : return false;
1616 : :
1617 : 124 : continue;
1618 : : }
1619 : :
1620 : : /*
1621 : : * Likewise for the final page, we must filter out TIDs greater than
1622 : : * maxtid.
1623 : : */
1624 [ + + ]: 6364 : if (ItemPointerCompare(&scan->rs_ctup.t_self, maxtid) > 0)
1625 : : {
1626 : 74 : ExecClearTuple(slot);
1627 : :
1628 : : /*
1629 : : * When scanning forward, the TIDs will be in ascending order.
1630 : : * Future tuples in this direction will be higher still, so we can
1631 : : * just return false to indicate there will be no more tuples.
1632 : : */
1633 [ + - ]: 74 : if (ScanDirectionIsForward(direction))
1634 : 74 : return false;
1635 : 0 : continue;
1636 : : }
1637 : :
1638 : 6290 : break;
1639 : : }
1640 : :
1641 : : /*
1642 : : * if we get here it means we have a new current scan tuple, so point to
1643 : : * the proper return buffer and return the tuple.
1644 : : */
1645 [ - + - - : 6290 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
+ - ]
1646 : :
1647 : 6290 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
1648 : 6290 : return true;
1649 : : }
1650 : :
1651 : : /*
1652 : : * heap_fetch - retrieve tuple with given tid
1653 : : *
1654 : : * On entry, tuple->t_self is the TID to fetch. We pin the buffer holding
1655 : : * the tuple, fill in the remaining fields of *tuple, and check the tuple
1656 : : * against the specified snapshot.
1657 : : *
1658 : : * If successful (tuple found and passes snapshot time qual), then *userbuf
1659 : : * is set to the buffer holding the tuple and true is returned. The caller
1660 : : * must unpin the buffer when done with the tuple.
1661 : : *
1662 : : * If the tuple is not found (ie, item number references a deleted slot),
1663 : : * then tuple->t_data is set to NULL, *userbuf is set to InvalidBuffer,
1664 : : * and false is returned.
1665 : : *
1666 : : * If the tuple is found but fails the time qual check, then the behavior
1667 : : * depends on the keep_buf parameter. If keep_buf is false, the results
1668 : : * are the same as for the tuple-not-found case. If keep_buf is true,
1669 : : * then tuple->t_data and *userbuf are returned as for the success case,
1670 : : * and again the caller must unpin the buffer; but false is returned.
1671 : : *
1672 : : * heap_fetch does not follow HOT chains: only the exact TID requested will
1673 : : * be fetched.
1674 : : *
1675 : : * It is somewhat inconsistent that we ereport() on invalid block number but
1676 : : * return false on invalid item number. There are a couple of reasons though.
1677 : : * One is that the caller can relatively easily check the block number for
1678 : : * validity, but cannot check the item number without reading the page
1679 : : * himself. Another is that when we are following a t_ctid link, we can be
1680 : : * reasonably confident that the page number is valid (since VACUUM shouldn't
1681 : : * truncate off the destination page without having killed the referencing
1682 : : * tuple first), but the item number might well not be good.
1683 : : */
1684 : : bool
1685 : 2841910 : heap_fetch(Relation relation,
1686 : : Snapshot snapshot,
1687 : : HeapTuple tuple,
1688 : : Buffer *userbuf,
1689 : : bool keep_buf)
1690 : : {
1691 : 2841910 : ItemPointer tid = &(tuple->t_self);
1692 : : ItemId lp;
1693 : : Buffer buffer;
1694 : : Page page;
1695 : : OffsetNumber offnum;
1696 : : bool valid;
1697 : :
1698 : : /*
1699 : : * Fetch and pin the appropriate page of the relation.
1700 : : */
1701 : 2841910 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
1702 : :
1703 : : /*
1704 : : * Need share lock on buffer to examine tuple commit status.
1705 : : */
1706 : 2841902 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1707 : 2841902 : page = BufferGetPage(buffer);
1708 : :
1709 : : /*
1710 : : * We'd better check for out-of-range offnum in case of VACUUM since the
1711 : : * TID was obtained.
1712 : : */
1713 : 2841902 : offnum = ItemPointerGetOffsetNumber(tid);
1714 [ + - + + ]: 2841902 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1715 : : {
1716 : 4 : UnlockReleaseBuffer(buffer);
1717 : 4 : *userbuf = InvalidBuffer;
1718 : 4 : tuple->t_data = NULL;
1719 : 4 : return false;
1720 : : }
1721 : :
1722 : : /*
1723 : : * get the item line pointer corresponding to the requested tid
1724 : : */
1725 : 2841898 : lp = PageGetItemId(page, offnum);
1726 : :
1727 : : /*
1728 : : * Must check for deleted tuple.
1729 : : */
1730 [ + + ]: 2841898 : if (!ItemIdIsNormal(lp))
1731 : : {
1732 : 293 : UnlockReleaseBuffer(buffer);
1733 : 293 : *userbuf = InvalidBuffer;
1734 : 293 : tuple->t_data = NULL;
1735 : 293 : return false;
1736 : : }
1737 : :
1738 : : /*
1739 : : * fill in *tuple fields
1740 : : */
1741 : 2841605 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
1742 : 2841605 : tuple->t_len = ItemIdGetLength(lp);
1743 : 2841605 : tuple->t_tableOid = RelationGetRelid(relation);
1744 : :
1745 : : /*
1746 : : * check tuple visibility, then release lock
1747 : : */
1748 : 2841605 : valid = HeapTupleSatisfiesVisibility(tuple, snapshot, buffer);
1749 : :
1750 [ + + ]: 2841605 : if (valid)
1751 : 2841543 : PredicateLockTID(relation, &(tuple->t_self), snapshot,
1752 : 2841543 : HeapTupleHeaderGetXmin(tuple->t_data));
1753 : :
1754 : 2841605 : HeapCheckForSerializableConflictOut(valid, relation, tuple, buffer, snapshot);
1755 : :
1756 : 2841605 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1757 : :
1758 [ + + ]: 2841605 : if (valid)
1759 : : {
1760 : : /*
1761 : : * All checks passed, so return the tuple as valid. Caller is now
1762 : : * responsible for releasing the buffer.
1763 : : */
1764 : 2841543 : *userbuf = buffer;
1765 : :
1766 : 2841543 : return true;
1767 : : }
1768 : :
1769 : : /* Tuple failed time qual, but maybe caller wants to see it anyway. */
1770 [ + + ]: 62 : if (keep_buf)
1771 : 38 : *userbuf = buffer;
1772 : : else
1773 : : {
1774 : 24 : ReleaseBuffer(buffer);
1775 : 24 : *userbuf = InvalidBuffer;
1776 : 24 : tuple->t_data = NULL;
1777 : : }
1778 : :
1779 : 62 : return false;
1780 : : }
1781 : :
1782 : : /*
1783 : : * heap_get_latest_tid - get the latest tid of a specified tuple
1784 : : *
1785 : : * Actually, this gets the latest version that is visible according to the
1786 : : * scan's snapshot. Create a scan using SnapshotDirty to get the very latest,
1787 : : * possibly uncommitted version.
1788 : : *
1789 : : * *tid is both an input and an output parameter: it is updated to
1790 : : * show the latest version of the row. Note that it will not be changed
1791 : : * if no version of the row passes the snapshot test.
1792 : : */
1793 : : void
1794 : 207 : heap_get_latest_tid(TableScanDesc sscan,
1795 : : ItemPointer tid)
1796 : : {
1797 : 207 : Relation relation = sscan->rs_rd;
1798 : 207 : Snapshot snapshot = sscan->rs_snapshot;
1799 : : ItemPointerData ctid;
1800 : : TransactionId priorXmax;
1801 : :
1802 : : /*
1803 : : * table_tuple_get_latest_tid() verified that the passed in tid is valid.
1804 : : * Assume that t_ctid links are valid however - there shouldn't be invalid
1805 : : * ones in the table.
1806 : : */
1807 : : Assert(ItemPointerIsValid(tid));
1808 : :
1809 : : /*
1810 : : * Loop to chase down t_ctid links. At top of loop, ctid is the tuple we
1811 : : * need to examine, and *tid is the TID we will return if ctid turns out
1812 : : * to be bogus.
1813 : : *
1814 : : * Note that we will loop until we reach the end of the t_ctid chain.
1815 : : * Depending on the snapshot passed, there might be at most one visible
1816 : : * version of the row, but we don't try to optimize for that.
1817 : : */
1818 : 207 : ctid = *tid;
1819 : 207 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
1820 : : for (;;)
1821 : 64 : {
1822 : : Buffer buffer;
1823 : : Page page;
1824 : : OffsetNumber offnum;
1825 : : ItemId lp;
1826 : : HeapTupleData tp;
1827 : : bool valid;
1828 : :
1829 : : /*
1830 : : * Read, pin, and lock the page.
1831 : : */
1832 : 271 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&ctid));
1833 : 271 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
1834 : 271 : page = BufferGetPage(buffer);
1835 : :
1836 : : /*
1837 : : * Check for bogus item number. This is not treated as an error
1838 : : * condition because it can happen while following a t_ctid link. We
1839 : : * just assume that the prior tid is OK and return it unchanged.
1840 : : */
1841 : 271 : offnum = ItemPointerGetOffsetNumber(&ctid);
1842 [ + - - + ]: 271 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1843 : : {
1844 : 0 : UnlockReleaseBuffer(buffer);
1845 : 0 : break;
1846 : : }
1847 : 271 : lp = PageGetItemId(page, offnum);
1848 [ - + ]: 271 : if (!ItemIdIsNormal(lp))
1849 : : {
1850 : 0 : UnlockReleaseBuffer(buffer);
1851 : 0 : break;
1852 : : }
1853 : :
1854 : : /* OK to access the tuple */
1855 : 271 : tp.t_self = ctid;
1856 : 271 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
1857 : 271 : tp.t_len = ItemIdGetLength(lp);
1858 : 271 : tp.t_tableOid = RelationGetRelid(relation);
1859 : :
1860 : : /*
1861 : : * After following a t_ctid link, we might arrive at an unrelated
1862 : : * tuple. Check for XMIN match.
1863 : : */
1864 [ + + - + ]: 335 : if (TransactionIdIsValid(priorXmax) &&
1865 : 64 : !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(tp.t_data)))
1866 : : {
1867 : 0 : UnlockReleaseBuffer(buffer);
1868 : 0 : break;
1869 : : }
1870 : :
1871 : : /*
1872 : : * Check tuple visibility; if visible, set it as the new result
1873 : : * candidate.
1874 : : */
1875 : 271 : valid = HeapTupleSatisfiesVisibility(&tp, snapshot, buffer);
1876 : 271 : HeapCheckForSerializableConflictOut(valid, relation, &tp, buffer, snapshot);
1877 [ + + ]: 271 : if (valid)
1878 : 191 : *tid = ctid;
1879 : :
1880 : : /*
1881 : : * If there's a valid t_ctid link, follow it, else we're done.
1882 : : */
1883 [ + + + + ]: 391 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
1884 [ + - ]: 200 : HeapTupleHeaderIsOnlyLocked(tp.t_data) ||
1885 [ + + ]: 160 : HeapTupleHeaderIndicatesMovedPartitions(tp.t_data) ||
1886 : 80 : ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
1887 : : {
1888 : 207 : UnlockReleaseBuffer(buffer);
1889 : 207 : break;
1890 : : }
1891 : :
1892 : 64 : ctid = tp.t_data->t_ctid;
1893 : 64 : priorXmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
1894 : 64 : UnlockReleaseBuffer(buffer);
1895 : : } /* end of loop */
1896 : 207 : }
1897 : :
1898 : :
1899 : : /*
1900 : : * UpdateXmaxHintBits - update tuple hint bits after xmax transaction ends
1901 : : *
1902 : : * This is called after we have waited for the XMAX transaction to terminate.
1903 : : * If the transaction aborted, we guarantee the XMAX_INVALID hint bit will
1904 : : * be set on exit. If the transaction committed, we set the XMAX_COMMITTED
1905 : : * hint bit if possible --- but beware that that may not yet be possible,
1906 : : * if the transaction committed asynchronously.
1907 : : *
1908 : : * Note that if the transaction was a locker only, we set HEAP_XMAX_INVALID
1909 : : * even if it commits.
1910 : : *
1911 : : * Hence callers should look only at XMAX_INVALID.
1912 : : *
1913 : : * Note this is not allowed for tuples whose xmax is a multixact.
1914 : : */
1915 : : static void
1916 : 281 : UpdateXmaxHintBits(HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
1917 : : {
1918 : : Assert(TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple), xid));
1919 : : Assert(!(tuple->t_infomask & HEAP_XMAX_IS_MULTI));
1920 : :
1921 [ + + ]: 281 : if (!(tuple->t_infomask & (HEAP_XMAX_COMMITTED | HEAP_XMAX_INVALID)))
1922 : : {
1923 [ + + + + ]: 510 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask) &&
1924 : 230 : TransactionIdDidCommit(xid))
1925 : 203 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_COMMITTED,
1926 : : xid);
1927 : : else
1928 : 77 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_INVALID,
1929 : : InvalidTransactionId);
1930 : : }
1931 : 281 : }
1932 : :
1933 : :
1934 : : /*
1935 : : * GetBulkInsertState - prepare status object for a bulk insert
1936 : : */
1937 : : BulkInsertState
1938 : 3606 : GetBulkInsertState(void)
1939 : : {
1940 : : BulkInsertState bistate;
1941 : :
1942 : 3606 : bistate = (BulkInsertState) palloc_object(BulkInsertStateData);
1943 : 3606 : bistate->strategy = GetAccessStrategy(BAS_BULKWRITE);
1944 : 3606 : bistate->current_buf = InvalidBuffer;
1945 : 3606 : bistate->next_free = InvalidBlockNumber;
1946 : 3606 : bistate->last_free = InvalidBlockNumber;
1947 : 3606 : bistate->already_extended_by = 0;
1948 : 3606 : return bistate;
1949 : : }
1950 : :
1951 : : /*
1952 : : * FreeBulkInsertState - clean up after finishing a bulk insert
1953 : : */
1954 : : void
1955 : 3367 : FreeBulkInsertState(BulkInsertState bistate)
1956 : : {
1957 [ + + ]: 3367 : if (bistate->current_buf != InvalidBuffer)
1958 : 2645 : ReleaseBuffer(bistate->current_buf);
1959 : 3367 : FreeAccessStrategy(bistate->strategy);
1960 : 3367 : pfree(bistate);
1961 : 3367 : }
1962 : :
1963 : : /*
1964 : : * ReleaseBulkInsertStatePin - release a buffer currently held in bistate
1965 : : */
1966 : : void
1967 : 90779 : ReleaseBulkInsertStatePin(BulkInsertState bistate)
1968 : : {
1969 [ + + ]: 90779 : if (bistate->current_buf != InvalidBuffer)
1970 : 40028 : ReleaseBuffer(bistate->current_buf);
1971 : 90779 : bistate->current_buf = InvalidBuffer;
1972 : :
1973 : : /*
1974 : : * Despite the name, we also reset bulk relation extension state.
1975 : : * Otherwise we can end up erroring out due to looking for free space in
1976 : : * ->next_free of one partition, even though ->next_free was set when
1977 : : * extending another partition. It could obviously also be bad for
1978 : : * efficiency to look at existing blocks at offsets from another
1979 : : * partition, even if we don't error out.
1980 : : */
1981 : 90779 : bistate->next_free = InvalidBlockNumber;
1982 : 90779 : bistate->last_free = InvalidBlockNumber;
1983 : 90779 : }
1984 : :
1985 : :
1986 : : /*
1987 : : * heap_insert - insert tuple into a heap
1988 : : *
1989 : : * The new tuple is stamped with current transaction ID and the specified
1990 : : * command ID.
1991 : : *
1992 : : * See table_tuple_insert for comments about most of the input flags, except
1993 : : * that this routine directly takes a tuple rather than a slot.
1994 : : *
1995 : : * There's corresponding HEAP_INSERT_ options to all the TABLE_INSERT_
1996 : : * options, and there additionally is HEAP_INSERT_SPECULATIVE which is used to
1997 : : * implement table_tuple_insert_speculative().
1998 : : *
1999 : : * On return the header fields of *tup are updated to match the stored tuple;
2000 : : * in particular tup->t_self receives the actual TID where the tuple was
2001 : : * stored. But note that any toasting of fields within the tuple data is NOT
2002 : : * reflected into *tup.
2003 : : */
2004 : : void
2005 : 11864276 : heap_insert(Relation relation, HeapTuple tup, CommandId cid,
2006 : : uint32 options, BulkInsertState bistate)
2007 : : {
2008 : 11864276 : TransactionId xid = GetCurrentTransactionId();
2009 : : HeapTuple heaptup;
2010 : : Buffer buffer;
2011 : : Page page;
2012 : 11864270 : Buffer vmbuffer = InvalidBuffer;
2013 : 11864270 : bool clear_all_visible = false;
2014 : 11864270 : bool vmbuffer_modified = false;
2015 : :
2016 : : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
2017 : : Assert(HeapTupleHeaderGetNatts(tup->t_data) <=
2018 : : RelationGetNumberOfAttributes(relation));
2019 : :
2020 : 11864270 : AssertHasSnapshotForToast(relation);
2021 : :
2022 : : /*
2023 : : * Fill in tuple header fields and toast the tuple if necessary.
2024 : : *
2025 : : * Note: below this point, heaptup is the data we actually intend to store
2026 : : * into the relation; tup is the caller's original untoasted data.
2027 : : */
2028 : 11864270 : heaptup = heap_prepare_insert(relation, tup, xid, cid, options);
2029 : :
2030 : : /*
2031 : : * Find buffer to insert this tuple into. If the page is all visible,
2032 : : * this will also pin the requisite visibility map page.
2033 : : */
2034 : 11864270 : buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
2035 : : InvalidBuffer, options, bistate,
2036 : : &vmbuffer, NULL,
2037 : : 0);
2038 : :
2039 : 11864268 : page = BufferGetPage(buffer);
2040 : :
2041 : : /*
2042 : : * We're about to do the actual insert -- but check for conflict first, to
2043 : : * avoid possibly having to roll back work we've just done.
2044 : : *
2045 : : * This is safe without a recheck as long as there is no possibility of
2046 : : * another process scanning the page between this check and the insert
2047 : : * being visible to the scan (i.e., an exclusive buffer content lock is
2048 : : * continuously held from this point until the tuple insert is visible).
2049 : : *
2050 : : * For a heap insert, we only need to check for table-level SSI locks. Our
2051 : : * new tuple can't possibly conflict with existing tuple locks, and heap
2052 : : * page locks are only consolidated versions of tuple locks; they do not
2053 : : * lock "gaps" as index page locks do. So we don't need to specify a
2054 : : * buffer when making the call, which makes for a faster check.
2055 : : */
2056 : 11864268 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2057 : :
2058 : : /* Lock the vmbuffer before the critical section */
2059 [ + + ]: 11864254 : if (PageIsAllVisible(page))
2060 : : {
2061 : 7433 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
2062 : 7433 : clear_all_visible = true;
2063 : : }
2064 : :
2065 : : /* NO EREPORT(ERROR) from here till changes are logged */
2066 : 11864254 : START_CRIT_SECTION();
2067 : :
2068 : 11864254 : RelationPutHeapTuple(relation, buffer, heaptup,
2069 : 11864254 : (options & HEAP_INSERT_SPECULATIVE) != 0);
2070 : :
2071 [ + + ]: 11864254 : if (clear_all_visible)
2072 : : {
2073 : : /* It's possible the VM bits were already clear */
2074 [ + + ]: 7433 : if (visibilitymap_clear(relation->rd_locator,
2075 : 7433 : ItemPointerGetBlockNumber(&(heaptup->t_self)),
2076 : : vmbuffer, VISIBILITYMAP_VALID_BITS))
2077 : 7431 : vmbuffer_modified = true;
2078 : :
2079 : 7433 : PageClearAllVisible(page);
2080 : : }
2081 : :
2082 : : /*
2083 : : * Set pd_prune_xid to trigger heap_page_prune_and_freeze() once the page
2084 : : * is full so that we can set the page all-visible in the VM on the next
2085 : : * page access.
2086 : : *
2087 : : * Setting pd_prune_xid is also handy if the inserting transaction
2088 : : * eventually aborts making this tuple DEAD and hence available for
2089 : : * pruning. If no other tuple in this page is UPDATEd/DELETEd, the aborted
2090 : : * tuple would never otherwise be pruned until next vacuum is triggered.
2091 : : *
2092 : : * Don't set it if we are in bootstrap mode or we are inserting a frozen
2093 : : * tuple, as there is no further pruning/freezing needed in those cases.
2094 : : */
2095 [ + + + + ]: 11864254 : if (TransactionIdIsNormal(xid) && !(options & HEAP_INSERT_FROZEN))
2096 [ + + + + ]: 11210911 : PageSetPrunable(page, xid);
2097 : :
2098 : 11864254 : MarkBufferDirty(buffer);
2099 : :
2100 : : /* XLOG stuff */
2101 [ + + + + : 11864254 : if (RelationNeedsWAL(relation))
+ + + + ]
2102 : : {
2103 : : xl_heap_insert xlrec;
2104 : : xl_heap_header xlhdr;
2105 : : XLogRecPtr recptr;
2106 : 10739883 : uint8 info = XLOG_HEAP_INSERT;
2107 : 10739883 : int bufflags = 0;
2108 : :
2109 : : /*
2110 : : * If this is a catalog, we need to transmit combo CIDs to properly
2111 : : * decode, so log that as well.
2112 : : */
2113 [ + + + + : 10739883 : if (RelationIsAccessibleInLogicalDecoding(relation))
+ - - + -
- - - + +
+ + - + -
- + + ]
2114 : 3713 : log_heap_new_cid(relation, heaptup);
2115 : :
2116 : : /*
2117 : : * If this is the single and first tuple on page, we can reinit the
2118 : : * page instead of restoring the whole thing. Set flag, and hide
2119 : : * buffer references from XLogInsert.
2120 : : */
2121 [ + + + + ]: 10864023 : if (ItemPointerGetOffsetNumber(&(heaptup->t_self)) == FirstOffsetNumber &&
2122 : 124140 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
2123 : : {
2124 : 123223 : info |= XLOG_HEAP_INIT_PAGE;
2125 : 123223 : bufflags |= REGBUF_WILL_INIT;
2126 : : }
2127 : :
2128 : 10739883 : xlrec.offnum = ItemPointerGetOffsetNumber(&heaptup->t_self);
2129 : 10739883 : xlrec.flags = 0;
2130 [ + + ]: 10739883 : if (clear_all_visible)
2131 : 7429 : xlrec.flags |= XLH_INSERT_ALL_VISIBLE_CLEARED;
2132 [ + + ]: 10739883 : if (options & HEAP_INSERT_SPECULATIVE)
2133 : 2207 : xlrec.flags |= XLH_INSERT_IS_SPECULATIVE;
2134 : : Assert(ItemPointerGetBlockNumber(&heaptup->t_self) == BufferGetBlockNumber(buffer));
2135 : :
2136 : : /*
2137 : : * For logical decoding, we need the tuple even if we're doing a full
2138 : : * page write, so make sure it's included even if we take a full-page
2139 : : * image. (XXX We could alternatively store a pointer into the FPW).
2140 : : */
2141 [ + + + + : 10739883 : if (RelationIsLogicallyLogged(relation) &&
+ - - + -
- - - + -
+ + ]
2142 [ + + ]: 276039 : !(options & HEAP_INSERT_NO_LOGICAL))
2143 : : {
2144 : 275920 : xlrec.flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2145 : 275920 : bufflags |= REGBUF_KEEP_DATA;
2146 : :
2147 [ + + ]: 275920 : if (IsToastRelation(relation))
2148 : 1855 : xlrec.flags |= XLH_INSERT_ON_TOAST_RELATION;
2149 : : }
2150 : :
2151 : 10739883 : XLogBeginInsert();
2152 : 10739883 : XLogRegisterData(&xlrec, SizeOfHeapInsert);
2153 : :
2154 : 10739883 : xlhdr.t_infomask2 = heaptup->t_data->t_infomask2;
2155 : 10739883 : xlhdr.t_infomask = heaptup->t_data->t_infomask;
2156 : 10739883 : xlhdr.t_hoff = heaptup->t_data->t_hoff;
2157 : :
2158 : : /*
2159 : : * note we mark xlhdr as belonging to buffer; if XLogInsert decides to
2160 : : * write the whole page to the xlog, we don't need to store
2161 : : * xl_heap_header in the xlog.
2162 : : */
2163 : 10739883 : XLogRegisterBuffer(HEAP_INSERT_BLKREF_HEAP, buffer,
2164 : 10739883 : REGBUF_STANDARD | bufflags);
2165 : 10739883 : XLogRegisterBufData(HEAP_INSERT_BLKREF_HEAP, &xlhdr,
2166 : : SizeOfHeapHeader);
2167 : : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
2168 : 10739883 : XLogRegisterBufData(HEAP_INSERT_BLKREF_HEAP,
2169 : 10739883 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2170 : 10739883 : heaptup->t_len - SizeofHeapTupleHeader);
2171 : :
2172 : : /* filtering by origin on a row level is much more efficient */
2173 : 10739883 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2174 : :
2175 [ + + ]: 10739883 : if (vmbuffer_modified)
2176 : 7427 : XLogRegisterBuffer(HEAP_INSERT_BLKREF_VM, vmbuffer, 0);
2177 : :
2178 : 10739883 : recptr = XLogInsert(RM_HEAP_ID, info);
2179 : :
2180 : 10739883 : PageSetLSN(page, recptr);
2181 : :
2182 [ + + ]: 10739883 : if (vmbuffer_modified)
2183 : 7427 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
2184 : : }
2185 : :
2186 : 11864254 : END_CRIT_SECTION();
2187 : :
2188 : 11864254 : UnlockReleaseBuffer(buffer);
2189 : :
2190 : : /*
2191 : : * We locked vmbuffer if clear_all_visible was true regardless of whether
2192 : : * or not we ended up modifying the vmbuffer.
2193 : : */
2194 [ + + ]: 11864254 : if (clear_all_visible)
2195 : 7433 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
2196 [ + + ]: 11864254 : if (BufferIsValid(vmbuffer))
2197 : 7809 : ReleaseBuffer(vmbuffer);
2198 : :
2199 : : /*
2200 : : * If tuple is cacheable, mark it for invalidation from the caches in case
2201 : : * we abort. Note it is OK to do this after releasing the buffer, because
2202 : : * the heaptup data structure is all in local memory, not in the shared
2203 : : * buffer.
2204 : : */
2205 : 11864254 : CacheInvalidateHeapTuple(relation, heaptup, NULL);
2206 : :
2207 : : /* Note: speculative insertions are counted too, even if aborted later */
2208 : 11864254 : pgstat_count_heap_insert(relation, 1);
2209 : :
2210 : : /*
2211 : : * If heaptup is a private copy, release it. Don't forget to copy t_self
2212 : : * back to the caller's image, too.
2213 : : */
2214 [ + + ]: 11864254 : if (heaptup != tup)
2215 : : {
2216 : 22650 : tup->t_self = heaptup->t_self;
2217 : 22650 : heap_freetuple(heaptup);
2218 : : }
2219 : 11864254 : }
2220 : :
2221 : : /*
2222 : : * Subroutine for heap_insert(). Prepares a tuple for insertion. This sets the
2223 : : * tuple header fields and toasts the tuple if necessary. Returns a toasted
2224 : : * version of the tuple if it was toasted, or the original tuple if not. Note
2225 : : * that in any case, the header fields are also set in the original tuple.
2226 : : */
2227 : : static HeapTuple
2228 : 13711033 : heap_prepare_insert(Relation relation, HeapTuple tup, TransactionId xid,
2229 : : CommandId cid, uint32 options)
2230 : : {
2231 : : /*
2232 : : * To allow parallel inserts, we need to ensure that they are safe to be
2233 : : * performed in workers. We have the infrastructure to allow parallel
2234 : : * inserts in general except for the cases where inserts generate a new
2235 : : * CommandId (eg. inserts into a table having a foreign key column).
2236 : : */
2237 [ - + ]: 13711033 : if (IsParallelWorker())
2238 [ # # ]: 0 : ereport(ERROR,
2239 : : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2240 : : errmsg("cannot insert tuples in a parallel worker")));
2241 : :
2242 : 13711033 : tup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
2243 : 13711033 : tup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
2244 : 13711033 : tup->t_data->t_infomask |= HEAP_XMAX_INVALID;
2245 : 13711033 : HeapTupleHeaderSetXmin(tup->t_data, xid);
2246 [ + + ]: 13711033 : if (options & HEAP_INSERT_FROZEN)
2247 : 102651 : HeapTupleHeaderSetXminFrozen(tup->t_data);
2248 : :
2249 : 13711033 : HeapTupleHeaderSetCmin(tup->t_data, cid);
2250 : 13711033 : HeapTupleHeaderSetXmax(tup->t_data, 0); /* for cleanliness */
2251 : 13711033 : tup->t_tableOid = RelationGetRelid(relation);
2252 : :
2253 : : /*
2254 : : * If the new tuple is too big for storage or contains already toasted
2255 : : * out-of-line attributes from some other relation, invoke the toaster.
2256 : : */
2257 [ + + ]: 13711033 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
2258 [ + + ]: 36423 : relation->rd_rel->relkind != RELKIND_MATVIEW)
2259 : : {
2260 : : /* toast table entries should never be recursively toasted */
2261 : : Assert(!HeapTupleHasExternal(tup));
2262 : 36362 : return tup;
2263 : : }
2264 [ + + + + ]: 13674671 : else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
2265 : 22709 : return heap_toast_insert_or_update(relation, tup, NULL, options);
2266 : : else
2267 : 13651962 : return tup;
2268 : : }
2269 : :
2270 : : /*
2271 : : * Helper for heap_multi_insert() that computes the number of entire pages
2272 : : * that inserting the remaining heaptuples requires. Used to determine how
2273 : : * much the relation needs to be extended by.
2274 : : */
2275 : : static int
2276 : 498607 : heap_multi_insert_pages(HeapTuple *heaptuples, int done, int ntuples, Size saveFreeSpace)
2277 : : {
2278 : 498607 : size_t page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2279 : 498607 : int npages = 1;
2280 : :
2281 [ + + ]: 3024959 : for (int i = done; i < ntuples; i++)
2282 : : {
2283 : 2526352 : size_t tup_sz = sizeof(ItemIdData) + MAXALIGN(heaptuples[i]->t_len);
2284 : :
2285 [ + + ]: 2526352 : if (page_avail < tup_sz)
2286 : : {
2287 : 17622 : npages++;
2288 : 17622 : page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2289 : : }
2290 : 2526352 : page_avail -= tup_sz;
2291 : : }
2292 : :
2293 : 498607 : return npages;
2294 : : }
2295 : :
2296 : : /*
2297 : : * heap_multi_insert - insert multiple tuples into a heap
2298 : : *
2299 : : * This is like heap_insert(), but inserts multiple tuples in one operation.
2300 : : * That's faster than calling heap_insert() in a loop, because when multiple
2301 : : * tuples can be inserted on a single page, we can write just a single WAL
2302 : : * record covering all of them, and only need to lock/unlock the page once.
2303 : : *
2304 : : * Note: this leaks memory into the current memory context. You can create a
2305 : : * temporary context before calling this, if that's a problem.
2306 : : */
2307 : : void
2308 : 490282 : heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
2309 : : CommandId cid, uint32 options, BulkInsertState bistate)
2310 : : {
2311 : 490282 : TransactionId xid = GetCurrentTransactionId();
2312 : : HeapTuple *heaptuples;
2313 : : int i;
2314 : : int ndone;
2315 : : PGAlignedBlock scratch;
2316 : : Page page;
2317 : 490282 : Buffer vmbuffer = InvalidBuffer;
2318 : : bool needwal;
2319 : : Size saveFreeSpace;
2320 [ + + + + : 490282 : bool need_tuple_data = RelationIsLogicallyLogged(relation);
+ - - + -
- - - + -
+ + ]
2321 [ + + + + : 490282 : bool need_cids = RelationIsAccessibleInLogicalDecoding(relation);
+ - - + -
- - - + +
- + - - -
- - - ]
2322 : 490282 : bool starting_with_empty_page = false;
2323 : 490282 : int npages = 0;
2324 : 490282 : int npages_used = 0;
2325 : :
2326 : : /* currently not needed (thus unsupported) for heap_multi_insert() */
2327 : : Assert(!(options & HEAP_INSERT_NO_LOGICAL));
2328 : :
2329 : 490282 : AssertHasSnapshotForToast(relation);
2330 : :
2331 [ + + + + : 490282 : needwal = RelationNeedsWAL(relation);
+ + + + ]
2332 [ + + ]: 490282 : saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
2333 : : HEAP_DEFAULT_FILLFACTOR);
2334 : :
2335 : : /* Toast and set header data in all the slots */
2336 : 490282 : heaptuples = palloc(ntuples * sizeof(HeapTuple));
2337 [ + + ]: 2337045 : for (i = 0; i < ntuples; i++)
2338 : : {
2339 : : HeapTuple tuple;
2340 : :
2341 : 1846763 : tuple = ExecFetchSlotHeapTuple(slots[i], true, NULL);
2342 : 1846763 : slots[i]->tts_tableOid = RelationGetRelid(relation);
2343 : 1846763 : tuple->t_tableOid = slots[i]->tts_tableOid;
2344 : 1846763 : heaptuples[i] = heap_prepare_insert(relation, tuple, xid, cid,
2345 : : options);
2346 : : }
2347 : :
2348 : : /*
2349 : : * We're about to do the actual inserts -- but check for conflict first,
2350 : : * to minimize the possibility of having to roll back work we've just
2351 : : * done.
2352 : : *
2353 : : * A check here does not definitively prevent a serialization anomaly;
2354 : : * that check MUST be done at least past the point of acquiring an
2355 : : * exclusive buffer content lock on every buffer that will be affected,
2356 : : * and MAY be done after all inserts are reflected in the buffers and
2357 : : * those locks are released; otherwise there is a race condition. Since
2358 : : * multiple buffers can be locked and unlocked in the loop below, and it
2359 : : * would not be feasible to identify and lock all of those buffers before
2360 : : * the loop, we must do a final check at the end.
2361 : : *
2362 : : * The check here could be omitted with no loss of correctness; it is
2363 : : * present strictly as an optimization.
2364 : : *
2365 : : * For heap inserts, we only need to check for table-level SSI locks. Our
2366 : : * new tuples can't possibly conflict with existing tuple locks, and heap
2367 : : * page locks are only consolidated versions of tuple locks; they do not
2368 : : * lock "gaps" as index page locks do. So we don't need to specify a
2369 : : * buffer when making the call, which makes for a faster check.
2370 : : */
2371 : 490282 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2372 : :
2373 : 490282 : ndone = 0;
2374 [ + + ]: 998224 : while (ndone < ntuples)
2375 : : {
2376 : : Buffer buffer;
2377 : 507942 : bool clear_all_visible = false;
2378 : 507942 : bool all_frozen_set = false;
2379 : 507942 : bool vmbuffer_modified = false;
2380 : : int nthispage;
2381 : :
2382 [ + + ]: 507942 : CHECK_FOR_INTERRUPTS();
2383 : :
2384 : : /*
2385 : : * Compute number of pages needed to fit the to-be-inserted tuples in
2386 : : * the worst case. This will be used to determine how much to extend
2387 : : * the relation by in RelationGetBufferForTuple(), if needed. If we
2388 : : * filled a prior page from scratch, we can just update our last
2389 : : * computation, but if we started with a partially filled page,
2390 : : * recompute from scratch, the number of potentially required pages
2391 : : * can vary due to tuples needing to fit onto the page, page headers
2392 : : * etc.
2393 : : */
2394 [ + + + + ]: 507942 : if (ndone == 0 || !starting_with_empty_page)
2395 : : {
2396 : 498607 : npages = heap_multi_insert_pages(heaptuples, ndone, ntuples,
2397 : : saveFreeSpace);
2398 : 498607 : npages_used = 0;
2399 : : }
2400 : : else
2401 : 9335 : npages_used++;
2402 : :
2403 : : /*
2404 : : * Find buffer where at least the next tuple will fit. If the page is
2405 : : * all-visible, this will also pin the requisite visibility map page.
2406 : : *
2407 : : * Also pin visibility map page if COPY FREEZE inserts tuples into an
2408 : : * empty page. See all_frozen_set below.
2409 : : */
2410 : 507942 : buffer = RelationGetBufferForTuple(relation, heaptuples[ndone]->t_len,
2411 : : InvalidBuffer, options, bistate,
2412 : : &vmbuffer, NULL,
2413 : : npages - npages_used);
2414 : 507942 : page = BufferGetPage(buffer);
2415 : :
2416 : 507942 : starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
2417 : :
2418 [ + + + + ]: 507942 : if (starting_with_empty_page && (options & HEAP_INSERT_FROZEN))
2419 : : {
2420 : 1665 : all_frozen_set = true;
2421 : : /* Lock the vmbuffer before entering the critical section */
2422 : 1665 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
2423 : : }
2424 [ + + + + ]: 506277 : else if (PageIsAllVisible(page) && !(options & HEAP_INSERT_FROZEN))
2425 : : {
2426 : 3513 : clear_all_visible = true;
2427 : : /* Lock the vmbuffer before entering the critical section */
2428 : 3513 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
2429 : : }
2430 : :
2431 : : /* NO EREPORT(ERROR) from here till changes are logged */
2432 : 507942 : START_CRIT_SECTION();
2433 : :
2434 : : /*
2435 : : * RelationGetBufferForTuple has ensured that the first tuple fits.
2436 : : * Put that on the page, and then as many other tuples as fit.
2437 : : */
2438 : 507942 : RelationPutHeapTuple(relation, buffer, heaptuples[ndone], false);
2439 : :
2440 : : /*
2441 : : * For logical decoding we need combo CIDs to properly decode the
2442 : : * catalog.
2443 : : */
2444 [ + + + + ]: 507942 : if (needwal && need_cids)
2445 : 5431 : log_heap_new_cid(relation, heaptuples[ndone]);
2446 : :
2447 [ + + ]: 1846763 : for (nthispage = 1; ndone + nthispage < ntuples; nthispage++)
2448 : : {
2449 : 1356481 : HeapTuple heaptup = heaptuples[ndone + nthispage];
2450 : :
2451 [ + + ]: 1356481 : if (PageGetHeapFreeSpace(page) < MAXALIGN(heaptup->t_len) + saveFreeSpace)
2452 : 17660 : break;
2453 : :
2454 : 1338821 : RelationPutHeapTuple(relation, buffer, heaptup, false);
2455 : :
2456 : : /*
2457 : : * For logical decoding we need combo CIDs to properly decode the
2458 : : * catalog.
2459 : : */
2460 [ + + + + ]: 1338821 : if (needwal && need_cids)
2461 : 5039 : log_heap_new_cid(relation, heaptup);
2462 : : }
2463 : :
2464 : : /*
2465 : : * If the page is all visible, need to clear that, unless we're only
2466 : : * going to add further frozen rows to it.
2467 : : *
2468 : : * If we're only adding already frozen rows to a previously empty
2469 : : * page, mark it as all-frozen and update the visibility map. We're
2470 : : * already holding a pin on the vmbuffer.
2471 : : */
2472 [ + + ]: 507942 : if (clear_all_visible)
2473 : : {
2474 : : Assert(!(options & HEAP_INSERT_FROZEN));
2475 : : /* It's possible the VM bits were already clear */
2476 [ + + ]: 3513 : if (visibilitymap_clear(relation->rd_locator,
2477 : : BufferGetBlockNumber(buffer),
2478 : : vmbuffer, VISIBILITYMAP_VALID_BITS))
2479 : 3512 : vmbuffer_modified = true;
2480 : :
2481 : 3513 : PageClearAllVisible(page);
2482 : : }
2483 [ + + ]: 504429 : else if (all_frozen_set)
2484 : : {
2485 : 1665 : PageSetAllVisible(page);
2486 : 1665 : PageClearPrunable(page);
2487 : 1665 : visibilitymap_set(BufferGetBlockNumber(buffer),
2488 : : vmbuffer,
2489 : : VISIBILITYMAP_ALL_VISIBLE |
2490 : : VISIBILITYMAP_ALL_FROZEN,
2491 : : relation->rd_locator);
2492 : : }
2493 : :
2494 : : /*
2495 : : * Set pd_prune_xid. See heap_insert() for more on why we do this when
2496 : : * inserting tuples. This only makes sense if we aren't already
2497 : : * setting the page frozen in the VM and we're not in bootstrap mode.
2498 : : */
2499 [ + + + + ]: 507942 : if (!all_frozen_set && TransactionIdIsNormal(xid))
2500 [ + + + + ]: 483591 : PageSetPrunable(page, xid);
2501 : :
2502 : 507942 : MarkBufferDirty(buffer);
2503 : :
2504 : : /* XLOG stuff */
2505 [ + + ]: 507942 : if (needwal)
2506 : : {
2507 : : XLogRecPtr recptr;
2508 : : xl_heap_multi_insert *xlrec;
2509 : 504042 : uint8 info = XLOG_HEAP2_MULTI_INSERT;
2510 : : char *tupledata;
2511 : : int totaldatalen;
2512 : 504042 : char *scratchptr = scratch.data;
2513 : : bool init;
2514 : 504042 : int bufflags = 0;
2515 : :
2516 : : /*
2517 : : * If the page was previously empty, we can reinit the page
2518 : : * instead of restoring the whole thing.
2519 : : */
2520 : 504042 : init = starting_with_empty_page;
2521 : :
2522 : : /* allocate xl_heap_multi_insert struct from the scratch area */
2523 : 504042 : xlrec = (xl_heap_multi_insert *) scratchptr;
2524 : 504042 : scratchptr += SizeOfHeapMultiInsert;
2525 : :
2526 : : /*
2527 : : * Allocate offsets array. Unless we're reinitializing the page,
2528 : : * in that case the tuples are stored in order starting at
2529 : : * FirstOffsetNumber and we don't need to store the offsets
2530 : : * explicitly.
2531 : : */
2532 [ + + ]: 504042 : if (!init)
2533 : 487350 : scratchptr += nthispage * sizeof(OffsetNumber);
2534 : :
2535 : : /* the rest of the scratch space is used for tuple data */
2536 : 504042 : tupledata = scratchptr;
2537 : :
2538 : : /* check that the mutually exclusive flags are not both set */
2539 : : Assert(!(clear_all_visible && all_frozen_set));
2540 : :
2541 : 504042 : xlrec->flags = 0;
2542 [ + + ]: 504042 : if (clear_all_visible)
2543 : 3513 : xlrec->flags = XLH_INSERT_ALL_VISIBLE_CLEARED;
2544 : :
2545 : : /*
2546 : : * We don't have to worry about including a conflict xid in the
2547 : : * WAL record, as HEAP_INSERT_FROZEN intentionally violates
2548 : : * visibility rules.
2549 : : */
2550 [ + + ]: 504042 : if (all_frozen_set)
2551 : 21 : xlrec->flags = XLH_INSERT_ALL_FROZEN_SET;
2552 : :
2553 : 504042 : xlrec->ntuples = nthispage;
2554 : :
2555 : : /*
2556 : : * Write out an xl_multi_insert_tuple and the tuple data itself
2557 : : * for each tuple.
2558 : : */
2559 [ + + ]: 2143627 : for (i = 0; i < nthispage; i++)
2560 : : {
2561 : 1639585 : HeapTuple heaptup = heaptuples[ndone + i];
2562 : : xl_multi_insert_tuple *tuphdr;
2563 : : int datalen;
2564 : :
2565 [ + + ]: 1639585 : if (!init)
2566 : 1003283 : xlrec->offsets[i] = ItemPointerGetOffsetNumber(&heaptup->t_self);
2567 : : /* xl_multi_insert_tuple needs two-byte alignment. */
2568 : 1639585 : tuphdr = (xl_multi_insert_tuple *) SHORTALIGN(scratchptr);
2569 : 1639585 : scratchptr = ((char *) tuphdr) + SizeOfMultiInsertTuple;
2570 : :
2571 : 1639585 : tuphdr->t_infomask2 = heaptup->t_data->t_infomask2;
2572 : 1639585 : tuphdr->t_infomask = heaptup->t_data->t_infomask;
2573 : 1639585 : tuphdr->t_hoff = heaptup->t_data->t_hoff;
2574 : :
2575 : : /* write bitmap [+ padding] [+ oid] + data */
2576 : 1639585 : datalen = heaptup->t_len - SizeofHeapTupleHeader;
2577 : 1639585 : memcpy(scratchptr,
2578 : 1639585 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2579 : : datalen);
2580 : 1639585 : tuphdr->datalen = datalen;
2581 : 1639585 : scratchptr += datalen;
2582 : : }
2583 : 504042 : totaldatalen = scratchptr - tupledata;
2584 : : Assert((scratchptr - scratch.data) < BLCKSZ);
2585 : :
2586 [ + + ]: 504042 : if (need_tuple_data)
2587 : 56 : xlrec->flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2588 : :
2589 : : /*
2590 : : * Signal that this is the last xl_heap_multi_insert record
2591 : : * emitted by this call to heap_multi_insert(). Needed for logical
2592 : : * decoding so it knows when to cleanup temporary data.
2593 : : */
2594 [ + + ]: 504042 : if (ndone + nthispage == ntuples)
2595 : 489827 : xlrec->flags |= XLH_INSERT_LAST_IN_MULTI;
2596 : :
2597 [ + + ]: 504042 : if (init)
2598 : : {
2599 : 16692 : info |= XLOG_HEAP_INIT_PAGE;
2600 : 16692 : bufflags |= REGBUF_WILL_INIT;
2601 : : }
2602 : :
2603 : : /*
2604 : : * If we're doing logical decoding, include the new tuple data
2605 : : * even if we take a full-page image of the page.
2606 : : */
2607 [ + + ]: 504042 : if (need_tuple_data)
2608 : 56 : bufflags |= REGBUF_KEEP_DATA;
2609 : :
2610 : 504042 : XLogBeginInsert();
2611 : 504042 : XLogRegisterData(xlrec, tupledata - scratch.data);
2612 : 504042 : XLogRegisterBuffer(HEAP_MULTI_INSERT_BLKREF_HEAP, buffer,
2613 : 504042 : REGBUF_STANDARD | bufflags);
2614 [ + + + + ]: 504042 : if (all_frozen_set || vmbuffer_modified)
2615 : 3533 : XLogRegisterBuffer(HEAP_MULTI_INSERT_BLKREF_VM, vmbuffer, 0);
2616 : :
2617 : 504042 : XLogRegisterBufData(HEAP_MULTI_INSERT_BLKREF_HEAP, tupledata,
2618 : : totaldatalen);
2619 : :
2620 : : /* filtering by origin on a row level is much more efficient */
2621 : 504042 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2622 : :
2623 : 504042 : recptr = XLogInsert(RM_HEAP2_ID, info);
2624 : :
2625 : 504042 : PageSetLSN(page, recptr);
2626 [ + + + + ]: 504042 : if (all_frozen_set || vmbuffer_modified)
2627 : : {
2628 : : Assert(BufferIsDirty(vmbuffer));
2629 : 3533 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
2630 : : }
2631 : : }
2632 : :
2633 : 507942 : END_CRIT_SECTION();
2634 : :
2635 : : /*
2636 : : * We locked vmbuffer if clear_all_visible was true regardless of
2637 : : * whether or not we ended up modifying the vmbuffer.
2638 : : */
2639 [ + + + + ]: 507942 : if (all_frozen_set || clear_all_visible)
2640 : 5178 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
2641 : :
2642 : 507942 : UnlockReleaseBuffer(buffer);
2643 : 507942 : ndone += nthispage;
2644 : :
2645 : : /*
2646 : : * NB: Only release vmbuffer after inserting all tuples - it's fairly
2647 : : * likely that we'll insert into subsequent heap pages that are likely
2648 : : * to use the same vm page.
2649 : : */
2650 : : }
2651 : :
2652 : : /* We're done with inserting all tuples, so release the last vmbuffer. */
2653 [ + + ]: 490282 : if (vmbuffer != InvalidBuffer)
2654 : 3618 : ReleaseBuffer(vmbuffer);
2655 : :
2656 : : /*
2657 : : * We're done with the actual inserts. Check for conflicts again, to
2658 : : * ensure that all rw-conflicts in to these inserts are detected. Without
2659 : : * this final check, a sequential scan of the heap may have locked the
2660 : : * table after the "before" check, missing one opportunity to detect the
2661 : : * conflict, and then scanned the table before the new tuples were there,
2662 : : * missing the other chance to detect the conflict.
2663 : : *
2664 : : * For heap inserts, we only need to check for table-level SSI locks. Our
2665 : : * new tuples can't possibly conflict with existing tuple locks, and heap
2666 : : * page locks are only consolidated versions of tuple locks; they do not
2667 : : * lock "gaps" as index page locks do. So we don't need to specify a
2668 : : * buffer when making the call.
2669 : : */
2670 : 490282 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2671 : :
2672 : : /*
2673 : : * If tuples are cacheable, mark them for invalidation from the caches in
2674 : : * case we abort. Note it is OK to do this after releasing the buffer,
2675 : : * because the heaptuples data structure is all in local memory, not in
2676 : : * the shared buffer.
2677 : : */
2678 [ + + ]: 490282 : if (IsCatalogRelation(relation))
2679 : : {
2680 [ + + ]: 1620443 : for (i = 0; i < ntuples; i++)
2681 : 1131632 : CacheInvalidateHeapTuple(relation, heaptuples[i], NULL);
2682 : : }
2683 : :
2684 : : /* copy t_self fields back to the caller's slots */
2685 [ + + ]: 2337045 : for (i = 0; i < ntuples; i++)
2686 : 1846763 : slots[i]->tts_tid = heaptuples[i]->t_self;
2687 : :
2688 : 490282 : pgstat_count_heap_insert(relation, ntuples);
2689 : 490282 : }
2690 : :
2691 : : /*
2692 : : * simple_heap_insert - insert a tuple
2693 : : *
2694 : : * Currently, this routine differs from heap_insert only in supplying
2695 : : * a default command ID and not allowing access to the speedup options.
2696 : : *
2697 : : * This should be used rather than using heap_insert directly in most places
2698 : : * where we are modifying system catalogs.
2699 : : */
2700 : : void
2701 : 1089622 : simple_heap_insert(Relation relation, HeapTuple tup)
2702 : : {
2703 : 1089622 : heap_insert(relation, tup, GetCurrentCommandId(true), 0, NULL);
2704 : 1089622 : }
2705 : :
2706 : : /*
2707 : : * Given infomask/infomask2, compute the bits that must be saved in the
2708 : : * "infobits" field of xl_heap_delete, xl_heap_update, xl_heap_lock,
2709 : : * xl_heap_lock_updated WAL records.
2710 : : *
2711 : : * See fix_infomask_from_infobits.
2712 : : */
2713 : : static uint8
2714 : 6565507 : compute_infobits(uint16 infomask, uint16 infomask2)
2715 : : {
2716 : : return
2717 : 6565507 : ((infomask & HEAP_XMAX_IS_MULTI) != 0 ? XLHL_XMAX_IS_MULTI : 0) |
2718 : 6565507 : ((infomask & HEAP_XMAX_LOCK_ONLY) != 0 ? XLHL_XMAX_LOCK_ONLY : 0) |
2719 : 6565507 : ((infomask & HEAP_XMAX_EXCL_LOCK) != 0 ? XLHL_XMAX_EXCL_LOCK : 0) |
2720 : : /* note we ignore HEAP_XMAX_SHR_LOCK here */
2721 : 13131014 : ((infomask & HEAP_XMAX_KEYSHR_LOCK) != 0 ? XLHL_XMAX_KEYSHR_LOCK : 0) |
2722 : : ((infomask2 & HEAP_KEYS_UPDATED) != 0 ?
2723 : 6565507 : XLHL_KEYS_UPDATED : 0);
2724 : : }
2725 : :
2726 : : /*
2727 : : * Given two versions of the same t_infomask for a tuple, compare them and
2728 : : * return whether the relevant status for a tuple Xmax has changed. This is
2729 : : * used after a buffer lock has been released and reacquired: we want to ensure
2730 : : * that the tuple state continues to be the same it was when we previously
2731 : : * examined it.
2732 : : *
2733 : : * Note the Xmax field itself must be compared separately.
2734 : : */
2735 : : static inline bool
2736 : 5452 : xmax_infomask_changed(uint16 new_infomask, uint16 old_infomask)
2737 : : {
2738 : 5452 : const uint16 interesting =
2739 : : HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY | HEAP_LOCK_MASK;
2740 : :
2741 [ + + ]: 5452 : if ((new_infomask & interesting) != (old_infomask & interesting))
2742 : 17 : return true;
2743 : :
2744 : 5435 : return false;
2745 : : }
2746 : :
2747 : : /*
2748 : : * heap_delete - delete a tuple
2749 : : *
2750 : : * See table_tuple_delete() for an explanation of the parameters, except that
2751 : : * this routine directly takes a tuple rather than a slot.
2752 : : *
2753 : : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
2754 : : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
2755 : : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
2756 : : * generated by another transaction).
2757 : : */
2758 : : TM_Result
2759 : 1923718 : heap_delete(Relation relation, const ItemPointerData *tid,
2760 : : CommandId cid, uint32 options, Snapshot crosscheck,
2761 : : bool wait, TM_FailureData *tmfd)
2762 : : {
2763 : : TM_Result result;
2764 : 1923718 : TransactionId xid = GetCurrentTransactionId();
2765 : : ItemId lp;
2766 : : HeapTupleData tp;
2767 : : Page page;
2768 : : BlockNumber block;
2769 : : Buffer buffer;
2770 : 1923718 : Buffer vmbuffer = InvalidBuffer;
2771 : 1923718 : bool vmbuffer_modified = false;
2772 : : TransactionId new_xmax;
2773 : : uint16 new_infomask,
2774 : : new_infomask2;
2775 : 1923718 : bool changingPart = (options & TABLE_DELETE_CHANGING_PARTITION) != 0;
2776 : 1923718 : bool walLogical = (options & TABLE_DELETE_NO_LOGICAL) == 0;
2777 : 1923718 : bool have_tuple_lock = false;
2778 : : bool iscombo;
2779 : 1923718 : bool clear_all_visible = false;
2780 : 1923718 : HeapTuple old_key_tuple = NULL; /* replica identity of the tuple */
2781 : 1923718 : bool old_key_copied = false;
2782 : :
2783 : : Assert(ItemPointerIsValid(tid));
2784 : :
2785 : 1923718 : AssertHasSnapshotForToast(relation);
2786 : :
2787 : : /*
2788 : : * Forbid this during a parallel operation, lest it allocate a combo CID.
2789 : : * Other workers might need that combo CID for visibility checks, and we
2790 : : * have no provision for broadcasting it to them.
2791 : : */
2792 [ - + ]: 1923718 : if (IsInParallelMode())
2793 [ # # ]: 0 : ereport(ERROR,
2794 : : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2795 : : errmsg("cannot delete tuples during a parallel operation")));
2796 : :
2797 : 1923718 : block = ItemPointerGetBlockNumber(tid);
2798 : 1923718 : buffer = ReadBuffer(relation, block);
2799 : 1923718 : page = BufferGetPage(buffer);
2800 : :
2801 : : /*
2802 : : * Before locking the buffer, pin the visibility map page if it appears to
2803 : : * be necessary. Since we haven't got the lock yet, someone else might be
2804 : : * in the middle of changing this, so we'll need to recheck after we have
2805 : : * the lock.
2806 : : */
2807 [ + + ]: 1923718 : if (PageIsAllVisible(page))
2808 : 1831 : visibilitymap_pin(relation, block, &vmbuffer);
2809 : :
2810 : 1923718 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2811 : :
2812 : 1923718 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
2813 : : Assert(ItemIdIsNormal(lp));
2814 : :
2815 : 1923718 : tp.t_tableOid = RelationGetRelid(relation);
2816 : 1923718 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2817 : 1923718 : tp.t_len = ItemIdGetLength(lp);
2818 : 1923718 : tp.t_self = *tid;
2819 : :
2820 : 1 : l1:
2821 : :
2822 : : /*
2823 : : * If we didn't pin the visibility map page and the page has become all
2824 : : * visible while we were busy locking the buffer, we'll have to unlock and
2825 : : * re-lock, to avoid holding the buffer lock across an I/O. That's a bit
2826 : : * unfortunate, but hopefully shouldn't happen often.
2827 : : */
2828 [ + + - + ]: 1923719 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
2829 : : {
2830 : 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2831 : 0 : visibilitymap_pin(relation, block, &vmbuffer);
2832 : 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2833 : : }
2834 : :
2835 : 1923719 : result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
2836 : :
2837 [ - + ]: 1923719 : if (result == TM_Invisible)
2838 : : {
2839 : 0 : UnlockReleaseBuffer(buffer);
2840 [ # # ]: 0 : ereport(ERROR,
2841 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2842 : : errmsg("attempted to delete invisible tuple")));
2843 : : }
2844 [ + + + - ]: 1923719 : else if (result == TM_BeingModified && wait)
2845 : : {
2846 : : TransactionId xwait;
2847 : : uint16 infomask;
2848 : :
2849 : : /* must copy state data before unlocking buffer */
2850 : 40672 : xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
2851 : 40672 : infomask = tp.t_data->t_infomask;
2852 : :
2853 : : /*
2854 : : * Sleep until concurrent transaction ends -- except when there's a
2855 : : * single locker and it's our own transaction. Note we don't care
2856 : : * which lock mode the locker has, because we need the strongest one.
2857 : : *
2858 : : * Before sleeping, we need to acquire tuple lock to establish our
2859 : : * priority for the tuple (see heap_lock_tuple). LockTuple will
2860 : : * release us when we are next-in-line for the tuple.
2861 : : *
2862 : : * If we are forced to "start over" below, we keep the tuple lock;
2863 : : * this arranges that we stay at the head of the line while rechecking
2864 : : * tuple state.
2865 : : */
2866 [ + + ]: 40672 : if (infomask & HEAP_XMAX_IS_MULTI)
2867 : : {
2868 : 8 : bool current_is_member = false;
2869 : :
2870 [ + - ]: 8 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
2871 : : LockTupleExclusive, ¤t_is_member))
2872 : : {
2873 : 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2874 : :
2875 : : /*
2876 : : * Acquire the lock, if necessary (but skip it when we're
2877 : : * requesting a lock and already have one; avoids deadlock).
2878 : : */
2879 [ + + ]: 8 : if (!current_is_member)
2880 : 6 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2881 : : LockWaitBlock, &have_tuple_lock);
2882 : :
2883 : : /* wait for multixact */
2884 : 8 : MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
2885 : : relation, &(tp.t_self), XLTW_Delete,
2886 : : NULL);
2887 : 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2888 : :
2889 : : /*
2890 : : * If xwait had just locked the tuple then some other xact
2891 : : * could update this tuple before we get to this point. Check
2892 : : * for xmax change, and start over if so.
2893 : : *
2894 : : * We also must start over if we didn't pin the VM page, and
2895 : : * the page has become all visible.
2896 : : */
2897 [ + - + - : 16 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
+ - ]
2898 [ - + ]: 16 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2899 : 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2900 : : xwait))
2901 : 0 : goto l1;
2902 : : }
2903 : :
2904 : : /*
2905 : : * You might think the multixact is necessarily done here, but not
2906 : : * so: it could have surviving members, namely our own xact or
2907 : : * other subxacts of this backend. It is legal for us to delete
2908 : : * the tuple in either case, however (the latter case is
2909 : : * essentially a situation of upgrading our former shared lock to
2910 : : * exclusive). We don't bother changing the on-disk hint bits
2911 : : * since we are about to overwrite the xmax altogether.
2912 : : */
2913 : : }
2914 [ + + ]: 40664 : else if (!TransactionIdIsCurrentTransactionId(xwait))
2915 : : {
2916 : : /*
2917 : : * Wait for regular transaction to end; but first, acquire tuple
2918 : : * lock.
2919 : : */
2920 : 71 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2921 : 71 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2922 : : LockWaitBlock, &have_tuple_lock);
2923 : 71 : XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
2924 : 67 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2925 : :
2926 : : /*
2927 : : * xwait is done, but if xwait had just locked the tuple then some
2928 : : * other xact could update this tuple before we get to this point.
2929 : : * Check for xmax change, and start over if so.
2930 : : *
2931 : : * We also must start over if we didn't pin the VM page, and the
2932 : : * page has become all visible.
2933 : : */
2934 [ + - + - : 134 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
+ + ]
2935 [ - + ]: 133 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
2936 : 66 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2937 : : xwait))
2938 : 1 : goto l1;
2939 : :
2940 : : /* Otherwise check if it committed or aborted */
2941 : 66 : UpdateXmaxHintBits(tp.t_data, buffer, xwait);
2942 : : }
2943 : :
2944 : : /*
2945 : : * We may overwrite if previous xmax aborted, or if it committed but
2946 : : * only locked the tuple without updating it.
2947 : : */
2948 [ + + + + ]: 81314 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
2949 [ + + ]: 40697 : HEAP_XMAX_IS_LOCKED_ONLY(tp.t_data->t_infomask) ||
2950 : 50 : HeapTupleHeaderIsOnlyLocked(tp.t_data))
2951 : 40621 : result = TM_Ok;
2952 [ + + ]: 46 : else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
2953 : 33 : result = TM_Updated;
2954 : : else
2955 : 13 : result = TM_Deleted;
2956 : : }
2957 : :
2958 : : /* sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
2959 : : if (result != TM_Ok)
2960 : : {
2961 : : Assert(result == TM_SelfModified ||
2962 : : result == TM_Updated ||
2963 : : result == TM_Deleted ||
2964 : : result == TM_BeingModified);
2965 : : Assert(!(tp.t_data->t_infomask & HEAP_XMAX_INVALID));
2966 : : Assert(result != TM_Updated ||
2967 : : !ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
2968 : : }
2969 : :
2970 [ + + + - ]: 1923714 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
2971 : : {
2972 : : /* Perform additional check for transaction-snapshot mode RI updates */
2973 [ + - ]: 1 : if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
2974 : 1 : result = TM_Updated;
2975 : : }
2976 : :
2977 [ + + ]: 1923714 : if (result != TM_Ok)
2978 : : {
2979 : 102 : tmfd->ctid = tp.t_data->t_ctid;
2980 : 102 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
2981 [ + + ]: 102 : if (result == TM_SelfModified)
2982 : 36 : tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
2983 : : else
2984 : 66 : tmfd->cmax = InvalidCommandId;
2985 : 102 : UnlockReleaseBuffer(buffer);
2986 [ + + ]: 102 : if (have_tuple_lock)
2987 : 46 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
2988 [ - + ]: 102 : if (vmbuffer != InvalidBuffer)
2989 : 0 : ReleaseBuffer(vmbuffer);
2990 : 102 : return result;
2991 : : }
2992 : :
2993 : : /*
2994 : : * We're about to do the actual delete -- check for conflict first, to
2995 : : * avoid possibly having to roll back work we've just done.
2996 : : *
2997 : : * This is safe without a recheck as long as there is no possibility of
2998 : : * another process scanning the page between this check and the delete
2999 : : * being visible to the scan (i.e., an exclusive buffer content lock is
3000 : : * continuously held from this point until the tuple delete is visible).
3001 : : */
3002 : 1923612 : CheckForSerializableConflictIn(relation, tid, BufferGetBlockNumber(buffer));
3003 : :
3004 : : /* replace cid with a combo CID if necessary */
3005 : 1923598 : HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
3006 : :
3007 : : /*
3008 : : * Compute replica identity tuple before entering the critical section so
3009 : : * we don't PANIC upon a memory allocation failure.
3010 : : */
3011 : 1923598 : old_key_tuple = walLogical ?
3012 [ + + ]: 1923598 : ExtractReplicaIdentity(relation, &tp, true, &old_key_copied) : NULL;
3013 : :
3014 : : /*
3015 : : * If this is the first possibly-multixact-able operation in the current
3016 : : * transaction, set my per-backend OldestMemberMXactId setting. We can be
3017 : : * certain that the transaction will never become a member of any older
3018 : : * MultiXactIds than that. (We have to do this even if we end up just
3019 : : * using our own TransactionId below, since some other backend could
3020 : : * incorporate our XID into a MultiXact immediately afterwards.)
3021 : : */
3022 : 1923598 : MultiXactIdSetOldestMember();
3023 : :
3024 : 1923598 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(tp.t_data),
3025 : 1923598 : tp.t_data->t_infomask, tp.t_data->t_infomask2,
3026 : : xid, LockTupleExclusive, true,
3027 : : &new_xmax, &new_infomask, &new_infomask2);
3028 : :
3029 : : /* Lock the VM before entering the critical section */
3030 [ + + ]: 1923598 : if (PageIsAllVisible(page))
3031 : : {
3032 : 1831 : clear_all_visible = true;
3033 : 1831 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
3034 : : }
3035 : :
3036 : 1923598 : START_CRIT_SECTION();
3037 : :
3038 : : /*
3039 : : * If this transaction commits, the tuple will become DEAD sooner or
3040 : : * later. Set flag that this page is a candidate for pruning once our xid
3041 : : * falls below the OldestXmin horizon. If the transaction finally aborts,
3042 : : * the subsequent page pruning will be a no-op and the hint will be
3043 : : * cleared.
3044 : : */
3045 [ + + + + ]: 1923598 : PageSetPrunable(page, xid);
3046 : :
3047 [ + + ]: 1923598 : if (clear_all_visible)
3048 : : {
3049 : : /* It's possible the VM bits were already clear */
3050 [ + - ]: 1831 : if (visibilitymap_clear(relation->rd_locator, BufferGetBlockNumber(buffer),
3051 : : vmbuffer, VISIBILITYMAP_VALID_BITS))
3052 : 1831 : vmbuffer_modified = true;
3053 : :
3054 : 1831 : PageClearAllVisible(page);
3055 : : }
3056 : :
3057 : : /* store transaction information of xact deleting the tuple */
3058 : 1923598 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3059 : 1923598 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3060 : 1923598 : tp.t_data->t_infomask |= new_infomask;
3061 : 1923598 : tp.t_data->t_infomask2 |= new_infomask2;
3062 : 1923598 : HeapTupleHeaderClearHotUpdated(tp.t_data);
3063 : 1923598 : HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
3064 : 1923598 : HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
3065 : : /* Make sure there is no forward chain link in t_ctid */
3066 : 1923598 : tp.t_data->t_ctid = tp.t_self;
3067 : :
3068 : : /* Signal that this is actually a move into another partition */
3069 [ + + ]: 1923598 : if (changingPart)
3070 : 666 : HeapTupleHeaderSetMovedPartitions(tp.t_data);
3071 : :
3072 : 1923598 : MarkBufferDirty(buffer);
3073 : :
3074 : : /*
3075 : : * XLOG stuff
3076 : : *
3077 : : * NB: heap_abort_speculative() uses the same xlog record and replay
3078 : : * routines.
3079 : : */
3080 [ + + + + : 1923598 : if (RelationNeedsWAL(relation))
+ - + + ]
3081 : : {
3082 : : xl_heap_delete xlrec;
3083 : : xl_heap_header xlhdr;
3084 : : XLogRecPtr recptr;
3085 : :
3086 : : /*
3087 : : * For logical decode we need combo CIDs to properly decode the
3088 : : * catalog
3089 : : */
3090 [ + + + + : 1840351 : if (RelationIsAccessibleInLogicalDecoding(relation))
+ - - + -
- - - + +
+ + - + -
- - + ]
3091 : 6971 : log_heap_new_cid(relation, &tp);
3092 : :
3093 : 1840351 : xlrec.flags = 0;
3094 [ + + ]: 1840351 : if (clear_all_visible)
3095 : 1831 : xlrec.flags |= XLH_DELETE_ALL_VISIBLE_CLEARED;
3096 [ + + ]: 1840351 : if (changingPart)
3097 : 666 : xlrec.flags |= XLH_DELETE_IS_PARTITION_MOVE;
3098 : 3680702 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
3099 : 1840351 : tp.t_data->t_infomask2);
3100 : 1840351 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
3101 : 1840351 : xlrec.xmax = new_xmax;
3102 : :
3103 [ + + ]: 1840351 : if (old_key_tuple != NULL)
3104 : : {
3105 [ + + ]: 47029 : if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
3106 : 135 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_TUPLE;
3107 : : else
3108 : 46894 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_KEY;
3109 : : }
3110 : :
3111 : : /*
3112 : : * Mark the change as not-for-logical-decoding if caller requested so.
3113 : : *
3114 : : * (This is used for changes that affect relations not visible to
3115 : : * other transactions, such as the transient table during concurrent
3116 : : * repack.)
3117 : : */
3118 [ + + ]: 1840351 : if (!walLogical)
3119 : 3 : xlrec.flags |= XLH_DELETE_NO_LOGICAL;
3120 : :
3121 : 1840351 : XLogBeginInsert();
3122 : 1840351 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
3123 : :
3124 : 1840351 : XLogRegisterBuffer(HEAP_DELETE_BLKREF_HEAP, buffer, REGBUF_STANDARD);
3125 : :
3126 : : /*
3127 : : * Log replica identity of the deleted tuple if there is one
3128 : : */
3129 [ + + ]: 1840351 : if (old_key_tuple != NULL)
3130 : : {
3131 : 47029 : xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
3132 : 47029 : xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
3133 : 47029 : xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
3134 : :
3135 : 47029 : XLogRegisterData(&xlhdr, SizeOfHeapHeader);
3136 : 47029 : XLogRegisterData((char *) old_key_tuple->t_data
3137 : : + SizeofHeapTupleHeader,
3138 : 47029 : old_key_tuple->t_len
3139 : : - SizeofHeapTupleHeader);
3140 : : }
3141 : :
3142 : : /* filtering by origin on a row level is much more efficient */
3143 : 1840351 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
3144 : :
3145 [ + + ]: 1840351 : if (vmbuffer_modified)
3146 : 1831 : XLogRegisterBuffer(HEAP_DELETE_BLKREF_VM, vmbuffer, 0);
3147 : :
3148 : 1840351 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
3149 : :
3150 : 1840351 : PageSetLSN(page, recptr);
3151 : :
3152 [ + + ]: 1840351 : if (vmbuffer_modified)
3153 : 1831 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
3154 : : }
3155 : :
3156 : 1923598 : END_CRIT_SECTION();
3157 : :
3158 : : /*
3159 : : * Release VM lock first, since it covers many heap blocks. We locked
3160 : : * vmbuffer if clear_all_visible was true regardless of whether or not we
3161 : : * ended up modifying the vmbuffer.
3162 : : */
3163 [ + + ]: 1923598 : if (clear_all_visible)
3164 : 1831 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
3165 : :
3166 : 1923598 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3167 : :
3168 [ + + ]: 1923598 : if (vmbuffer != InvalidBuffer)
3169 : 1831 : ReleaseBuffer(vmbuffer);
3170 : :
3171 : : /*
3172 : : * If the tuple has toasted out-of-line attributes, we need to delete
3173 : : * those items too. We have to do this before releasing the buffer
3174 : : * because we need to look at the contents of the tuple, but it's OK to
3175 : : * release the content lock on the buffer first.
3176 : : */
3177 [ + + ]: 1927458 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3178 [ + + ]: 3873 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3179 : : {
3180 : : /* toast table entries should never be recursively toasted */
3181 : : Assert(!HeapTupleHasExternal(&tp));
3182 : : }
3183 [ + + ]: 1919738 : else if (HeapTupleHasExternal(&tp))
3184 : 546 : heap_toast_delete(relation, &tp, false);
3185 : :
3186 : : /*
3187 : : * Mark tuple for invalidation from system caches at next command
3188 : : * boundary. We have to do this before releasing the buffer because we
3189 : : * need to look at the contents of the tuple.
3190 : : */
3191 : 1923598 : CacheInvalidateHeapTuple(relation, &tp, NULL);
3192 : :
3193 : : /* Now we can release the buffer */
3194 : 1923598 : ReleaseBuffer(buffer);
3195 : :
3196 : : /*
3197 : : * Release the lmgr tuple lock, if we had it.
3198 : : */
3199 [ + + ]: 1923598 : if (have_tuple_lock)
3200 : 26 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3201 : :
3202 : 1923598 : pgstat_count_heap_delete(relation);
3203 : :
3204 [ + + + + ]: 1923598 : if (old_key_tuple != NULL && old_key_copied)
3205 : 46895 : heap_freetuple(old_key_tuple);
3206 : :
3207 : 1923598 : return TM_Ok;
3208 : : }
3209 : :
3210 : : /*
3211 : : * simple_heap_delete - delete a tuple
3212 : : *
3213 : : * This routine may be used to delete a tuple when concurrent updates of
3214 : : * the target tuple are not expected (for example, because we have a lock
3215 : : * on the relation associated with the tuple). Any failure is reported
3216 : : * via ereport().
3217 : : */
3218 : : void
3219 : 839483 : simple_heap_delete(Relation relation, const ItemPointerData *tid)
3220 : : {
3221 : : TM_Result result;
3222 : : TM_FailureData tmfd;
3223 : :
3224 : 839483 : result = heap_delete(relation, tid,
3225 : : GetCurrentCommandId(true),
3226 : : 0,
3227 : : InvalidSnapshot,
3228 : : true /* wait for commit */ ,
3229 : : &tmfd);
3230 [ - + - - : 839483 : switch (result)
- ]
3231 : : {
3232 : 0 : case TM_SelfModified:
3233 : : /* Tuple was already updated in current command? */
3234 [ # # ]: 0 : elog(ERROR, "tuple already updated by self");
3235 : : break;
3236 : :
3237 : 839483 : case TM_Ok:
3238 : : /* done successfully */
3239 : 839483 : break;
3240 : :
3241 : 0 : case TM_Updated:
3242 [ # # ]: 0 : elog(ERROR, "tuple concurrently updated");
3243 : : break;
3244 : :
3245 : 0 : case TM_Deleted:
3246 [ # # ]: 0 : elog(ERROR, "tuple concurrently deleted");
3247 : : break;
3248 : :
3249 : 0 : default:
3250 [ # # ]: 0 : elog(ERROR, "unrecognized heap_delete status: %u", result);
3251 : : break;
3252 : : }
3253 : 839483 : }
3254 : :
3255 : : /*
3256 : : * heap_update - replace a tuple
3257 : : *
3258 : : * See table_tuple_update() for an explanation of the parameters, except that
3259 : : * this routine directly takes a tuple rather than a slot.
3260 : : *
3261 : : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
3262 : : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
3263 : : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
3264 : : * generated by another transaction).
3265 : : */
3266 : : TM_Result
3267 : 2386848 : heap_update(Relation relation, const ItemPointerData *otid, HeapTuple newtup,
3268 : : CommandId cid, uint32 options pg_attribute_unused(), Snapshot crosscheck, bool wait,
3269 : : TM_FailureData *tmfd, LockTupleMode *lockmode,
3270 : : TU_UpdateIndexes *update_indexes)
3271 : : {
3272 : : TM_Result result;
3273 : 2386848 : TransactionId xid = GetCurrentTransactionId();
3274 : : Bitmapset *hot_attrs;
3275 : : Bitmapset *sum_attrs;
3276 : : Bitmapset *key_attrs;
3277 : : Bitmapset *id_attrs;
3278 : : Bitmapset *interesting_attrs;
3279 : : Bitmapset *modified_attrs;
3280 : : ItemId lp;
3281 : : HeapTupleData oldtup;
3282 : : HeapTuple heaptup;
3283 : 2386848 : HeapTuple old_key_tuple = NULL;
3284 : 2386848 : bool old_key_copied = false;
3285 : 2386848 : bool walLogical = (options & TABLE_UPDATE_NO_LOGICAL) == 0;
3286 : : Page page,
3287 : : newpage;
3288 : : BlockNumber block;
3289 : : MultiXactStatus mxact_status;
3290 : : Buffer buffer,
3291 : : newbuf,
3292 : 2386848 : vmbuffer = InvalidBuffer,
3293 : 2386848 : vmbuffer_new = InvalidBuffer;
3294 : 2386848 : bool unlock_vmbuffer = false;
3295 : 2386848 : bool unlock_vmbuffer_new = false;
3296 : : bool need_toast;
3297 : : Size newtupsize,
3298 : : pagefree;
3299 : 2386848 : bool have_tuple_lock = false;
3300 : : bool iscombo;
3301 : 2386848 : bool use_hot_update = false;
3302 : 2386848 : bool summarized_update = false;
3303 : : bool key_intact;
3304 : 2386848 : bool clear_all_visible = false;
3305 : 2386848 : bool clear_all_visible_new = false;
3306 : 2386848 : bool vmbuffer_modified = false;
3307 : 2386848 : bool vmbuffer_new_modified = false;
3308 : : bool checked_lockers;
3309 : : bool locker_remains;
3310 : 2386848 : bool id_has_external = false;
3311 : : TransactionId xmax_new_tuple,
3312 : : xmax_old_tuple;
3313 : : uint16 infomask_old_tuple,
3314 : : infomask2_old_tuple,
3315 : : infomask_new_tuple,
3316 : : infomask2_new_tuple;
3317 : :
3318 : : Assert(ItemPointerIsValid(otid));
3319 : :
3320 : : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
3321 : : Assert(HeapTupleHeaderGetNatts(newtup->t_data) <=
3322 : : RelationGetNumberOfAttributes(relation));
3323 : :
3324 : 2386848 : AssertHasSnapshotForToast(relation);
3325 : :
3326 : : /*
3327 : : * Forbid this during a parallel operation, lest it allocate a combo CID.
3328 : : * Other workers might need that combo CID for visibility checks, and we
3329 : : * have no provision for broadcasting it to them.
3330 : : */
3331 [ - + ]: 2386848 : if (IsInParallelMode())
3332 [ # # ]: 0 : ereport(ERROR,
3333 : : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
3334 : : errmsg("cannot update tuples during a parallel operation")));
3335 : :
3336 : : #ifdef USE_ASSERT_CHECKING
3337 : : check_lock_if_inplace_updateable_rel(relation, otid, newtup);
3338 : : #endif
3339 : :
3340 : : /*
3341 : : * Fetch the list of attributes to be checked for various operations.
3342 : : *
3343 : : * For HOT considerations, this is wasted effort if we fail to update or
3344 : : * have to put the new tuple on a different page. But we must compute the
3345 : : * list before obtaining buffer lock --- in the worst case, if we are
3346 : : * doing an update on one of the relevant system catalogs, we could
3347 : : * deadlock if we try to fetch the list later. In any case, the relcache
3348 : : * caches the data so this is usually pretty cheap.
3349 : : *
3350 : : * We also need columns used by the replica identity and columns that are
3351 : : * considered the "key" of rows in the table.
3352 : : *
3353 : : * Note that we get copies of each bitmap, so we need not worry about
3354 : : * relcache flush happening midway through.
3355 : : */
3356 : 2386848 : hot_attrs = RelationGetIndexAttrBitmap(relation,
3357 : : INDEX_ATTR_BITMAP_HOT_BLOCKING);
3358 : 2386848 : sum_attrs = RelationGetIndexAttrBitmap(relation,
3359 : : INDEX_ATTR_BITMAP_SUMMARIZED);
3360 : 2386848 : key_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_KEY);
3361 : 2386848 : id_attrs = RelationGetIndexAttrBitmap(relation,
3362 : : INDEX_ATTR_BITMAP_IDENTITY_KEY);
3363 : 2386848 : interesting_attrs = NULL;
3364 : 2386848 : interesting_attrs = bms_add_members(interesting_attrs, hot_attrs);
3365 : 2386848 : interesting_attrs = bms_add_members(interesting_attrs, sum_attrs);
3366 : 2386848 : interesting_attrs = bms_add_members(interesting_attrs, key_attrs);
3367 : 2386848 : interesting_attrs = bms_add_members(interesting_attrs, id_attrs);
3368 : :
3369 : 2386848 : block = ItemPointerGetBlockNumber(otid);
3370 : 2386848 : INJECTION_POINT("heap_update-before-pin", NULL);
3371 : 2386848 : buffer = ReadBuffer(relation, block);
3372 : 2386848 : page = BufferGetPage(buffer);
3373 : :
3374 : : /*
3375 : : * Before locking the buffer, pin the visibility map page if it appears to
3376 : : * be necessary. Since we haven't got the lock yet, someone else might be
3377 : : * in the middle of changing this, so we'll need to recheck after we have
3378 : : * the lock.
3379 : : */
3380 [ + + ]: 2386848 : if (PageIsAllVisible(page))
3381 : 2172 : visibilitymap_pin(relation, block, &vmbuffer);
3382 : :
3383 : 2386848 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3384 : :
3385 : 2386848 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(otid));
3386 : :
3387 : : /*
3388 : : * Usually, a buffer pin and/or snapshot blocks pruning of otid, ensuring
3389 : : * we see LP_NORMAL here. When the otid origin is a syscache, we may have
3390 : : * neither a pin nor a snapshot. Hence, we may see other LP_ states, each
3391 : : * of which indicates concurrent pruning.
3392 : : *
3393 : : * Failing with TM_Updated would be most accurate. However, unlike other
3394 : : * TM_Updated scenarios, we don't know the successor ctid in LP_UNUSED and
3395 : : * LP_DEAD cases. While the distinction between TM_Updated and TM_Deleted
3396 : : * does matter to SQL statements UPDATE and MERGE, those SQL statements
3397 : : * hold a snapshot that ensures LP_NORMAL. Hence, the choice between
3398 : : * TM_Updated and TM_Deleted affects only the wording of error messages.
3399 : : * Settle on TM_Deleted, for two reasons. First, it avoids complicating
3400 : : * the specification of when tmfd->ctid is valid. Second, it creates
3401 : : * error log evidence that we took this branch.
3402 : : *
3403 : : * Since it's possible to see LP_UNUSED at otid, it's also possible to see
3404 : : * LP_NORMAL for a tuple that replaced LP_UNUSED. If it's a tuple for an
3405 : : * unrelated row, we'll fail with "duplicate key value violates unique".
3406 : : * XXX if otid is the live, newer version of the newtup row, we'll discard
3407 : : * changes originating in versions of this catalog row after the version
3408 : : * the caller got from syscache. See syscache-update-pruned.spec.
3409 : : */
3410 [ + + ]: 2386848 : if (!ItemIdIsNormal(lp))
3411 : : {
3412 : : Assert(RelationSupportsSysCache(RelationGetRelid(relation)));
3413 : :
3414 : 1 : UnlockReleaseBuffer(buffer);
3415 : : Assert(!have_tuple_lock);
3416 [ + - ]: 1 : if (vmbuffer != InvalidBuffer)
3417 : 1 : ReleaseBuffer(vmbuffer);
3418 : 1 : tmfd->ctid = *otid;
3419 : 1 : tmfd->xmax = InvalidTransactionId;
3420 : 1 : tmfd->cmax = InvalidCommandId;
3421 : 1 : *update_indexes = TU_None;
3422 : :
3423 : 1 : bms_free(hot_attrs);
3424 : 1 : bms_free(sum_attrs);
3425 : 1 : bms_free(key_attrs);
3426 : 1 : bms_free(id_attrs);
3427 : : /* modified_attrs not yet initialized */
3428 : 1 : bms_free(interesting_attrs);
3429 : 1 : return TM_Deleted;
3430 : : }
3431 : :
3432 : : /*
3433 : : * Fill in enough data in oldtup for HeapDetermineColumnsInfo to work
3434 : : * properly.
3435 : : */
3436 : 2386847 : oldtup.t_tableOid = RelationGetRelid(relation);
3437 : 2386847 : oldtup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
3438 : 2386847 : oldtup.t_len = ItemIdGetLength(lp);
3439 : 2386847 : oldtup.t_self = *otid;
3440 : :
3441 : : /* the new tuple is ready, except for this: */
3442 : 2386847 : newtup->t_tableOid = RelationGetRelid(relation);
3443 : :
3444 : : /*
3445 : : * Determine columns modified by the update. Additionally, identify
3446 : : * whether any of the unmodified replica identity key attributes in the
3447 : : * old tuple is externally stored or not. This is required because for
3448 : : * such attributes the flattened value won't be WAL logged as part of the
3449 : : * new tuple so we must include it as part of the old_key_tuple. See
3450 : : * ExtractReplicaIdentity.
3451 : : */
3452 : 2386847 : modified_attrs = HeapDetermineColumnsInfo(relation, interesting_attrs,
3453 : : id_attrs, &oldtup,
3454 : : newtup, &id_has_external);
3455 : :
3456 : : /*
3457 : : * If we're not updating any "key" column, we can grab a weaker lock type.
3458 : : * This allows for more concurrency when we are running simultaneously
3459 : : * with foreign key checks.
3460 : : *
3461 : : * Note that if a column gets detoasted while executing the update, but
3462 : : * the value ends up being the same, this test will fail and we will use
3463 : : * the stronger lock. This is acceptable; the important case to optimize
3464 : : * is updates that don't manipulate key columns, not those that
3465 : : * serendipitously arrive at the same key values.
3466 : : */
3467 [ + + ]: 2386847 : if (!bms_overlap(modified_attrs, key_attrs))
3468 : : {
3469 : 2380904 : *lockmode = LockTupleNoKeyExclusive;
3470 : 2380904 : mxact_status = MultiXactStatusNoKeyUpdate;
3471 : 2380904 : key_intact = true;
3472 : :
3473 : : /*
3474 : : * If this is the first possibly-multixact-able operation in the
3475 : : * current transaction, set my per-backend OldestMemberMXactId
3476 : : * setting. We can be certain that the transaction will never become a
3477 : : * member of any older MultiXactIds than that. (We have to do this
3478 : : * even if we end up just using our own TransactionId below, since
3479 : : * some other backend could incorporate our XID into a MultiXact
3480 : : * immediately afterwards.)
3481 : : */
3482 : 2380904 : MultiXactIdSetOldestMember();
3483 : : }
3484 : : else
3485 : : {
3486 : 5943 : *lockmode = LockTupleExclusive;
3487 : 5943 : mxact_status = MultiXactStatusUpdate;
3488 : 5943 : key_intact = false;
3489 : : }
3490 : :
3491 : : /*
3492 : : * Note: beyond this point, use oldtup not otid to refer to old tuple.
3493 : : * otid may very well point at newtup->t_self, which we will overwrite
3494 : : * with the new tuple's location, so there's great risk of confusion if we
3495 : : * use otid anymore.
3496 : : */
3497 : :
3498 : 1 : l2:
3499 : 2386848 : checked_lockers = false;
3500 : 2386848 : locker_remains = false;
3501 : 2386848 : result = HeapTupleSatisfiesUpdate(&oldtup, cid, buffer);
3502 : :
3503 : : /* see below about the "no wait" case */
3504 : : Assert(result != TM_BeingModified || wait);
3505 : :
3506 [ - + ]: 2386848 : if (result == TM_Invisible)
3507 : : {
3508 : 0 : UnlockReleaseBuffer(buffer);
3509 [ # # ]: 0 : ereport(ERROR,
3510 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3511 : : errmsg("attempted to update invisible tuple")));
3512 : : }
3513 [ + + + - ]: 2386848 : else if (result == TM_BeingModified && wait)
3514 : : {
3515 : : TransactionId xwait;
3516 : : uint16 infomask;
3517 : 36552 : bool can_continue = false;
3518 : :
3519 : : /*
3520 : : * XXX note that we don't consider the "no wait" case here. This
3521 : : * isn't a problem currently because no caller uses that case, but it
3522 : : * should be fixed if such a caller is introduced. It wasn't a
3523 : : * problem previously because this code would always wait, but now
3524 : : * that some tuple locks do not conflict with one of the lock modes we
3525 : : * use, it is possible that this case is interesting to handle
3526 : : * specially.
3527 : : *
3528 : : * This may cause failures with third-party code that calls
3529 : : * heap_update directly.
3530 : : */
3531 : :
3532 : : /* must copy state data before unlocking buffer */
3533 : 36552 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3534 : 36552 : infomask = oldtup.t_data->t_infomask;
3535 : :
3536 : : /*
3537 : : * Now we have to do something about the existing locker. If it's a
3538 : : * multi, sleep on it; we might be awakened before it is completely
3539 : : * gone (or even not sleep at all in some cases); we need to preserve
3540 : : * it as locker, unless it is gone completely.
3541 : : *
3542 : : * If it's not a multi, we need to check for sleeping conditions
3543 : : * before actually going to sleep. If the update doesn't conflict
3544 : : * with the locks, we just continue without sleeping (but making sure
3545 : : * it is preserved).
3546 : : *
3547 : : * Before sleeping, we need to acquire tuple lock to establish our
3548 : : * priority for the tuple (see heap_lock_tuple). LockTuple will
3549 : : * release us when we are next-in-line for the tuple. Note we must
3550 : : * not acquire the tuple lock until we're sure we're going to sleep;
3551 : : * otherwise we're open for race conditions with other transactions
3552 : : * holding the tuple lock which sleep on us.
3553 : : *
3554 : : * If we are forced to "start over" below, we keep the tuple lock;
3555 : : * this arranges that we stay at the head of the line while rechecking
3556 : : * tuple state.
3557 : : */
3558 [ + + ]: 36552 : if (infomask & HEAP_XMAX_IS_MULTI)
3559 : : {
3560 : : TransactionId update_xact;
3561 : : int remain;
3562 : 179 : bool current_is_member = false;
3563 : :
3564 [ + + ]: 179 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
3565 : : *lockmode, ¤t_is_member))
3566 : : {
3567 : 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3568 : :
3569 : : /*
3570 : : * Acquire the lock, if necessary (but skip it when we're
3571 : : * requesting a lock and already have one; avoids deadlock).
3572 : : */
3573 [ - + ]: 8 : if (!current_is_member)
3574 : 0 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3575 : : LockWaitBlock, &have_tuple_lock);
3576 : :
3577 : : /* wait for multixact */
3578 : 8 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
3579 : : relation, &oldtup.t_self, XLTW_Update,
3580 : : &remain);
3581 : 8 : checked_lockers = true;
3582 : 8 : locker_remains = remain != 0;
3583 : 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3584 : :
3585 : : /*
3586 : : * If xwait had just locked the tuple then some other xact
3587 : : * could update this tuple before we get to this point. Check
3588 : : * for xmax change, and start over if so.
3589 : : */
3590 [ + - ]: 8 : if (xmax_infomask_changed(oldtup.t_data->t_infomask,
3591 [ - + ]: 8 : infomask) ||
3592 : 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3593 : : xwait))
3594 : 0 : goto l2;
3595 : : }
3596 : :
3597 : : /*
3598 : : * Note that the multixact may not be done by now. It could have
3599 : : * surviving members; our own xact or other subxacts of this
3600 : : * backend, and also any other concurrent transaction that locked
3601 : : * the tuple with LockTupleKeyShare if we only got
3602 : : * LockTupleNoKeyExclusive. If this is the case, we have to be
3603 : : * careful to mark the updated tuple with the surviving members in
3604 : : * Xmax.
3605 : : *
3606 : : * Note that there could have been another update in the
3607 : : * MultiXact. In that case, we need to check whether it committed
3608 : : * or aborted. If it aborted we are safe to update it again;
3609 : : * otherwise there is an update conflict, and we have to return
3610 : : * TableTuple{Deleted, Updated} below.
3611 : : *
3612 : : * In the LockTupleExclusive case, we still need to preserve the
3613 : : * surviving members: those would include the tuple locks we had
3614 : : * before this one, which are important to keep in case this
3615 : : * subxact aborts.
3616 : : */
3617 [ + + ]: 179 : if (!HEAP_XMAX_IS_LOCKED_ONLY(oldtup.t_data->t_infomask))
3618 : 8 : update_xact = HeapTupleGetUpdateXid(oldtup.t_data);
3619 : : else
3620 : 171 : update_xact = InvalidTransactionId;
3621 : :
3622 : : /*
3623 : : * There was no UPDATE in the MultiXact; or it aborted. No
3624 : : * TransactionIdIsInProgress() call needed here, since we called
3625 : : * MultiXactIdWait() above.
3626 : : */
3627 [ + + + + ]: 187 : if (!TransactionIdIsValid(update_xact) ||
3628 : 8 : TransactionIdDidAbort(update_xact))
3629 : 172 : can_continue = true;
3630 : : }
3631 [ + + ]: 36373 : else if (TransactionIdIsCurrentTransactionId(xwait))
3632 : : {
3633 : : /*
3634 : : * The only locker is ourselves; we can avoid grabbing the tuple
3635 : : * lock here, but must preserve our locking information.
3636 : : */
3637 : 36245 : checked_lockers = true;
3638 : 36245 : locker_remains = true;
3639 : 36245 : can_continue = true;
3640 : : }
3641 [ + + + + ]: 128 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) && key_intact)
3642 : : {
3643 : : /*
3644 : : * If it's just a key-share locker, and we're not changing the key
3645 : : * columns, we don't need to wait for it to end; but we need to
3646 : : * preserve it as locker.
3647 : : */
3648 : 29 : checked_lockers = true;
3649 : 29 : locker_remains = true;
3650 : 29 : can_continue = true;
3651 : : }
3652 : : else
3653 : : {
3654 : : /*
3655 : : * Wait for regular transaction to end; but first, acquire tuple
3656 : : * lock.
3657 : : */
3658 : 99 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3659 : 99 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3660 : : LockWaitBlock, &have_tuple_lock);
3661 : 99 : XactLockTableWait(xwait, relation, &oldtup.t_self,
3662 : : XLTW_Update);
3663 : 99 : checked_lockers = true;
3664 : 99 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3665 : :
3666 : : /*
3667 : : * xwait is done, but if xwait had just locked the tuple then some
3668 : : * other xact could update this tuple before we get to this point.
3669 : : * Check for xmax change, and start over if so.
3670 : : */
3671 [ + + - + ]: 197 : if (xmax_infomask_changed(oldtup.t_data->t_infomask, infomask) ||
3672 : 98 : !TransactionIdEquals(xwait,
3673 : : HeapTupleHeaderGetRawXmax(oldtup.t_data)))
3674 : 1 : goto l2;
3675 : :
3676 : : /* Otherwise check if it committed or aborted */
3677 : 98 : UpdateXmaxHintBits(oldtup.t_data, buffer, xwait);
3678 [ + + ]: 98 : if (oldtup.t_data->t_infomask & HEAP_XMAX_INVALID)
3679 : 22 : can_continue = true;
3680 : : }
3681 : :
3682 [ + + ]: 36551 : if (can_continue)
3683 : 36468 : result = TM_Ok;
3684 [ + + ]: 83 : else if (!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid))
3685 : 69 : result = TM_Updated;
3686 : : else
3687 : 14 : result = TM_Deleted;
3688 : : }
3689 : :
3690 : : /* Sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3691 : : if (result != TM_Ok)
3692 : : {
3693 : : Assert(result == TM_SelfModified ||
3694 : : result == TM_Updated ||
3695 : : result == TM_Deleted ||
3696 : : result == TM_BeingModified);
3697 : : Assert(!(oldtup.t_data->t_infomask & HEAP_XMAX_INVALID));
3698 : : Assert(result != TM_Updated ||
3699 : : !ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
3700 : : }
3701 : :
3702 [ + + + - ]: 2386847 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3703 : : {
3704 : : /* Perform additional check for transaction-snapshot mode RI updates */
3705 [ + - ]: 1 : if (!HeapTupleSatisfiesVisibility(&oldtup, crosscheck, buffer))
3706 : 1 : result = TM_Updated;
3707 : : }
3708 : :
3709 [ + + ]: 2386847 : if (result != TM_Ok)
3710 : : {
3711 : 209 : tmfd->ctid = oldtup.t_data->t_ctid;
3712 : 209 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(oldtup.t_data);
3713 [ + + ]: 209 : if (result == TM_SelfModified)
3714 : 73 : tmfd->cmax = HeapTupleHeaderGetCmax(oldtup.t_data);
3715 : : else
3716 : 136 : tmfd->cmax = InvalidCommandId;
3717 : 209 : UnlockReleaseBuffer(buffer);
3718 [ + + ]: 209 : if (have_tuple_lock)
3719 : 76 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
3720 [ - + ]: 209 : if (vmbuffer != InvalidBuffer)
3721 : 0 : ReleaseBuffer(vmbuffer);
3722 : 209 : *update_indexes = TU_None;
3723 : :
3724 : 209 : bms_free(hot_attrs);
3725 : 209 : bms_free(sum_attrs);
3726 : 209 : bms_free(key_attrs);
3727 : 209 : bms_free(id_attrs);
3728 : 209 : bms_free(modified_attrs);
3729 : 209 : bms_free(interesting_attrs);
3730 : 209 : return result;
3731 : : }
3732 : :
3733 : : /*
3734 : : * If we didn't pin the visibility map page and the page has become all
3735 : : * visible while we were busy locking the buffer, or during some
3736 : : * subsequent window during which we had it unlocked, we'll have to unlock
3737 : : * and re-lock, to avoid holding the buffer lock across an I/O. That's a
3738 : : * bit unfortunate, especially since we'll now have to recheck whether the
3739 : : * tuple has been locked or updated under us, but hopefully it won't
3740 : : * happen very often.
3741 : : */
3742 [ + + - + ]: 2386638 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3743 : : {
3744 : 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3745 : 0 : visibilitymap_pin(relation, block, &vmbuffer);
3746 : 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3747 : 0 : goto l2;
3748 : : }
3749 : :
3750 : : /* Fill in transaction status data */
3751 : :
3752 : : /*
3753 : : * If the tuple we're updating is locked, we need to preserve the locking
3754 : : * info in the old tuple's Xmax. Prepare a new Xmax value for this.
3755 : : */
3756 : 2386638 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3757 : 2386638 : oldtup.t_data->t_infomask,
3758 : 2386638 : oldtup.t_data->t_infomask2,
3759 : : xid, *lockmode, true,
3760 : : &xmax_old_tuple, &infomask_old_tuple,
3761 : : &infomask2_old_tuple);
3762 : :
3763 : : /*
3764 : : * And also prepare an Xmax value for the new copy of the tuple. If there
3765 : : * was no xmax previously, or there was one but all lockers are now gone,
3766 : : * then use InvalidTransactionId; otherwise, get the xmax from the old
3767 : : * tuple. (In rare cases that might also be InvalidTransactionId and yet
3768 : : * not have the HEAP_XMAX_INVALID bit set; that's fine.)
3769 : : */
3770 [ + + + - ]: 2423084 : if ((oldtup.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3771 [ + + ]: 72892 : HEAP_LOCKED_UPGRADED(oldtup.t_data->t_infomask) ||
3772 [ - + ]: 36275 : (checked_lockers && !locker_remains))
3773 : 2350192 : xmax_new_tuple = InvalidTransactionId;
3774 : : else
3775 : 36446 : xmax_new_tuple = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3776 : :
3777 [ + + ]: 2386638 : if (!TransactionIdIsValid(xmax_new_tuple))
3778 : : {
3779 : 2350192 : infomask_new_tuple = HEAP_XMAX_INVALID;
3780 : 2350192 : infomask2_new_tuple = 0;
3781 : : }
3782 : : else
3783 : : {
3784 : : /*
3785 : : * If we found a valid Xmax for the new tuple, then the infomask bits
3786 : : * to use on the new tuple depend on what was there on the old one.
3787 : : * Note that since we're doing an update, the only possibility is that
3788 : : * the lockers had FOR KEY SHARE lock.
3789 : : */
3790 [ + + ]: 36446 : if (oldtup.t_data->t_infomask & HEAP_XMAX_IS_MULTI)
3791 : : {
3792 : 172 : GetMultiXactIdHintBits(xmax_new_tuple, &infomask_new_tuple,
3793 : : &infomask2_new_tuple);
3794 : : }
3795 : : else
3796 : : {
3797 : 36274 : infomask_new_tuple = HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_LOCK_ONLY;
3798 : 36274 : infomask2_new_tuple = 0;
3799 : : }
3800 : : }
3801 : :
3802 : : /*
3803 : : * Prepare the new tuple with the appropriate initial values of Xmin and
3804 : : * Xmax, as well as initial infomask bits as computed above.
3805 : : */
3806 : 2386638 : newtup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
3807 : 2386638 : newtup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
3808 : 2386638 : HeapTupleHeaderSetXmin(newtup->t_data, xid);
3809 : 2386638 : HeapTupleHeaderSetCmin(newtup->t_data, cid);
3810 : 2386638 : newtup->t_data->t_infomask |= HEAP_UPDATED | infomask_new_tuple;
3811 : 2386638 : newtup->t_data->t_infomask2 |= infomask2_new_tuple;
3812 : 2386638 : HeapTupleHeaderSetXmax(newtup->t_data, xmax_new_tuple);
3813 : :
3814 : : /*
3815 : : * Replace cid with a combo CID if necessary. Note that we already put
3816 : : * the plain cid into the new tuple.
3817 : : */
3818 : 2386638 : HeapTupleHeaderAdjustCmax(oldtup.t_data, &cid, &iscombo);
3819 : :
3820 : : /*
3821 : : * If the toaster needs to be activated, OR if the new tuple will not fit
3822 : : * on the same page as the old, then we need to release the content lock
3823 : : * (but not the pin!) on the old tuple's buffer while we are off doing
3824 : : * TOAST and/or table-file-extension work. We must mark the old tuple to
3825 : : * show that it's locked, else other processes may try to update it
3826 : : * themselves.
3827 : : *
3828 : : * We need to invoke the toaster if there are already any out-of-line
3829 : : * toasted values present, or if the new tuple is over-threshold.
3830 : : */
3831 [ - + ]: 2386638 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3832 [ # # ]: 0 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3833 : : {
3834 : : /* toast table entries should never be recursively toasted */
3835 : : Assert(!HeapTupleHasExternal(&oldtup));
3836 : : Assert(!HeapTupleHasExternal(newtup));
3837 : 0 : need_toast = false;
3838 : : }
3839 : : else
3840 [ + + ]: 7159382 : need_toast = (HeapTupleHasExternal(&oldtup) ||
3841 [ + + ]: 4772744 : HeapTupleHasExternal(newtup) ||
3842 [ + + ]: 2386074 : newtup->t_len > TOAST_TUPLE_THRESHOLD);
3843 : :
3844 : 2386638 : pagefree = PageGetHeapFreeSpace(page);
3845 : :
3846 : 2386638 : newtupsize = MAXALIGN(newtup->t_len);
3847 : :
3848 [ + + + + ]: 2386638 : if (need_toast || newtupsize > pagefree)
3849 : 2198139 : {
3850 : : TransactionId xmax_lock_old_tuple;
3851 : : uint16 infomask_lock_old_tuple,
3852 : : infomask2_lock_old_tuple;
3853 : 2198139 : bool cleared_all_frozen = false;
3854 : :
3855 : : /*
3856 : : * To prevent concurrent sessions from updating the tuple, we have to
3857 : : * temporarily mark it locked, while we release the page-level lock.
3858 : : *
3859 : : * To satisfy the rule that any xid potentially appearing in a buffer
3860 : : * written out to disk, we unfortunately have to WAL log this
3861 : : * temporary modification. We can reuse xl_heap_lock for this
3862 : : * purpose. If we crash/error before following through with the
3863 : : * actual update, xmax will be of an aborted transaction, allowing
3864 : : * other sessions to proceed.
3865 : : */
3866 : :
3867 : : /*
3868 : : * Compute xmax / infomask appropriate for locking the tuple. This has
3869 : : * to be done separately from the combo that's going to be used for
3870 : : * updating, because the potentially created multixact would otherwise
3871 : : * be wrong.
3872 : : */
3873 : 2198139 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3874 : 2198139 : oldtup.t_data->t_infomask,
3875 : 2198139 : oldtup.t_data->t_infomask2,
3876 : : xid, *lockmode, false,
3877 : : &xmax_lock_old_tuple, &infomask_lock_old_tuple,
3878 : : &infomask2_lock_old_tuple);
3879 : :
3880 : : Assert(HEAP_XMAX_IS_LOCKED_ONLY(infomask_lock_old_tuple));
3881 : :
3882 [ + + ]: 2198139 : if (PageIsAllVisible(page))
3883 : : {
3884 : 1274 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
3885 : 1274 : unlock_vmbuffer = true;
3886 : : }
3887 : :
3888 : 2198139 : START_CRIT_SECTION();
3889 : :
3890 : : /* Clear obsolete visibility flags ... */
3891 : 2198139 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3892 : 2198139 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3893 : 2198139 : HeapTupleClearHotUpdated(&oldtup);
3894 : : /* ... and store info about transaction updating this tuple */
3895 : : Assert(TransactionIdIsValid(xmax_lock_old_tuple));
3896 : 2198139 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_lock_old_tuple);
3897 : 2198139 : oldtup.t_data->t_infomask |= infomask_lock_old_tuple;
3898 : 2198139 : oldtup.t_data->t_infomask2 |= infomask2_lock_old_tuple;
3899 : 2198139 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
3900 : :
3901 : : /* temporarily make it look not-updated, but locked */
3902 : 2198139 : oldtup.t_data->t_ctid = oldtup.t_self;
3903 : :
3904 : : /*
3905 : : * Clear all-frozen bit on visibility map if needed. We could
3906 : : * immediately reset ALL_VISIBLE, but given that the WAL logging
3907 : : * overhead would be unchanged, that doesn't seem necessarily
3908 : : * worthwhile.
3909 : : */
3910 [ + + ]: 2198139 : if (PageIsAllVisible(page))
3911 : : {
3912 : : /* It's possible all-frozen was already clear */
3913 [ + + ]: 1274 : if (visibilitymap_clear(relation->rd_locator, block, vmbuffer,
3914 : : VISIBILITYMAP_ALL_FROZEN))
3915 : 841 : cleared_all_frozen = true;
3916 : : }
3917 : :
3918 : 2198139 : MarkBufferDirty(buffer);
3919 : :
3920 [ + + + + : 2198139 : if (RelationNeedsWAL(relation))
+ - + + ]
3921 : : {
3922 : : xl_heap_lock xlrec;
3923 : : XLogRecPtr recptr;
3924 : :
3925 : 2188002 : XLogBeginInsert();
3926 : 2188002 : XLogRegisterBuffer(HEAP_LOCK_BLKREF_HEAP, buffer, REGBUF_STANDARD);
3927 : :
3928 : 2188002 : xlrec.offnum = ItemPointerGetOffsetNumber(&oldtup.t_self);
3929 : 2188002 : xlrec.xmax = xmax_lock_old_tuple;
3930 : 4376004 : xlrec.infobits_set = compute_infobits(oldtup.t_data->t_infomask,
3931 : 2188002 : oldtup.t_data->t_infomask2);
3932 : 2188002 : xlrec.flags =
3933 : 2188002 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
3934 : 2188002 : XLogRegisterData(&xlrec, SizeOfHeapLock);
3935 : :
3936 [ + + ]: 2188002 : if (cleared_all_frozen)
3937 : 841 : XLogRegisterBuffer(HEAP_LOCK_BLKREF_VM, vmbuffer, 0);
3938 : :
3939 : 2188002 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
3940 : 2188002 : PageSetLSN(page, recptr);
3941 : :
3942 [ + + ]: 2188002 : if (cleared_all_frozen)
3943 : 841 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
3944 : : }
3945 : :
3946 : 2198139 : END_CRIT_SECTION();
3947 : :
3948 : : /* release VM lock first, since it covers many heap blocks */
3949 [ + + ]: 2198139 : if (unlock_vmbuffer)
3950 : 1274 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
3951 : 2198139 : unlock_vmbuffer = false;
3952 : :
3953 : 2198139 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3954 : :
3955 : : /*
3956 : : * Let the toaster do its thing, if needed.
3957 : : *
3958 : : * Note: below this point, heaptup is the data we actually intend to
3959 : : * store into the relation; newtup is the caller's original untoasted
3960 : : * data.
3961 : : */
3962 [ + + ]: 2198139 : if (need_toast)
3963 : : {
3964 : : /* Note we always use WAL and FSM during updates */
3965 : 1958 : heaptup = heap_toast_insert_or_update(relation, newtup, &oldtup, 0);
3966 : 1958 : newtupsize = MAXALIGN(heaptup->t_len);
3967 : : }
3968 : : else
3969 : 2196181 : heaptup = newtup;
3970 : :
3971 : : /*
3972 : : * Now, do we need a new page for the tuple, or not? This is a bit
3973 : : * tricky since someone else could have added tuples to the page while
3974 : : * we weren't looking. We have to recheck the available space after
3975 : : * reacquiring the buffer lock. But don't bother to do that if the
3976 : : * former amount of free space is still not enough; it's unlikely
3977 : : * there's more free now than before.
3978 : : *
3979 : : * What's more, if we need to get a new page, we will need to acquire
3980 : : * buffer locks on both old and new pages. To avoid deadlock against
3981 : : * some other backend trying to get the same two locks in the other
3982 : : * order, we must be consistent about the order we get the locks in.
3983 : : * We use the rule "lock the lower-numbered page of the relation
3984 : : * first". To implement this, we must do RelationGetBufferForTuple
3985 : : * while not holding the lock on the old page, and we must rely on it
3986 : : * to get the locks on both pages in the correct order.
3987 : : *
3988 : : * Another consideration is that we need visibility map page pin(s) if
3989 : : * we will have to clear the all-visible flag on either page. If we
3990 : : * call RelationGetBufferForTuple, we rely on it to acquire any such
3991 : : * pins; but if we don't, we have to handle that here. Hence we need
3992 : : * a loop.
3993 : : */
3994 : : for (;;)
3995 : : {
3996 [ + + ]: 2198139 : if (newtupsize > pagefree)
3997 : : {
3998 : : /* It doesn't fit, must use RelationGetBufferForTuple. */
3999 : 2197446 : newbuf = RelationGetBufferForTuple(relation, heaptup->t_len,
4000 : : buffer, 0, NULL,
4001 : : &vmbuffer_new, &vmbuffer,
4002 : : 0);
4003 : : /* We're all done. */
4004 : 2197446 : break;
4005 : : }
4006 : : /* Acquire VM page pin if needed and we don't have it. */
4007 [ + + - + ]: 693 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
4008 : 0 : visibilitymap_pin(relation, block, &vmbuffer);
4009 : : /* Re-acquire the lock on the old tuple's page. */
4010 : 693 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
4011 : : /* Re-check using the up-to-date free space */
4012 : 693 : pagefree = PageGetHeapFreeSpace(page);
4013 [ + - ]: 693 : if (newtupsize > pagefree ||
4014 [ + + - + ]: 693 : (vmbuffer == InvalidBuffer && PageIsAllVisible(page)))
4015 : : {
4016 : : /*
4017 : : * Rats, it doesn't fit anymore, or somebody just now set the
4018 : : * all-visible flag. We must now unlock and loop to avoid
4019 : : * deadlock. Fortunately, this path should seldom be taken.
4020 : : */
4021 : 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4022 : : }
4023 : : else
4024 : : {
4025 : : /* We're all done. */
4026 : 693 : newbuf = buffer;
4027 : 693 : break;
4028 : : }
4029 : : }
4030 : : }
4031 : : else
4032 : : {
4033 : : /* No TOAST work needed, and it'll fit on same page */
4034 : 188499 : newbuf = buffer;
4035 : 188499 : heaptup = newtup;
4036 : : }
4037 : :
4038 : 2386638 : newpage = BufferGetPage(newbuf);
4039 : :
4040 : : /*
4041 : : * We're about to do the actual update -- check for conflict first, to
4042 : : * avoid possibly having to roll back work we've just done.
4043 : : *
4044 : : * This is safe without a recheck as long as there is no possibility of
4045 : : * another process scanning the pages between this check and the update
4046 : : * being visible to the scan (i.e., exclusive buffer content lock(s) are
4047 : : * continuously held from this point until the tuple update is visible).
4048 : : *
4049 : : * For the new tuple the only check needed is at the relation level, but
4050 : : * since both tuples are in the same relation and the check for oldtup
4051 : : * will include checking the relation level, there is no benefit to a
4052 : : * separate check for the new tuple.
4053 : : */
4054 : 2386638 : CheckForSerializableConflictIn(relation, &oldtup.t_self,
4055 : : BufferGetBlockNumber(buffer));
4056 : :
4057 : : /*
4058 : : * At this point newbuf and buffer are both pinned and locked, and newbuf
4059 : : * has enough space for the new tuple. If they are the same buffer, only
4060 : : * one pin is held.
4061 : : */
4062 : :
4063 [ + + ]: 2386626 : if (newbuf == buffer)
4064 : : {
4065 : : /*
4066 : : * Since the new tuple is going into the same page, we might be able
4067 : : * to do a HOT update. Check if any of the index columns have been
4068 : : * changed.
4069 : : */
4070 [ + + ]: 189180 : if (!bms_overlap(modified_attrs, hot_attrs))
4071 : : {
4072 : 172080 : use_hot_update = true;
4073 : :
4074 : : /*
4075 : : * If none of the columns that are used in hot-blocking indexes
4076 : : * were updated, we can apply HOT, but we do still need to check
4077 : : * if we need to update the summarizing indexes, and update those
4078 : : * indexes if the columns were updated, or we may fail to detect
4079 : : * e.g. value bound changes in BRIN minmax indexes.
4080 : : */
4081 [ + + ]: 172080 : if (bms_overlap(modified_attrs, sum_attrs))
4082 : 2188 : summarized_update = true;
4083 : : }
4084 : : }
4085 : : else
4086 : : {
4087 : : /* Set a hint that the old page could use prune/defrag */
4088 : 2197446 : PageSetFull(page);
4089 : : }
4090 : :
4091 : : /*
4092 : : * Compute replica identity tuple before entering the critical section so
4093 : : * we don't PANIC upon a memory allocation failure.
4094 : : * ExtractReplicaIdentity() will return NULL if nothing needs to be
4095 : : * logged. Pass old key required as true only if the replica identity key
4096 : : * columns are modified or it has external data.
4097 : : */
4098 : 2386626 : old_key_tuple = ExtractReplicaIdentity(relation, &oldtup,
4099 [ + + + + ]: 2386626 : bms_overlap(modified_attrs, id_attrs) ||
4100 : : id_has_external,
4101 : 2386626 : &old_key_copied);
4102 : :
4103 : 2386626 : clear_all_visible = PageIsAllVisible(page);
4104 [ + + + + ]: 2386626 : clear_all_visible_new = newbuf != buffer && PageIsAllVisible(newpage);
4105 : :
4106 : : /*
4107 : : * Clear PD_ALL_VISIBLE flags and reset visibility map bits for any heap
4108 : : * pages that were all-visible. If there are two heap pages, we may need
4109 : : * to clear VM bits for both.
4110 : : */
4111 [ + + + + ]: 2386626 : if (clear_all_visible && clear_all_visible_new &&
4112 [ + - ]: 144 : vmbuffer_new == vmbuffer)
4113 : : {
4114 : : /*
4115 : : * This is the more complicated case: both the new and old heap pages
4116 : : * are all-visible and both their VM bits are on the same page of the
4117 : : * VM, so we register a single VM buffer as HEAP_UPDATE_BLKREF_VM_NEW
4118 : : * in the WAL record. We must be careful to only lock and register one
4119 : : * buffer, even though we modify it twice -- once for each heap
4120 : : * block's VM bits.
4121 : : */
4122 : 144 : LockBuffer(vmbuffer_new, BUFFER_LOCK_EXCLUSIVE);
4123 : 144 : unlock_vmbuffer_new = true;
4124 : :
4125 : : /* We will not lock or attempt to modify old VM buffer */
4126 : : }
4127 : : else
4128 : : {
4129 : : /*
4130 : : * In all the remaining cases, we will clear at most one heap block's
4131 : : * VM bits per VM page.
4132 : : */
4133 : 4772964 : Buffer vmbuffers[2] = {
4134 [ + + ]: 2386482 : clear_all_visible ? vmbuffer : InvalidBuffer,
4135 [ + + ]: 2386482 : clear_all_visible_new ? vmbuffer_new : InvalidBuffer
4136 : : };
4137 : :
4138 : : /*
4139 : : * When both pages need different VM pages cleared, acquire the VM
4140 : : * buffer locks in VM block order to avoid deadlocks between backends
4141 : : * updating tuples in opposite directions across VM pages.
4142 : : */
4143 [ + + - + : 2386482 : if (clear_all_visible && clear_all_visible_new &&
- - ]
4144 : 0 : BufferGetBlockNumber(vmbuffers[0]) > BufferGetBlockNumber(vmbuffers[1]))
4145 : : {
4146 : 0 : Buffer swap = vmbuffers[0];
4147 : :
4148 : 0 : vmbuffers[0] = vmbuffers[1];
4149 : 0 : vmbuffers[1] = swap;
4150 : : }
4151 : :
4152 : : Assert((!BufferIsValid(vmbuffers[0]) && !BufferIsValid(vmbuffers[1])) ||
4153 : : vmbuffers[0] != vmbuffers[1]);
4154 : :
4155 [ + + ]: 2386482 : if (BufferIsValid(vmbuffers[0]))
4156 : 2027 : LockBuffer(vmbuffers[0], BUFFER_LOCK_EXCLUSIVE);
4157 [ + + ]: 2386482 : if (BufferIsValid(vmbuffers[1]))
4158 : 681 : LockBuffer(vmbuffers[1], BUFFER_LOCK_EXCLUSIVE);
4159 : :
4160 [ + + ]: 2386482 : if (clear_all_visible)
4161 : 2027 : unlock_vmbuffer = true;
4162 [ + + ]: 2386482 : if (clear_all_visible_new)
4163 : 681 : unlock_vmbuffer_new = true;
4164 : : }
4165 : :
4166 : : /* NO EREPORT(ERROR) from here till changes are logged */
4167 : 2386626 : START_CRIT_SECTION();
4168 : :
4169 : : /*
4170 : : * If this transaction commits, the old tuple will become DEAD sooner or
4171 : : * later. Set flag that this page is a candidate for pruning once our xid
4172 : : * falls below the OldestXmin horizon. If the transaction finally aborts,
4173 : : * the subsequent page pruning will be a no-op and the hint will be
4174 : : * cleared.
4175 : : *
4176 : : * We set the new page prunable as well. See heap_insert() for more on why
4177 : : * we do this when inserting tuples.
4178 : : */
4179 [ + + + + ]: 2386626 : PageSetPrunable(page, xid);
4180 [ + + ]: 2386626 : if (newbuf != buffer)
4181 [ + + + + ]: 2197446 : PageSetPrunable(newpage, xid);
4182 : :
4183 [ + + ]: 2386626 : if (use_hot_update)
4184 : : {
4185 : : /* Mark the old tuple as HOT-updated */
4186 : 172080 : HeapTupleSetHotUpdated(&oldtup);
4187 : : /* And mark the new tuple as heap-only */
4188 : 172080 : HeapTupleSetHeapOnly(heaptup);
4189 : : /* Mark the caller's copy too, in case different from heaptup */
4190 : 172080 : HeapTupleSetHeapOnly(newtup);
4191 : : }
4192 : : else
4193 : : {
4194 : : /* Make sure tuples are correctly marked as not-HOT */
4195 : 2214546 : HeapTupleClearHotUpdated(&oldtup);
4196 : 2214546 : HeapTupleClearHeapOnly(heaptup);
4197 : 2214546 : HeapTupleClearHeapOnly(newtup);
4198 : : }
4199 : :
4200 : 2386626 : RelationPutHeapTuple(relation, newbuf, heaptup, false); /* insert new tuple */
4201 : :
4202 : :
4203 : : /* Clear obsolete visibility flags, possibly set by ourselves above... */
4204 : 2386626 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
4205 : 2386626 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
4206 : : /* ... and store info about transaction updating this tuple */
4207 : : Assert(TransactionIdIsValid(xmax_old_tuple));
4208 : 2386626 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_old_tuple);
4209 : 2386626 : oldtup.t_data->t_infomask |= infomask_old_tuple;
4210 : 2386626 : oldtup.t_data->t_infomask2 |= infomask2_old_tuple;
4211 : 2386626 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
4212 : :
4213 : : /* record address of new tuple in t_ctid of old one */
4214 : 2386626 : oldtup.t_data->t_ctid = heaptup->t_self;
4215 : :
4216 : : /*
4217 : : * Clear PD_ALL_VISIBLE flags and reset all visibilitymap bits. In all
4218 : : * cases, it's possible that PD_ALL_VISIBLE was set but the corresponding
4219 : : * visibility map bits were already clear.
4220 : : */
4221 [ + + ]: 2386626 : if (clear_all_visible)
4222 : : {
4223 [ + - ]: 2171 : if (visibilitymap_clear(relation->rd_locator, block,
4224 : : vmbuffer, VISIBILITYMAP_VALID_BITS))
4225 : : {
4226 : : /*
4227 : : * When old and new heap blocks' VM bits are on the same VM page,
4228 : : * that page is registered in the WAL record only once. If both
4229 : : * heap pages were PD_ALL_VISIBLE and either VM bit needs
4230 : : * clearing, we register the VM buffer as
4231 : : * HEAP_UPDATE_BLKREF_VM_NEW.
4232 : : */
4233 [ + + + - ]: 2171 : if (clear_all_visible_new && vmbuffer == vmbuffer_new)
4234 : 144 : vmbuffer_new_modified = true;
4235 : : else
4236 : 2027 : vmbuffer_modified = true;
4237 : : }
4238 : :
4239 : 2171 : PageClearAllVisible(page);
4240 : : }
4241 [ + + ]: 2386626 : if (clear_all_visible_new)
4242 : : {
4243 : : /*
4244 : : * If both heap blocks' VM bits are on the same VM buffer, this will
4245 : : * clear the new heap block's VM bits from the shared vmbuffer.
4246 : : */
4247 [ + - ]: 825 : if (visibilitymap_clear(relation->rd_locator, BufferGetBlockNumber(newbuf),
4248 : : vmbuffer_new, VISIBILITYMAP_VALID_BITS))
4249 : 825 : vmbuffer_new_modified = true;
4250 : :
4251 : 825 : PageClearAllVisible(newpage);
4252 : : }
4253 : :
4254 [ + + ]: 2386626 : if (newbuf != buffer)
4255 : 2197446 : MarkBufferDirty(newbuf);
4256 : 2386626 : MarkBufferDirty(buffer);
4257 : :
4258 : : /* XLOG stuff */
4259 [ + + + + : 2386626 : if (RelationNeedsWAL(relation))
+ - + + ]
4260 : : {
4261 : : XLogRecPtr recptr;
4262 : :
4263 : : /*
4264 : : * For logical decoding we need combo CIDs to properly decode the
4265 : : * catalog.
4266 : : */
4267 [ + + + + : 2374838 : if (RelationIsAccessibleInLogicalDecoding(relation))
+ - - + -
- - - + +
+ + - + -
- + + ]
4268 : : {
4269 : 2728 : log_heap_new_cid(relation, &oldtup);
4270 : 2728 : log_heap_new_cid(relation, heaptup);
4271 : : }
4272 : :
4273 [ + + + + ]: 2374838 : recptr = log_heap_update(relation, buffer,
4274 : : vmbuffer_modified ? vmbuffer : InvalidBuffer,
4275 : : newbuf,
4276 : : vmbuffer_new_modified ? vmbuffer_new : InvalidBuffer,
4277 : : &oldtup, heaptup,
4278 : : old_key_tuple,
4279 : : clear_all_visible,
4280 : : clear_all_visible_new,
4281 : : walLogical);
4282 [ + + ]: 2374838 : if (newbuf != buffer)
4283 : : {
4284 : 2187317 : PageSetLSN(newpage, recptr);
4285 : : }
4286 : 2374838 : PageSetLSN(page, recptr);
4287 : :
4288 [ + + ]: 2374838 : if (vmbuffer_modified)
4289 : 2026 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
4290 [ + + ]: 2374838 : if (vmbuffer_new_modified)
4291 : 825 : PageSetLSN(BufferGetPage(vmbuffer_new), recptr);
4292 : : }
4293 : :
4294 : 2386626 : END_CRIT_SECTION();
4295 : :
4296 [ + + ]: 2386626 : if (unlock_vmbuffer)
4297 : 2027 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
4298 [ + + ]: 2386626 : if (unlock_vmbuffer_new)
4299 : 825 : LockBuffer(vmbuffer_new, BUFFER_LOCK_UNLOCK);
4300 : :
4301 [ + + ]: 2386626 : if (newbuf != buffer)
4302 : 2197446 : LockBuffer(newbuf, BUFFER_LOCK_UNLOCK);
4303 : 2386626 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4304 : :
4305 : : /*
4306 : : * Mark old tuple for invalidation from system caches at next command
4307 : : * boundary, and mark the new tuple for invalidation in case we abort. We
4308 : : * have to do this before releasing the buffer because oldtup is in the
4309 : : * buffer. (heaptup is all in local memory, but it's necessary to process
4310 : : * both tuple versions in one call to inval.c so we can avoid redundant
4311 : : * sinval messages.)
4312 : : */
4313 : 2386626 : CacheInvalidateHeapTuple(relation, &oldtup, heaptup);
4314 : :
4315 : : /* Now we can release the buffer(s) */
4316 [ + + ]: 2386626 : if (newbuf != buffer)
4317 : 2197446 : ReleaseBuffer(newbuf);
4318 : 2386626 : ReleaseBuffer(buffer);
4319 [ + + ]: 2386626 : if (BufferIsValid(vmbuffer_new))
4320 : 827 : ReleaseBuffer(vmbuffer_new);
4321 [ + + ]: 2386626 : if (BufferIsValid(vmbuffer))
4322 : 2171 : ReleaseBuffer(vmbuffer);
4323 : :
4324 : : /*
4325 : : * Release the lmgr tuple lock, if we had it.
4326 : : */
4327 [ + + ]: 2386626 : if (have_tuple_lock)
4328 : 22 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
4329 : :
4330 : 2386626 : pgstat_count_heap_update(relation, use_hot_update, newbuf != buffer);
4331 : :
4332 : : /*
4333 : : * If heaptup is a private copy, release it. Don't forget to copy t_self
4334 : : * back to the caller's image, too.
4335 : : */
4336 [ + + ]: 2386626 : if (heaptup != newtup)
4337 : : {
4338 : 1892 : newtup->t_self = heaptup->t_self;
4339 : 1892 : heap_freetuple(heaptup);
4340 : : }
4341 : :
4342 : : /*
4343 : : * If it is a HOT update, the update may still need to update summarized
4344 : : * indexes, lest we fail to update those summaries and get incorrect
4345 : : * results (for example, minmax bounds of the block may change with this
4346 : : * update).
4347 : : */
4348 [ + + ]: 2386626 : if (use_hot_update)
4349 : : {
4350 [ + + ]: 172080 : if (summarized_update)
4351 : 2188 : *update_indexes = TU_Summarizing;
4352 : : else
4353 : 169892 : *update_indexes = TU_None;
4354 : : }
4355 : : else
4356 : 2214546 : *update_indexes = TU_All;
4357 : :
4358 [ + + + + ]: 2386626 : if (old_key_tuple != NULL && old_key_copied)
4359 : 96 : heap_freetuple(old_key_tuple);
4360 : :
4361 : 2386626 : bms_free(hot_attrs);
4362 : 2386626 : bms_free(sum_attrs);
4363 : 2386626 : bms_free(key_attrs);
4364 : 2386626 : bms_free(id_attrs);
4365 : 2386626 : bms_free(modified_attrs);
4366 : 2386626 : bms_free(interesting_attrs);
4367 : :
4368 : 2386626 : return TM_Ok;
4369 : : }
4370 : :
4371 : : #ifdef USE_ASSERT_CHECKING
4372 : : /*
4373 : : * Confirm adequate lock held during heap_update(), per rules from
4374 : : * README.tuplock section "Locking to write inplace-updated tables".
4375 : : */
4376 : : static void
4377 : : check_lock_if_inplace_updateable_rel(Relation relation,
4378 : : const ItemPointerData *otid,
4379 : : HeapTuple newtup)
4380 : : {
4381 : : /* LOCKTAG_TUPLE acceptable for any catalog */
4382 : : switch (RelationGetRelid(relation))
4383 : : {
4384 : : case RelationRelationId:
4385 : : case DatabaseRelationId:
4386 : : {
4387 : : LOCKTAG tuptag;
4388 : :
4389 : : SET_LOCKTAG_TUPLE(tuptag,
4390 : : relation->rd_lockInfo.lockRelId.dbId,
4391 : : relation->rd_lockInfo.lockRelId.relId,
4392 : : ItemPointerGetBlockNumber(otid),
4393 : : ItemPointerGetOffsetNumber(otid));
4394 : : if (LockHeldByMe(&tuptag, InplaceUpdateTupleLock, false))
4395 : : return;
4396 : : }
4397 : : break;
4398 : : default:
4399 : : Assert(!IsInplaceUpdateRelation(relation));
4400 : : return;
4401 : : }
4402 : :
4403 : : switch (RelationGetRelid(relation))
4404 : : {
4405 : : case RelationRelationId:
4406 : : {
4407 : : /* LOCKTAG_TUPLE or LOCKTAG_RELATION ok */
4408 : : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(newtup);
4409 : : Oid relid = classForm->oid;
4410 : : Oid dbid;
4411 : : LOCKTAG tag;
4412 : :
4413 : : if (IsSharedRelation(relid))
4414 : : dbid = InvalidOid;
4415 : : else
4416 : : dbid = MyDatabaseId;
4417 : :
4418 : : if (classForm->relkind == RELKIND_INDEX)
4419 : : {
4420 : : Relation irel = index_open(relid, AccessShareLock);
4421 : :
4422 : : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4423 : : index_close(irel, AccessShareLock);
4424 : : }
4425 : : else
4426 : : SET_LOCKTAG_RELATION(tag, dbid, relid);
4427 : :
4428 : : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, false) &&
4429 : : !LockHeldByMe(&tag, ShareRowExclusiveLock, true))
4430 : : elog(WARNING,
4431 : : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4432 : : NameStr(classForm->relname),
4433 : : relid,
4434 : : classForm->relkind,
4435 : : ItemPointerGetBlockNumber(otid),
4436 : : ItemPointerGetOffsetNumber(otid));
4437 : : }
4438 : : break;
4439 : : case DatabaseRelationId:
4440 : : {
4441 : : /* LOCKTAG_TUPLE required */
4442 : : Form_pg_database dbForm = (Form_pg_database) GETSTRUCT(newtup);
4443 : :
4444 : : elog(WARNING,
4445 : : "missing lock on database \"%s\" (OID %u) @ TID (%u,%u)",
4446 : : NameStr(dbForm->datname),
4447 : : dbForm->oid,
4448 : : ItemPointerGetBlockNumber(otid),
4449 : : ItemPointerGetOffsetNumber(otid));
4450 : : }
4451 : : break;
4452 : : }
4453 : : }
4454 : :
4455 : : /*
4456 : : * Confirm adequate relation lock held, per rules from README.tuplock section
4457 : : * "Locking to write inplace-updated tables".
4458 : : */
4459 : : static void
4460 : : check_inplace_rel_lock(HeapTuple oldtup)
4461 : : {
4462 : : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(oldtup);
4463 : : Oid relid = classForm->oid;
4464 : : Oid dbid;
4465 : : LOCKTAG tag;
4466 : :
4467 : : if (IsSharedRelation(relid))
4468 : : dbid = InvalidOid;
4469 : : else
4470 : : dbid = MyDatabaseId;
4471 : :
4472 : : if (classForm->relkind == RELKIND_INDEX)
4473 : : {
4474 : : Relation irel = index_open(relid, AccessShareLock);
4475 : :
4476 : : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4477 : : index_close(irel, AccessShareLock);
4478 : : }
4479 : : else
4480 : : SET_LOCKTAG_RELATION(tag, dbid, relid);
4481 : :
4482 : : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, true))
4483 : : elog(WARNING,
4484 : : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4485 : : NameStr(classForm->relname),
4486 : : relid,
4487 : : classForm->relkind,
4488 : : ItemPointerGetBlockNumber(&oldtup->t_self),
4489 : : ItemPointerGetOffsetNumber(&oldtup->t_self));
4490 : : }
4491 : : #endif
4492 : :
4493 : : /*
4494 : : * Check if the specified attribute's values are the same. Subroutine for
4495 : : * HeapDetermineColumnsInfo.
4496 : : */
4497 : : static bool
4498 : 993968 : heap_attr_equals(TupleDesc tupdesc, int attrnum, Datum value1, Datum value2,
4499 : : bool isnull1, bool isnull2)
4500 : : {
4501 : : /*
4502 : : * If one value is NULL and other is not, then they are certainly not
4503 : : * equal
4504 : : */
4505 [ + + ]: 993968 : if (isnull1 != isnull2)
4506 : 60 : return false;
4507 : :
4508 : : /*
4509 : : * If both are NULL, they can be considered equal.
4510 : : */
4511 [ + + ]: 993908 : if (isnull1)
4512 : 6641 : return true;
4513 : :
4514 : : /*
4515 : : * We do simple binary comparison of the two datums. This may be overly
4516 : : * strict because there can be multiple binary representations for the
4517 : : * same logical value. But we should be OK as long as there are no false
4518 : : * positives. Using a type-specific equality operator is messy because
4519 : : * there could be multiple notions of equality in different operator
4520 : : * classes; furthermore, we cannot safely invoke user-defined functions
4521 : : * while holding exclusive buffer lock.
4522 : : */
4523 [ - + ]: 987267 : if (attrnum <= 0)
4524 : : {
4525 : : /* The only allowed system columns are OIDs, so do this */
4526 : 0 : return (DatumGetObjectId(value1) == DatumGetObjectId(value2));
4527 : : }
4528 : : else
4529 : : {
4530 : : CompactAttribute *att;
4531 : :
4532 : : Assert(attrnum <= tupdesc->natts);
4533 : 987267 : att = TupleDescCompactAttr(tupdesc, attrnum - 1);
4534 : 987267 : return datumIsEqual(value1, value2, att->attbyval, att->attlen);
4535 : : }
4536 : : }
4537 : :
4538 : : /*
4539 : : * Check which columns are being updated.
4540 : : *
4541 : : * Given an updated tuple, determine (and return into the output bitmapset),
4542 : : * from those listed as interesting, the set of columns that changed.
4543 : : *
4544 : : * has_external indicates if any of the unmodified attributes (from those
4545 : : * listed as interesting) of the old tuple is a member of external_cols and is
4546 : : * stored externally.
4547 : : */
4548 : : static Bitmapset *
4549 : 2386847 : HeapDetermineColumnsInfo(Relation relation,
4550 : : Bitmapset *interesting_cols,
4551 : : Bitmapset *external_cols,
4552 : : HeapTuple oldtup, HeapTuple newtup,
4553 : : bool *has_external)
4554 : : {
4555 : : int attidx;
4556 : 2386847 : Bitmapset *modified = NULL;
4557 : 2386847 : TupleDesc tupdesc = RelationGetDescr(relation);
4558 : :
4559 : 2386847 : attidx = -1;
4560 [ + + ]: 3380815 : while ((attidx = bms_next_member(interesting_cols, attidx)) >= 0)
4561 : : {
4562 : : /* attidx is zero-based, attrnum is the normal attribute number */
4563 : 993968 : AttrNumber attrnum = attidx + FirstLowInvalidHeapAttributeNumber;
4564 : : Datum value1,
4565 : : value2;
4566 : : bool isnull1,
4567 : : isnull2;
4568 : :
4569 : : /*
4570 : : * If it's a whole-tuple reference, say "not equal". It's not really
4571 : : * worth supporting this case, since it could only succeed after a
4572 : : * no-op update, which is hardly a case worth optimizing for.
4573 : : */
4574 [ - + ]: 993968 : if (attrnum == 0)
4575 : : {
4576 : 0 : modified = bms_add_member(modified, attidx);
4577 : 961535 : continue;
4578 : : }
4579 : :
4580 : : /*
4581 : : * Likewise, automatically say "not equal" for any system attribute
4582 : : * other than tableOID; we cannot expect these to be consistent in a
4583 : : * HOT chain, or even to be set correctly yet in the new tuple.
4584 : : */
4585 [ - + ]: 993968 : if (attrnum < 0)
4586 : : {
4587 [ # # ]: 0 : if (attrnum != TableOidAttributeNumber)
4588 : : {
4589 : 0 : modified = bms_add_member(modified, attidx);
4590 : 0 : continue;
4591 : : }
4592 : : }
4593 : :
4594 : : /*
4595 : : * Extract the corresponding values. XXX this is pretty inefficient
4596 : : * if there are many indexed columns. Should we do a single
4597 : : * heap_deform_tuple call on each tuple, instead? But that doesn't
4598 : : * work for system columns ...
4599 : : */
4600 : 993968 : value1 = heap_getattr(oldtup, attrnum, tupdesc, &isnull1);
4601 : 993968 : value2 = heap_getattr(newtup, attrnum, tupdesc, &isnull2);
4602 : :
4603 [ + + ]: 993968 : if (!heap_attr_equals(tupdesc, attrnum, value1,
4604 : : value2, isnull1, isnull2))
4605 : : {
4606 : 54908 : modified = bms_add_member(modified, attidx);
4607 : 54908 : continue;
4608 : : }
4609 : :
4610 : : /*
4611 : : * No need to check attributes that can't be stored externally. Note
4612 : : * that system attributes can't be stored externally.
4613 : : */
4614 [ + - + + ]: 939060 : if (attrnum < 0 || isnull1 ||
4615 [ + + ]: 932419 : TupleDescCompactAttr(tupdesc, attrnum - 1)->attlen != -1)
4616 : 906627 : continue;
4617 : :
4618 : : /*
4619 : : * Check if the old tuple's attribute is stored externally and is a
4620 : : * member of external_cols.
4621 : : */
4622 [ + + + + ]: 32438 : if (VARATT_IS_EXTERNAL((varlena *) DatumGetPointer(value1)) &&
4623 : 5 : bms_is_member(attidx, external_cols))
4624 : 2 : *has_external = true;
4625 : : }
4626 : :
4627 : 2386847 : return modified;
4628 : : }
4629 : :
4630 : : /*
4631 : : * simple_heap_update - replace a tuple
4632 : : *
4633 : : * This routine may be used to update a tuple when concurrent updates of
4634 : : * the target tuple are not expected (for example, because we have a lock
4635 : : * on the relation associated with the tuple). Any failure is reported
4636 : : * via ereport().
4637 : : */
4638 : : void
4639 : 142334 : simple_heap_update(Relation relation, const ItemPointerData *otid, HeapTuple tup,
4640 : : TU_UpdateIndexes *update_indexes)
4641 : : {
4642 : : TM_Result result;
4643 : : TM_FailureData tmfd;
4644 : : LockTupleMode lockmode;
4645 : :
4646 : 142334 : result = heap_update(relation, otid, tup,
4647 : : GetCurrentCommandId(true), 0,
4648 : : InvalidSnapshot,
4649 : : true /* wait for commit */ ,
4650 : : &tmfd, &lockmode, update_indexes);
4651 [ - + - + : 142334 : switch (result)
- ]
4652 : : {
4653 : 0 : case TM_SelfModified:
4654 : : /* Tuple was already updated in current command? */
4655 [ # # ]: 0 : elog(ERROR, "tuple already updated by self");
4656 : : break;
4657 : :
4658 : 142333 : case TM_Ok:
4659 : : /* done successfully */
4660 : 142333 : break;
4661 : :
4662 : 0 : case TM_Updated:
4663 [ # # ]: 0 : elog(ERROR, "tuple concurrently updated");
4664 : : break;
4665 : :
4666 : 1 : case TM_Deleted:
4667 [ + - ]: 1 : elog(ERROR, "tuple concurrently deleted");
4668 : : break;
4669 : :
4670 : 0 : default:
4671 [ # # ]: 0 : elog(ERROR, "unrecognized heap_update status: %u", result);
4672 : : break;
4673 : : }
4674 : 142333 : }
4675 : :
4676 : :
4677 : : /*
4678 : : * Return the MultiXactStatus corresponding to the given tuple lock mode.
4679 : : */
4680 : : static MultiXactStatus
4681 : 115529 : get_mxact_status_for_lock(LockTupleMode mode, bool is_update)
4682 : : {
4683 : : int retval;
4684 : :
4685 [ + + ]: 115529 : if (is_update)
4686 : 216 : retval = tupleLockExtraInfo[mode].updstatus;
4687 : : else
4688 : 115313 : retval = tupleLockExtraInfo[mode].lockstatus;
4689 : :
4690 [ - + ]: 115529 : if (retval == -1)
4691 [ # # # # ]: 0 : elog(ERROR, "invalid lock tuple mode %d/%s", mode,
4692 : : is_update ? "true" : "false");
4693 : :
4694 : 115529 : return (MultiXactStatus) retval;
4695 : : }
4696 : :
4697 : : /*
4698 : : * heap_lock_tuple - lock a tuple in shared or exclusive mode
4699 : : *
4700 : : * Note that this acquires a buffer pin, which the caller must release.
4701 : : *
4702 : : * Input parameters:
4703 : : * relation: relation containing tuple (caller must hold suitable lock)
4704 : : * cid: current command ID (used for visibility test, and stored into
4705 : : * tuple's cmax if lock is successful)
4706 : : * mode: indicates if shared or exclusive tuple lock is desired
4707 : : * wait_policy: what to do if tuple lock is not available
4708 : : * follow_updates: if true, follow the update chain to also lock descendant
4709 : : * tuples.
4710 : : *
4711 : : * Output parameters:
4712 : : * *tuple: all fields filled in
4713 : : * *buffer: set to buffer holding tuple (pinned but not locked at exit)
4714 : : * *tmfd: filled in failure cases (see below)
4715 : : *
4716 : : * Function results are the same as the ones for table_tuple_lock().
4717 : : *
4718 : : * In the failure cases other than TM_Invisible, the routine fills
4719 : : * *tmfd with the tuple's t_ctid, t_xmax (resolving a possible MultiXact,
4720 : : * if necessary), and t_cmax (the last only for TM_SelfModified,
4721 : : * since we cannot obtain cmax from a combo CID generated by another
4722 : : * transaction).
4723 : : * See comments for struct TM_FailureData for additional info.
4724 : : *
4725 : : * See README.tuplock for a thorough explanation of this mechanism.
4726 : : */
4727 : : TM_Result
4728 : 571732 : heap_lock_tuple(Relation relation, HeapTuple tuple,
4729 : : CommandId cid, LockTupleMode mode, LockWaitPolicy wait_policy,
4730 : : bool follow_updates,
4731 : : Buffer *buffer, TM_FailureData *tmfd)
4732 : : {
4733 : : TM_Result result;
4734 : 571732 : ItemPointer tid = &(tuple->t_self);
4735 : : ItemId lp;
4736 : : Page page;
4737 : 571732 : Buffer vmbuffer = InvalidBuffer;
4738 : 571732 : bool unlock_vmbuffer = false;
4739 : : BlockNumber block;
4740 : : TransactionId xid,
4741 : : xmax;
4742 : : uint16 old_infomask,
4743 : : new_infomask,
4744 : : new_infomask2;
4745 : 571732 : bool first_time = true;
4746 : 571732 : bool skip_tuple_lock = false;
4747 : 571732 : bool have_tuple_lock = false;
4748 : 571732 : bool cleared_all_frozen = false;
4749 : :
4750 : 571732 : *buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
4751 : 571732 : block = ItemPointerGetBlockNumber(tid);
4752 : 571732 : page = BufferGetPage(*buffer);
4753 : :
4754 : : /*
4755 : : * Before locking the buffer, pin the visibility map page if it appears to
4756 : : * be necessary. Since we haven't got the lock yet, someone else might be
4757 : : * in the middle of changing this, so we'll need to recheck after we have
4758 : : * the lock.
4759 : : */
4760 [ + + ]: 571732 : if (PageIsAllVisible(page))
4761 : 413512 : visibilitymap_pin(relation, block, &vmbuffer);
4762 : :
4763 : 571732 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4764 : :
4765 : 571732 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
4766 : : Assert(ItemIdIsNormal(lp));
4767 : :
4768 : 571732 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
4769 : 571732 : tuple->t_len = ItemIdGetLength(lp);
4770 : 571732 : tuple->t_tableOid = RelationGetRelid(relation);
4771 : :
4772 : 16 : l3:
4773 : 571748 : result = HeapTupleSatisfiesUpdate(tuple, cid, *buffer);
4774 : :
4775 [ + + ]: 571748 : if (result == TM_Invisible)
4776 : : {
4777 : : /*
4778 : : * This is possible, but only when locking a tuple for ON CONFLICT DO
4779 : : * SELECT/UPDATE. We return this value here rather than throwing an
4780 : : * error in order to give that case the opportunity to throw a more
4781 : : * specific error.
4782 : : */
4783 : 28 : result = TM_Invisible;
4784 : 28 : goto out_locked;
4785 : : }
4786 [ + + + + ]: 571720 : else if (result == TM_BeingModified ||
4787 [ + + ]: 81063 : result == TM_Updated ||
4788 : : result == TM_Deleted)
4789 : : {
4790 : : TransactionId xwait;
4791 : : uint16 infomask;
4792 : : uint16 infomask2;
4793 : : bool require_sleep;
4794 : : ItemPointerData t_ctid;
4795 : :
4796 : : /* must copy state data before unlocking buffer */
4797 : 490659 : xwait = HeapTupleHeaderGetRawXmax(tuple->t_data);
4798 : 490659 : infomask = tuple->t_data->t_infomask;
4799 : 490659 : infomask2 = tuple->t_data->t_infomask2;
4800 : 490659 : ItemPointerCopy(&tuple->t_data->t_ctid, &t_ctid);
4801 : :
4802 : 490659 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4803 : :
4804 : : /*
4805 : : * If any subtransaction of the current top transaction already holds
4806 : : * a lock as strong as or stronger than what we're requesting, we
4807 : : * effectively hold the desired lock already. We *must* succeed
4808 : : * without trying to take the tuple lock, else we will deadlock
4809 : : * against anyone wanting to acquire a stronger lock.
4810 : : *
4811 : : * Note we only do this the first time we loop on the HTSU result;
4812 : : * there is no point in testing in subsequent passes, because
4813 : : * evidently our own transaction cannot have acquired a new lock after
4814 : : * the first time we checked.
4815 : : */
4816 [ + + ]: 490659 : if (first_time)
4817 : : {
4818 : 490647 : first_time = false;
4819 : :
4820 [ + + ]: 490647 : if (infomask & HEAP_XMAX_IS_MULTI)
4821 : : {
4822 : : int i;
4823 : : int nmembers;
4824 : : MultiXactMember *members;
4825 : :
4826 : : /*
4827 : : * We don't need to allow old multixacts here; if that had
4828 : : * been the case, HeapTupleSatisfiesUpdate would have returned
4829 : : * MayBeUpdated and we wouldn't be here.
4830 : : */
4831 : : nmembers =
4832 : 73307 : GetMultiXactIdMembers(xwait, &members, false,
4833 : 73307 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
4834 : :
4835 [ + + ]: 1422693 : for (i = 0; i < nmembers; i++)
4836 : : {
4837 : : /* only consider members of our own transaction */
4838 [ + + ]: 1349400 : if (!TransactionIdIsCurrentTransactionId(members[i].xid))
4839 : 1349350 : continue;
4840 : :
4841 [ + + ]: 50 : if (TUPLOCK_from_mxstatus(members[i].status) >= mode)
4842 : : {
4843 : 14 : pfree(members);
4844 : 14 : result = TM_Ok;
4845 : 14 : goto out_unlocked;
4846 : : }
4847 : : else
4848 : : {
4849 : : /*
4850 : : * Disable acquisition of the heavyweight tuple lock.
4851 : : * Otherwise, when promoting a weaker lock, we might
4852 : : * deadlock with another locker that has acquired the
4853 : : * heavyweight tuple lock and is waiting for our
4854 : : * transaction to finish.
4855 : : *
4856 : : * Note that in this case we still need to wait for
4857 : : * the multixact if required, to avoid acquiring
4858 : : * conflicting locks.
4859 : : */
4860 : 36 : skip_tuple_lock = true;
4861 : : }
4862 : : }
4863 : :
4864 [ + - ]: 73293 : if (members)
4865 : 73293 : pfree(members);
4866 : : }
4867 [ + + ]: 417340 : else if (TransactionIdIsCurrentTransactionId(xwait))
4868 : : {
4869 [ + + + + : 415918 : switch (mode)
- ]
4870 : : {
4871 : 409468 : case LockTupleKeyShare:
4872 : : Assert(HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) ||
4873 : : HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4874 : : HEAP_XMAX_IS_EXCL_LOCKED(infomask));
4875 : 409468 : result = TM_Ok;
4876 : 409468 : goto out_unlocked;
4877 : 43 : case LockTupleShare:
4878 [ + + - + ]: 49 : if (HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4879 : 6 : HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4880 : : {
4881 : 37 : result = TM_Ok;
4882 : 37 : goto out_unlocked;
4883 : : }
4884 : 6 : break;
4885 : 85 : case LockTupleNoKeyExclusive:
4886 [ + + ]: 85 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4887 : : {
4888 : 72 : result = TM_Ok;
4889 : 72 : goto out_unlocked;
4890 : : }
4891 : 13 : break;
4892 : 6322 : case LockTupleExclusive:
4893 [ + + ]: 6322 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask) &&
4894 [ + + ]: 1280 : infomask2 & HEAP_KEYS_UPDATED)
4895 : : {
4896 : 1251 : result = TM_Ok;
4897 : 1251 : goto out_unlocked;
4898 : : }
4899 : 5071 : break;
4900 : : }
4901 : : }
4902 : : }
4903 : :
4904 : : /*
4905 : : * Initially assume that we will have to wait for the locking
4906 : : * transaction(s) to finish. We check various cases below in which
4907 : : * this can be turned off.
4908 : : */
4909 : 79817 : require_sleep = true;
4910 [ + + ]: 79817 : if (mode == LockTupleKeyShare)
4911 : : {
4912 : : /*
4913 : : * If we're requesting KeyShare, and there's no update present, we
4914 : : * don't need to wait. Even if there is an update, we can still
4915 : : * continue if the key hasn't been modified.
4916 : : *
4917 : : * However, if there are updates, we need to walk the update chain
4918 : : * to mark future versions of the row as locked, too. That way,
4919 : : * if somebody deletes that future version, we're protected
4920 : : * against the key going away. This locking of future versions
4921 : : * could block momentarily, if a concurrent transaction is
4922 : : * deleting a key; or it could return a value to the effect that
4923 : : * the transaction deleting the key has already committed. So we
4924 : : * do this before re-locking the buffer; otherwise this would be
4925 : : * prone to deadlocks.
4926 : : *
4927 : : * Note that the TID we're locking was grabbed before we unlocked
4928 : : * the buffer. For it to change while we're not looking, the
4929 : : * other properties we're testing for below after re-locking the
4930 : : * buffer would also change, in which case we would restart this
4931 : : * loop above.
4932 : : */
4933 [ + + ]: 73904 : if (!(infomask2 & HEAP_KEYS_UPDATED))
4934 : : {
4935 : : bool updated;
4936 : :
4937 : 73857 : updated = !HEAP_XMAX_IS_LOCKED_ONLY(infomask);
4938 : :
4939 : : /*
4940 : : * If there are updates, follow the update chain; bail out if
4941 : : * that cannot be done.
4942 : : */
4943 [ + - + + ]: 73857 : if (follow_updates && updated &&
4944 [ + - ]: 2171 : !ItemPointerEquals(&tuple->t_self, &t_ctid))
4945 : : {
4946 : : TM_Result res;
4947 : :
4948 : 2171 : res = heap_lock_updated_tuple(relation,
4949 : : infomask, xwait, &t_ctid,
4950 : : GetCurrentTransactionId(),
4951 : : mode);
4952 [ + + ]: 2171 : if (res != TM_Ok)
4953 : : {
4954 : 6 : result = res;
4955 : : /* recovery code expects to have buffer lock held */
4956 : 6 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4957 : 208 : goto failed;
4958 : : }
4959 : : }
4960 : :
4961 : 73851 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4962 : :
4963 : : /*
4964 : : * Make sure it's still an appropriate lock, else start over.
4965 : : * Also, if it wasn't updated before we released the lock, but
4966 : : * is updated now, we start over too; the reason is that we
4967 : : * now need to follow the update chain to lock the new
4968 : : * versions.
4969 : : */
4970 [ + + ]: 73851 : if (!HeapTupleHeaderIsOnlyLocked(tuple->t_data) &&
4971 [ + - ]: 2153 : ((tuple->t_data->t_infomask2 & HEAP_KEYS_UPDATED) ||
4972 [ - + ]: 2153 : !updated))
4973 : 16 : goto l3;
4974 : :
4975 : : /* Things look okay, so we can skip sleeping */
4976 : 73851 : require_sleep = false;
4977 : :
4978 : : /*
4979 : : * Note we allow Xmax to change here; other updaters/lockers
4980 : : * could have modified it before we grabbed the buffer lock.
4981 : : * However, this is not a problem, because with the recheck we
4982 : : * just did we ensure that they still don't conflict with the
4983 : : * lock we want.
4984 : : */
4985 : : }
4986 : : }
4987 [ + + ]: 5913 : else if (mode == LockTupleShare)
4988 : : {
4989 : : /*
4990 : : * If we're requesting Share, we can similarly avoid sleeping if
4991 : : * there's no update and no exclusive lock present.
4992 : : */
4993 [ + - ]: 481 : if (HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
4994 [ + + ]: 481 : !HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4995 : : {
4996 : 475 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4997 : :
4998 : : /*
4999 : : * Make sure it's still an appropriate lock, else start over.
5000 : : * See above about allowing xmax to change.
5001 : : */
5002 [ + - - + ]: 950 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
5003 : 475 : HEAP_XMAX_IS_EXCL_LOCKED(tuple->t_data->t_infomask))
5004 : 0 : goto l3;
5005 : 475 : require_sleep = false;
5006 : : }
5007 : : }
5008 [ + + ]: 5432 : else if (mode == LockTupleNoKeyExclusive)
5009 : : {
5010 : : /*
5011 : : * If we're requesting NoKeyExclusive, we might also be able to
5012 : : * avoid sleeping; just ensure that there no conflicting lock
5013 : : * already acquired.
5014 : : */
5015 [ + + ]: 175 : if (infomask & HEAP_XMAX_IS_MULTI)
5016 : : {
5017 [ + + ]: 26 : if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
5018 : : mode, NULL))
5019 : : {
5020 : : /*
5021 : : * No conflict, but if the xmax changed under us in the
5022 : : * meantime, start over.
5023 : : */
5024 : 13 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5025 [ + - - + ]: 26 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
5026 : 13 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
5027 : : xwait))
5028 : 0 : goto l3;
5029 : :
5030 : : /* otherwise, we're good */
5031 : 13 : require_sleep = false;
5032 : : }
5033 : : }
5034 [ + + ]: 149 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
5035 : : {
5036 : 19 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5037 : :
5038 : : /* if the xmax changed in the meantime, start over */
5039 [ + - - + ]: 38 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
5040 : 19 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
5041 : : xwait))
5042 : 0 : goto l3;
5043 : : /* otherwise, we're good */
5044 : 19 : require_sleep = false;
5045 : : }
5046 : : }
5047 : :
5048 : : /*
5049 : : * As a check independent from those above, we can also avoid sleeping
5050 : : * if the current transaction is the sole locker of the tuple. Note
5051 : : * that the strength of the lock already held is irrelevant; this is
5052 : : * not about recording the lock in Xmax (which will be done regardless
5053 : : * of this optimization, below). Also, note that the cases where we
5054 : : * hold a lock stronger than we are requesting are already handled
5055 : : * above by not doing anything.
5056 : : *
5057 : : * Note we only deal with the non-multixact case here; MultiXactIdWait
5058 : : * is well equipped to deal with this situation on its own.
5059 : : */
5060 [ + + + + : 85220 : if (require_sleep && !(infomask & HEAP_XMAX_IS_MULTI) &&
+ + ]
5061 : 5409 : TransactionIdIsCurrentTransactionId(xwait))
5062 : : {
5063 : : /* ... but if the xmax changed in the meantime, start over */
5064 : 5071 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5065 [ + - - + ]: 10142 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
5066 : 5071 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
5067 : : xwait))
5068 : 0 : goto l3;
5069 : : Assert(HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask));
5070 : 5071 : require_sleep = false;
5071 : : }
5072 : :
5073 : : /*
5074 : : * Time to sleep on the other transaction/multixact, if necessary.
5075 : : *
5076 : : * If the other transaction is an update/delete that's already
5077 : : * committed, then sleeping cannot possibly do any good: if we're
5078 : : * required to sleep, get out to raise an error instead.
5079 : : *
5080 : : * By here, we either have already acquired the buffer exclusive lock,
5081 : : * or we must wait for the locking transaction or multixact; so below
5082 : : * we ensure that we grab buffer lock after the sleep.
5083 : : */
5084 [ + + + + : 79811 : if (require_sleep && (result == TM_Updated || result == TM_Deleted))
+ + ]
5085 : : {
5086 : 163 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5087 : 163 : goto failed;
5088 : : }
5089 [ + + ]: 79648 : else if (require_sleep)
5090 : : {
5091 : : /*
5092 : : * Acquire tuple lock to establish our priority for the tuple, or
5093 : : * die trying. LockTuple will release us when we are next-in-line
5094 : : * for the tuple. We must do this even if we are share-locking,
5095 : : * but not if we already have a weaker lock on the tuple.
5096 : : *
5097 : : * If we are forced to "start over" below, we keep the tuple lock;
5098 : : * this arranges that we stay at the head of the line while
5099 : : * rechecking tuple state.
5100 : : */
5101 [ + + ]: 219 : if (!skip_tuple_lock &&
5102 [ + + ]: 202 : !heap_acquire_tuplock(relation, tid, mode, wait_policy,
5103 : : &have_tuple_lock))
5104 : : {
5105 : : /*
5106 : : * This can only happen if wait_policy is Skip and the lock
5107 : : * couldn't be obtained.
5108 : : */
5109 : 1 : result = TM_WouldBlock;
5110 : : /* recovery code expects to have buffer lock held */
5111 : 1 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5112 : 1 : goto failed;
5113 : : }
5114 : :
5115 [ + + ]: 217 : if (infomask & HEAP_XMAX_IS_MULTI)
5116 : : {
5117 : 43 : MultiXactStatus status = get_mxact_status_for_lock(mode, false);
5118 : :
5119 : : /* We only ever lock tuples, never update them */
5120 [ - + ]: 43 : if (status >= MultiXactStatusNoKeyUpdate)
5121 [ # # ]: 0 : elog(ERROR, "invalid lock mode in heap_lock_tuple");
5122 : :
5123 : : /* wait for multixact to end, or die trying */
5124 [ + + + - ]: 43 : switch (wait_policy)
5125 : : {
5126 : 37 : case LockWaitBlock:
5127 : 37 : MultiXactIdWait((MultiXactId) xwait, status, infomask,
5128 : 37 : relation, &tuple->t_self, XLTW_Lock, NULL);
5129 : 37 : break;
5130 : 2 : case LockWaitSkip:
5131 [ + - ]: 2 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5132 : : status, infomask, relation,
5133 : : NULL, false))
5134 : : {
5135 : 2 : result = TM_WouldBlock;
5136 : : /* recovery code expects to have buffer lock held */
5137 : 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5138 : 2 : goto failed;
5139 : : }
5140 : 0 : break;
5141 : 4 : case LockWaitError:
5142 [ + - ]: 4 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5143 : : status, infomask, relation,
5144 : : NULL, log_lock_failures))
5145 [ + - ]: 4 : ereport(ERROR,
5146 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5147 : : errmsg("could not obtain lock on row in relation \"%s\"",
5148 : : RelationGetRelationName(relation))));
5149 : :
5150 : 0 : break;
5151 : : }
5152 : :
5153 : : /*
5154 : : * Of course, the multixact might not be done here: if we're
5155 : : * requesting a light lock mode, other transactions with light
5156 : : * locks could still be alive, as well as locks owned by our
5157 : : * own xact or other subxacts of this backend. We need to
5158 : : * preserve the surviving MultiXact members. Note that it
5159 : : * isn't absolutely necessary in the latter case, but doing so
5160 : : * is simpler.
5161 : : */
5162 : : }
5163 : : else
5164 : : {
5165 : : /* wait for regular transaction to end, or die trying */
5166 [ + + + - ]: 174 : switch (wait_policy)
5167 : : {
5168 : 133 : case LockWaitBlock:
5169 : 133 : XactLockTableWait(xwait, relation, &tuple->t_self,
5170 : : XLTW_Lock);
5171 : 133 : break;
5172 : 33 : case LockWaitSkip:
5173 [ + - ]: 33 : if (!ConditionalXactLockTableWait(xwait, false))
5174 : : {
5175 : 33 : result = TM_WouldBlock;
5176 : : /* recovery code expects to have buffer lock held */
5177 : 33 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5178 : 33 : goto failed;
5179 : : }
5180 : 0 : break;
5181 : 8 : case LockWaitError:
5182 [ + - ]: 8 : if (!ConditionalXactLockTableWait(xwait, log_lock_failures))
5183 [ + - ]: 8 : ereport(ERROR,
5184 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5185 : : errmsg("could not obtain lock on row in relation \"%s\"",
5186 : : RelationGetRelationName(relation))));
5187 : 0 : break;
5188 : : }
5189 : : }
5190 : :
5191 : : /* if there are updates, follow the update chain */
5192 [ + + + + ]: 170 : if (follow_updates && !HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
5193 [ + + ]: 76 : !ItemPointerEquals(&tuple->t_self, &t_ctid))
5194 : : {
5195 : : TM_Result res;
5196 : :
5197 : 55 : res = heap_lock_updated_tuple(relation,
5198 : : infomask, xwait, &t_ctid,
5199 : : GetCurrentTransactionId(),
5200 : : mode);
5201 [ + + ]: 55 : if (res != TM_Ok)
5202 : : {
5203 : 3 : result = res;
5204 : : /* recovery code expects to have buffer lock held */
5205 : 3 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5206 : 3 : goto failed;
5207 : : }
5208 : : }
5209 : :
5210 : 167 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5211 : :
5212 : : /*
5213 : : * xwait is done, but if xwait had just locked the tuple then some
5214 : : * other xact could update this tuple before we get to this point.
5215 : : * Check for xmax change, and start over if so.
5216 : : */
5217 [ + + + + ]: 319 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
5218 : 152 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
5219 : : xwait))
5220 : 16 : goto l3;
5221 : :
5222 [ + + ]: 151 : if (!(infomask & HEAP_XMAX_IS_MULTI))
5223 : : {
5224 : : /*
5225 : : * Otherwise check if it committed or aborted. Note we cannot
5226 : : * be here if the tuple was only locked by somebody who didn't
5227 : : * conflict with us; that would have been handled above. So
5228 : : * that transaction must necessarily be gone by now. But
5229 : : * don't check for this in the multixact case, because some
5230 : : * locker transactions might still be running.
5231 : : */
5232 : 117 : UpdateXmaxHintBits(tuple->t_data, *buffer, xwait);
5233 : : }
5234 : : }
5235 : :
5236 : : /* By here, we're certain that we hold buffer exclusive lock again */
5237 : :
5238 : : /*
5239 : : * We may lock if previous xmax aborted, or if it committed but only
5240 : : * locked the tuple without updating it; or if we didn't have to wait
5241 : : * at all for whatever reason.
5242 : : */
5243 [ + + ]: 79580 : if (!require_sleep ||
5244 [ + + + + ]: 266 : (tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
5245 [ + + ]: 215 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
5246 : 100 : HeapTupleHeaderIsOnlyLocked(tuple->t_data))
5247 : 79489 : result = TM_Ok;
5248 [ + + ]: 91 : else if (!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
5249 : 66 : result = TM_Updated;
5250 : : else
5251 : 25 : result = TM_Deleted;
5252 : : }
5253 : :
5254 : 81061 : failed:
5255 [ + + ]: 160849 : if (result != TM_Ok)
5256 : : {
5257 : : Assert(result == TM_SelfModified || result == TM_Updated ||
5258 : : result == TM_Deleted || result == TM_WouldBlock);
5259 : :
5260 : : /*
5261 : : * When locking a tuple under LockWaitSkip semantics and we fail with
5262 : : * TM_WouldBlock above, it's possible for concurrent transactions to
5263 : : * release the lock and set HEAP_XMAX_INVALID in the meantime. So
5264 : : * this assert is slightly different from the equivalent one in
5265 : : * heap_delete and heap_update.
5266 : : */
5267 : : Assert((result == TM_WouldBlock) ||
5268 : : !(tuple->t_data->t_infomask & HEAP_XMAX_INVALID));
5269 : : Assert(result != TM_Updated ||
5270 : : !ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid));
5271 : 307 : tmfd->ctid = tuple->t_data->t_ctid;
5272 : 307 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
5273 [ + + ]: 307 : if (result == TM_SelfModified)
5274 : 8 : tmfd->cmax = HeapTupleHeaderGetCmax(tuple->t_data);
5275 : : else
5276 : 299 : tmfd->cmax = InvalidCommandId;
5277 : 307 : goto out_locked;
5278 : : }
5279 : :
5280 : : /*
5281 : : * If we didn't pin the visibility map page and the page has become all
5282 : : * visible while we were busy locking the buffer, or during some
5283 : : * subsequent window during which we had it unlocked, we'll have to unlock
5284 : : * and re-lock, to avoid holding the buffer lock across I/O. That's a bit
5285 : : * unfortunate, especially since we'll now have to recheck whether the
5286 : : * tuple has been locked or updated under us, but hopefully it won't
5287 : : * happen very often.
5288 : : */
5289 [ + + - + ]: 160542 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
5290 : : {
5291 : 0 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5292 : 0 : visibilitymap_pin(relation, block, &vmbuffer);
5293 : 0 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5294 : 0 : goto l3;
5295 : : }
5296 : :
5297 : 160542 : xmax = HeapTupleHeaderGetRawXmax(tuple->t_data);
5298 : 160542 : old_infomask = tuple->t_data->t_infomask;
5299 : :
5300 : : /*
5301 : : * If this is the first possibly-multixact-able operation in the current
5302 : : * transaction, set my per-backend OldestMemberMXactId setting. We can be
5303 : : * certain that the transaction will never become a member of any older
5304 : : * MultiXactIds than that. (We have to do this even if we end up just
5305 : : * using our own TransactionId below, since some other backend could
5306 : : * incorporate our XID into a MultiXact immediately afterwards.)
5307 : : */
5308 : 160542 : MultiXactIdSetOldestMember();
5309 : :
5310 : : /*
5311 : : * Compute the new xmax and infomask to store into the tuple. Note we do
5312 : : * not modify the tuple just yet, because that would leave it in the wrong
5313 : : * state if multixact.c elogs.
5314 : : */
5315 : 160542 : compute_new_xmax_infomask(xmax, old_infomask, tuple->t_data->t_infomask2,
5316 : : GetCurrentTransactionId(), mode, false,
5317 : : &xid, &new_infomask, &new_infomask2);
5318 : :
5319 : : /* Lock VM buffer before entering critical section */
5320 [ + + ]: 160542 : if (PageIsAllVisible(page))
5321 : : {
5322 : 3615 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
5323 : 3615 : unlock_vmbuffer = true;
5324 : : }
5325 : :
5326 : 160542 : START_CRIT_SECTION();
5327 : :
5328 : : /*
5329 : : * Store transaction information of xact locking the tuple.
5330 : : *
5331 : : * Note: Cmax is meaningless in this context, so don't set it; this avoids
5332 : : * possibly generating a useless combo CID. Moreover, if we're locking a
5333 : : * previously updated tuple, it's important to preserve the Cmax.
5334 : : *
5335 : : * Also reset the HOT UPDATE bit, but only if there's no update; otherwise
5336 : : * we would break the HOT chain.
5337 : : */
5338 : 160542 : tuple->t_data->t_infomask &= ~HEAP_XMAX_BITS;
5339 : 160542 : tuple->t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5340 : 160542 : tuple->t_data->t_infomask |= new_infomask;
5341 : 160542 : tuple->t_data->t_infomask2 |= new_infomask2;
5342 [ + + ]: 160542 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5343 : 158393 : HeapTupleHeaderClearHotUpdated(tuple->t_data);
5344 : 160542 : HeapTupleHeaderSetXmax(tuple->t_data, xid);
5345 : :
5346 : : /*
5347 : : * Make sure there is no forward chain link in t_ctid. Note that in the
5348 : : * cases where the tuple has been updated, we must not overwrite t_ctid,
5349 : : * because it was set by the updater. Moreover, if the tuple has been
5350 : : * updated, we need to follow the update chain to lock the new versions of
5351 : : * the tuple as well.
5352 : : */
5353 [ + + ]: 160542 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5354 : 158393 : tuple->t_data->t_ctid = *tid;
5355 : :
5356 : : /* Clear only the all-frozen bit on visibility map if needed */
5357 [ + + ]: 160542 : if (PageIsAllVisible(page))
5358 : : {
5359 [ + + ]: 3615 : if (visibilitymap_clear(relation->rd_locator, block, vmbuffer,
5360 : : VISIBILITYMAP_ALL_FROZEN))
5361 : 18 : cleared_all_frozen = true;
5362 : : }
5363 : :
5364 : :
5365 : 160542 : MarkBufferDirty(*buffer);
5366 : :
5367 : : /*
5368 : : * XLOG stuff. You might think that we don't need an XLOG record because
5369 : : * there is no state change worth restoring after a crash. You would be
5370 : : * wrong however: we have just written either a TransactionId or a
5371 : : * MultiXactId that may never have been seen on disk before, and we need
5372 : : * to make sure that there are XLOG entries covering those ID numbers.
5373 : : * Else the same IDs might be re-used after a crash, which would be
5374 : : * disastrous if this page made it to disk before the crash. Essentially
5375 : : * we have to enforce the WAL log-before-data rule even in this case.
5376 : : * (Also, in a PITR log-shipping or 2PC environment, we have to have XLOG
5377 : : * entries for everything anyway.)
5378 : : */
5379 [ + + + + : 160542 : if (RelationNeedsWAL(relation))
+ - + - ]
5380 : : {
5381 : : xl_heap_lock xlrec;
5382 : : XLogRecPtr recptr;
5383 : :
5384 : 160113 : XLogBeginInsert();
5385 : 160113 : XLogRegisterBuffer(HEAP_LOCK_BLKREF_HEAP, *buffer, REGBUF_STANDARD);
5386 : :
5387 : 160113 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
5388 : 160113 : xlrec.xmax = xid;
5389 : 320226 : xlrec.infobits_set = compute_infobits(new_infomask,
5390 : 160113 : tuple->t_data->t_infomask2);
5391 : 160113 : xlrec.flags = cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
5392 : 160113 : XLogRegisterData(&xlrec, SizeOfHeapLock);
5393 : :
5394 [ + + ]: 160113 : if (cleared_all_frozen)
5395 : 18 : XLogRegisterBuffer(HEAP_LOCK_BLKREF_VM, vmbuffer, 0);
5396 : :
5397 : : /* we don't decode row locks atm, so no need to log the origin */
5398 : :
5399 : 160113 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
5400 : :
5401 : 160113 : PageSetLSN(page, recptr);
5402 : :
5403 [ + + ]: 160113 : if (cleared_all_frozen)
5404 : 18 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
5405 : : }
5406 : :
5407 : 160542 : END_CRIT_SECTION();
5408 : :
5409 : : /* release VM lock first, since it covers many heap blocks */
5410 [ + + ]: 160542 : if (unlock_vmbuffer)
5411 : : {
5412 : 3615 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
5413 : 3615 : unlock_vmbuffer = false;
5414 : : }
5415 : :
5416 : 160542 : result = TM_Ok;
5417 : :
5418 : 160877 : out_locked:
5419 : 160877 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5420 : : Assert(!unlock_vmbuffer);
5421 : :
5422 : 571719 : out_unlocked:
5423 [ + + ]: 571719 : if (BufferIsValid(vmbuffer))
5424 : 413512 : ReleaseBuffer(vmbuffer);
5425 : :
5426 : : /*
5427 : : * Don't update the visibility map here. Locking a tuple doesn't change
5428 : : * visibility info.
5429 : : */
5430 : :
5431 : : /*
5432 : : * Now that we have successfully marked the tuple as locked, we can
5433 : : * release the lmgr tuple lock, if we had it.
5434 : : */
5435 [ + + ]: 571719 : if (have_tuple_lock)
5436 : 183 : UnlockTupleTuplock(relation, tid, mode);
5437 : :
5438 : 571719 : return result;
5439 : : }
5440 : :
5441 : : /*
5442 : : * Acquire heavyweight lock on the given tuple, in preparation for acquiring
5443 : : * its normal, Xmax-based tuple lock.
5444 : : *
5445 : : * have_tuple_lock is an input and output parameter: on input, it indicates
5446 : : * whether the lock has previously been acquired (and this function does
5447 : : * nothing in that case). If this function returns success, have_tuple_lock
5448 : : * has been flipped to true.
5449 : : *
5450 : : * Returns false if it was unable to obtain the lock; this can only happen if
5451 : : * wait_policy is Skip.
5452 : : */
5453 : : static bool
5454 : 378 : heap_acquire_tuplock(Relation relation, const ItemPointerData *tid, LockTupleMode mode,
5455 : : LockWaitPolicy wait_policy, bool *have_tuple_lock)
5456 : : {
5457 [ + + ]: 378 : if (*have_tuple_lock)
5458 : 9 : return true;
5459 : :
5460 [ + + + - ]: 369 : switch (wait_policy)
5461 : : {
5462 : 324 : case LockWaitBlock:
5463 : 324 : LockTupleTuplock(relation, tid, mode);
5464 : 324 : break;
5465 : :
5466 : 34 : case LockWaitSkip:
5467 [ + + ]: 34 : if (!ConditionalLockTupleTuplock(relation, tid, mode, false))
5468 : 1 : return false;
5469 : 33 : break;
5470 : :
5471 : 11 : case LockWaitError:
5472 [ + + ]: 11 : if (!ConditionalLockTupleTuplock(relation, tid, mode, log_lock_failures))
5473 [ + - ]: 1 : ereport(ERROR,
5474 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5475 : : errmsg("could not obtain lock on row in relation \"%s\"",
5476 : : RelationGetRelationName(relation))));
5477 : 10 : break;
5478 : : }
5479 : 367 : *have_tuple_lock = true;
5480 : :
5481 : 367 : return true;
5482 : : }
5483 : :
5484 : : /*
5485 : : * Given an original set of Xmax and infomask, and a transaction (identified by
5486 : : * add_to_xmax) acquiring a new lock of some mode, compute the new Xmax and
5487 : : * corresponding infomasks to use on the tuple.
5488 : : *
5489 : : * Note that this might have side effects such as creating a new MultiXactId.
5490 : : *
5491 : : * Most callers will have called HeapTupleSatisfiesUpdate before this function;
5492 : : * that will have set the HEAP_XMAX_INVALID bit if the xmax was a MultiXactId
5493 : : * but it was not running anymore. There is a race condition, which is that the
5494 : : * MultiXactId may have finished since then, but that uncommon case is handled
5495 : : * either here, or within MultiXactIdExpand.
5496 : : *
5497 : : * There is a similar race condition possible when the old xmax was a regular
5498 : : * TransactionId. We test TransactionIdIsInProgress again just to narrow the
5499 : : * window, but it's still possible to end up creating an unnecessary
5500 : : * MultiXactId. Fortunately this is harmless.
5501 : : */
5502 : : static void
5503 : 6671108 : compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
5504 : : uint16 old_infomask2, TransactionId add_to_xmax,
5505 : : LockTupleMode mode, bool is_update,
5506 : : TransactionId *result_xmax, uint16 *result_infomask,
5507 : : uint16 *result_infomask2)
5508 : : {
5509 : : TransactionId new_xmax;
5510 : : uint16 new_infomask,
5511 : : new_infomask2;
5512 : :
5513 : : Assert(TransactionIdIsCurrentTransactionId(add_to_xmax));
5514 : :
5515 : 104549 : l5:
5516 : 6775657 : new_infomask = 0;
5517 : 6775657 : new_infomask2 = 0;
5518 [ + + ]: 6775657 : if (old_infomask & HEAP_XMAX_INVALID)
5519 : : {
5520 : : /*
5521 : : * No previous locker; we just insert our own TransactionId.
5522 : : *
5523 : : * Note that it's critical that this case be the first one checked,
5524 : : * because there are several blocks below that come back to this one
5525 : : * to implement certain optimizations; old_infomask might contain
5526 : : * other dirty bits in those cases, but we don't really care.
5527 : : */
5528 [ + + ]: 6594403 : if (is_update)
5529 : : {
5530 : 4310020 : new_xmax = add_to_xmax;
5531 [ + + ]: 4310020 : if (mode == LockTupleExclusive)
5532 : 1961597 : new_infomask2 |= HEAP_KEYS_UPDATED;
5533 : : }
5534 : : else
5535 : : {
5536 : 2284383 : new_infomask |= HEAP_XMAX_LOCK_ONLY;
5537 [ + + + + : 2284383 : switch (mode)
- ]
5538 : : {
5539 : 5812 : case LockTupleKeyShare:
5540 : 5812 : new_xmax = add_to_xmax;
5541 : 5812 : new_infomask |= HEAP_XMAX_KEYSHR_LOCK;
5542 : 5812 : break;
5543 : 817 : case LockTupleShare:
5544 : 817 : new_xmax = add_to_xmax;
5545 : 817 : new_infomask |= HEAP_XMAX_SHR_LOCK;
5546 : 817 : break;
5547 : 2181706 : case LockTupleNoKeyExclusive:
5548 : 2181706 : new_xmax = add_to_xmax;
5549 : 2181706 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5550 : 2181706 : break;
5551 : 96048 : case LockTupleExclusive:
5552 : 96048 : new_xmax = add_to_xmax;
5553 : 96048 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5554 : 96048 : new_infomask2 |= HEAP_KEYS_UPDATED;
5555 : 96048 : break;
5556 : 0 : default:
5557 : 0 : new_xmax = InvalidTransactionId; /* silence compiler */
5558 [ # # ]: 0 : elog(ERROR, "invalid lock mode");
5559 : : }
5560 : : }
5561 : : }
5562 [ + + ]: 181254 : else if (old_infomask & HEAP_XMAX_IS_MULTI)
5563 : : {
5564 : : MultiXactStatus new_status;
5565 : :
5566 : : /*
5567 : : * Currently we don't allow XMAX_COMMITTED to be set for multis, so
5568 : : * cross-check.
5569 : : */
5570 : : Assert(!(old_infomask & HEAP_XMAX_COMMITTED));
5571 : :
5572 : : /*
5573 : : * A multixact together with LOCK_ONLY set but neither lock bit set
5574 : : * (i.e. a pg_upgraded share locked tuple) cannot possibly be running
5575 : : * anymore. This check is critical for databases upgraded by
5576 : : * pg_upgrade; both MultiXactIdIsRunning and MultiXactIdExpand assume
5577 : : * that such multis are never passed.
5578 : : */
5579 [ - + ]: 75570 : if (HEAP_LOCKED_UPGRADED(old_infomask))
5580 : : {
5581 : 0 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5582 : 0 : old_infomask |= HEAP_XMAX_INVALID;
5583 : 0 : goto l5;
5584 : : }
5585 : :
5586 : : /*
5587 : : * If the XMAX is already a MultiXactId, then we need to expand it to
5588 : : * include add_to_xmax; but if all the members were lockers and are
5589 : : * all gone, we can do away with the IS_MULTI bit and just set
5590 : : * add_to_xmax as the only locker/updater. If all lockers are gone
5591 : : * and we have an updater that aborted, we can also do without a
5592 : : * multi.
5593 : : *
5594 : : * The cost of doing GetMultiXactIdMembers would be paid by
5595 : : * MultiXactIdExpand if we weren't to do this, so this check is not
5596 : : * incurring extra work anyhow.
5597 : : */
5598 [ + + ]: 75570 : if (!MultiXactIdIsRunning(xmax, HEAP_XMAX_IS_LOCKED_ONLY(old_infomask)))
5599 : : {
5600 [ + + ]: 25 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) ||
5601 [ + - ]: 10 : !TransactionIdDidCommit(MultiXactIdGetUpdateXid(xmax,
5602 : : old_infomask)))
5603 : : {
5604 : : /*
5605 : : * Reset these bits and restart; otherwise fall through to
5606 : : * create a new multi below.
5607 : : */
5608 : 25 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5609 : 25 : old_infomask |= HEAP_XMAX_INVALID;
5610 : 25 : goto l5;
5611 : : }
5612 : : }
5613 : :
5614 : 75545 : new_status = get_mxact_status_for_lock(mode, is_update);
5615 : :
5616 : 75545 : new_xmax = MultiXactIdExpand((MultiXactId) xmax, add_to_xmax,
5617 : : new_status);
5618 : 75545 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5619 : : }
5620 [ + + ]: 105684 : else if (old_infomask & HEAP_XMAX_COMMITTED)
5621 : : {
5622 : : /*
5623 : : * It's a committed update, so we need to preserve him as updater of
5624 : : * the tuple.
5625 : : */
5626 : : MultiXactStatus status;
5627 : : MultiXactStatus new_status;
5628 : :
5629 [ - + ]: 13 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5630 : 0 : status = MultiXactStatusUpdate;
5631 : : else
5632 : 13 : status = MultiXactStatusNoKeyUpdate;
5633 : :
5634 : 13 : new_status = get_mxact_status_for_lock(mode, is_update);
5635 : :
5636 : : /*
5637 : : * since it's not running, it's obviously impossible for the old
5638 : : * updater to be identical to the current one, so we need not check
5639 : : * for that case as we do in the block above.
5640 : : */
5641 : 13 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5642 : 13 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5643 : : }
5644 [ + + ]: 105671 : else if (TransactionIdIsInProgress(xmax))
5645 : : {
5646 : : /*
5647 : : * If the XMAX is a valid, in-progress TransactionId, then we need to
5648 : : * create a new MultiXactId that includes both the old locker or
5649 : : * updater and our own TransactionId.
5650 : : */
5651 : : MultiXactStatus new_status;
5652 : : MultiXactStatus old_status;
5653 : : LockTupleMode old_mode;
5654 : :
5655 [ + + ]: 105656 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5656 : : {
5657 [ + + ]: 105628 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5658 : 5721 : old_status = MultiXactStatusForKeyShare;
5659 [ + + ]: 99907 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5660 : 472 : old_status = MultiXactStatusForShare;
5661 [ + - ]: 99435 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5662 : : {
5663 [ + + ]: 99435 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5664 : 92939 : old_status = MultiXactStatusForUpdate;
5665 : : else
5666 : 6496 : old_status = MultiXactStatusForNoKeyUpdate;
5667 : : }
5668 : : else
5669 : : {
5670 : : /*
5671 : : * LOCK_ONLY can be present alone only when a page has been
5672 : : * upgraded by pg_upgrade. But in that case,
5673 : : * TransactionIdIsInProgress() should have returned false. We
5674 : : * assume it's no longer locked in this case.
5675 : : */
5676 [ # # ]: 0 : elog(WARNING, "LOCK_ONLY found for Xid in progress %u", xmax);
5677 : 0 : old_infomask |= HEAP_XMAX_INVALID;
5678 : 0 : old_infomask &= ~HEAP_XMAX_LOCK_ONLY;
5679 : 0 : goto l5;
5680 : : }
5681 : : }
5682 : : else
5683 : : {
5684 : : /* it's an update, but which kind? */
5685 [ - + ]: 28 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5686 : 0 : old_status = MultiXactStatusUpdate;
5687 : : else
5688 : 28 : old_status = MultiXactStatusNoKeyUpdate;
5689 : : }
5690 : :
5691 : 105656 : old_mode = TUPLOCK_from_mxstatus(old_status);
5692 : :
5693 : : /*
5694 : : * If the lock to be acquired is for the same TransactionId as the
5695 : : * existing lock, there's an optimization possible: consider only the
5696 : : * strongest of both locks as the only one present, and restart.
5697 : : */
5698 [ + + ]: 105656 : if (xmax == add_to_xmax)
5699 : : {
5700 : : /*
5701 : : * Note that it's not possible for the original tuple to be
5702 : : * updated: we wouldn't be here because the tuple would have been
5703 : : * invisible and we wouldn't try to update it. As a subtlety,
5704 : : * this code can also run when traversing an update chain to lock
5705 : : * future versions of a tuple. But we wouldn't be here either,
5706 : : * because the add_to_xmax would be different from the original
5707 : : * updater.
5708 : : */
5709 : : Assert(HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5710 : :
5711 : : /* acquire the strongest of both */
5712 [ + + ]: 104510 : if (mode < old_mode)
5713 : 52236 : mode = old_mode;
5714 : : /* mustn't touch is_update */
5715 : :
5716 : 104510 : old_infomask |= HEAP_XMAX_INVALID;
5717 : 104510 : goto l5;
5718 : : }
5719 : :
5720 : : /* otherwise, just fall back to creating a new multixact */
5721 : 1146 : new_status = get_mxact_status_for_lock(mode, is_update);
5722 : 1146 : new_xmax = MultiXactIdCreate(xmax, old_status,
5723 : : add_to_xmax, new_status);
5724 : 1146 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5725 : : }
5726 [ + + + + ]: 20 : else if (!HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) &&
5727 : 5 : TransactionIdDidCommit(xmax))
5728 : 1 : {
5729 : : /*
5730 : : * It's a committed update, so we gotta preserve him as updater of the
5731 : : * tuple.
5732 : : */
5733 : : MultiXactStatus status;
5734 : : MultiXactStatus new_status;
5735 : :
5736 [ - + ]: 1 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5737 : 0 : status = MultiXactStatusUpdate;
5738 : : else
5739 : 1 : status = MultiXactStatusNoKeyUpdate;
5740 : :
5741 : 1 : new_status = get_mxact_status_for_lock(mode, is_update);
5742 : :
5743 : : /*
5744 : : * since it's not running, it's obviously impossible for the old
5745 : : * updater to be identical to the current one, so we need not check
5746 : : * for that case as we do in the block above.
5747 : : */
5748 : 1 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5749 : 1 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5750 : : }
5751 : : else
5752 : : {
5753 : : /*
5754 : : * Can get here iff the locking/updating transaction was running when
5755 : : * the infomask was extracted from the tuple, but finished before
5756 : : * TransactionIdIsInProgress got to run. Deal with it as if there was
5757 : : * no locker at all in the first place.
5758 : : */
5759 : 14 : old_infomask |= HEAP_XMAX_INVALID;
5760 : 14 : goto l5;
5761 : : }
5762 : :
5763 : 6671108 : *result_infomask = new_infomask;
5764 : 6671108 : *result_infomask2 = new_infomask2;
5765 : 6671108 : *result_xmax = new_xmax;
5766 : 6671108 : }
5767 : :
5768 : : /*
5769 : : * Subroutine for heap_lock_updated_tuple_rec.
5770 : : *
5771 : : * Given a hypothetical multixact status held by the transaction identified
5772 : : * with the given xid, does the current transaction need to wait, fail, or can
5773 : : * it continue if it wanted to acquire a lock of the given mode? "needwait"
5774 : : * is set to true if waiting is necessary; if it can continue, then TM_Ok is
5775 : : * returned. If the lock is already held by the current transaction, return
5776 : : * TM_SelfModified. In case of a conflict with another transaction, a
5777 : : * different HeapTupleSatisfiesUpdate return code is returned.
5778 : : *
5779 : : * The held status is said to be hypothetical because it might correspond to a
5780 : : * lock held by a single Xid, i.e. not a real MultiXactId; we express it this
5781 : : * way for simplicity of API.
5782 : : */
5783 : : static TM_Result
5784 : 38781 : test_lockmode_for_conflict(MultiXactStatus status, TransactionId xid,
5785 : : LockTupleMode mode, HeapTuple tup,
5786 : : bool *needwait)
5787 : : {
5788 : : MultiXactStatus wantedstatus;
5789 : :
5790 : 38781 : *needwait = false;
5791 : 38781 : wantedstatus = get_mxact_status_for_lock(mode, false);
5792 : :
5793 : : /*
5794 : : * Note: we *must* check TransactionIdIsInProgress before
5795 : : * TransactionIdDidAbort/Commit; see comment at top of heapam_visibility.c
5796 : : * for an explanation.
5797 : : */
5798 [ - + ]: 38781 : if (TransactionIdIsCurrentTransactionId(xid))
5799 : : {
5800 : : /*
5801 : : * The tuple has already been locked by our own transaction. This is
5802 : : * very rare but can happen if multiple transactions are trying to
5803 : : * lock an ancient version of the same tuple.
5804 : : */
5805 : 0 : return TM_SelfModified;
5806 : : }
5807 [ + + ]: 38781 : else if (TransactionIdIsInProgress(xid))
5808 : : {
5809 : : /*
5810 : : * If the locking transaction is running, what we do depends on
5811 : : * whether the lock modes conflict: if they do, then we must wait for
5812 : : * it to finish; otherwise we can fall through to lock this tuple
5813 : : * version without waiting.
5814 : : */
5815 [ + + ]: 36539 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5816 : 36539 : LOCKMODE_from_mxstatus(wantedstatus)))
5817 : : {
5818 : 8 : *needwait = true;
5819 : : }
5820 : :
5821 : : /*
5822 : : * If we set needwait above, then this value doesn't matter;
5823 : : * otherwise, this value signals to caller that it's okay to proceed.
5824 : : */
5825 : 36539 : return TM_Ok;
5826 : : }
5827 [ + + ]: 2242 : else if (TransactionIdDidAbort(xid))
5828 : 206 : return TM_Ok;
5829 [ + - ]: 2036 : else if (TransactionIdDidCommit(xid))
5830 : : {
5831 : : /*
5832 : : * The other transaction committed. If it was only a locker, then the
5833 : : * lock is completely gone now and we can return success; but if it
5834 : : * was an update, then what we do depends on whether the two lock
5835 : : * modes conflict. If they conflict, then we must report error to
5836 : : * caller. But if they don't, we can fall through to allow the current
5837 : : * transaction to lock the tuple.
5838 : : *
5839 : : * Note: the reason we worry about ISUPDATE here is because as soon as
5840 : : * a transaction ends, all its locks are gone and meaningless, and
5841 : : * thus we can ignore them; whereas its updates persist. In the
5842 : : * TransactionIdIsInProgress case, above, we don't need to check
5843 : : * because we know the lock is still "alive" and thus a conflict needs
5844 : : * always be checked.
5845 : : */
5846 [ + + ]: 2036 : if (!ISUPDATE_from_mxstatus(status))
5847 : 2026 : return TM_Ok;
5848 : :
5849 [ + + ]: 10 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5850 : 10 : LOCKMODE_from_mxstatus(wantedstatus)))
5851 : : {
5852 : : /* bummer */
5853 [ + + ]: 9 : if (!ItemPointerEquals(&tup->t_self, &tup->t_data->t_ctid))
5854 : 7 : return TM_Updated;
5855 : : else
5856 : 2 : return TM_Deleted;
5857 : : }
5858 : :
5859 : 1 : return TM_Ok;
5860 : : }
5861 : :
5862 : : /* Not in progress, not aborted, not committed -- must have crashed */
5863 : 0 : return TM_Ok;
5864 : : }
5865 : :
5866 : :
5867 : : /*
5868 : : * Recursive part of heap_lock_updated_tuple
5869 : : *
5870 : : * Fetch the tuple pointed to by tid in rel, and mark it as locked by the given
5871 : : * xid with the given mode; if this tuple is updated, recurse to lock the new
5872 : : * version as well.
5873 : : */
5874 : : static TM_Result
5875 : 2224 : heap_lock_updated_tuple_rec(Relation rel, TransactionId priorXmax,
5876 : : const ItemPointerData *tid, TransactionId xid,
5877 : : LockTupleMode mode)
5878 : : {
5879 : : TM_Result result;
5880 : : ItemPointerData tupid;
5881 : : HeapTupleData mytup;
5882 : : Buffer buf;
5883 : : Page page;
5884 : : uint16 new_infomask,
5885 : : new_infomask2,
5886 : : old_infomask,
5887 : : old_infomask2;
5888 : : TransactionId xmax,
5889 : : new_xmax;
5890 : 2224 : bool cleared_all_frozen = false;
5891 : : bool pinned_desired_page;
5892 : 2224 : Buffer vmbuffer = InvalidBuffer;
5893 : 2224 : bool unlock_vmbuffer = false;
5894 : : BlockNumber block;
5895 : :
5896 : 2224 : ItemPointerCopy(tid, &tupid);
5897 : :
5898 : : for (;;)
5899 : : {
5900 : 2227 : new_infomask = 0;
5901 : 2227 : new_xmax = InvalidTransactionId;
5902 : 2227 : cleared_all_frozen = false;
5903 : 2227 : block = ItemPointerGetBlockNumber(&tupid);
5904 : 2227 : ItemPointerCopy(&tupid, &(mytup.t_self));
5905 : :
5906 [ + - ]: 2227 : if (!heap_fetch(rel, SnapshotAny, &mytup, &buf, false))
5907 : : {
5908 : : /*
5909 : : * if we fail to find the updated version of the tuple, it's
5910 : : * because it was vacuumed/pruned away after its creator
5911 : : * transaction aborted. So behave as if we got to the end of the
5912 : : * chain, and there's no further tuple to lock: return success to
5913 : : * caller.
5914 : : */
5915 : 0 : result = TM_Ok;
5916 : 0 : goto out_unlocked;
5917 : : }
5918 : :
5919 : 2227 : l4:
5920 [ - + ]: 2235 : CHECK_FOR_INTERRUPTS();
5921 : :
5922 : 2235 : page = BufferGetPage(buf);
5923 : :
5924 : : /*
5925 : : * Before locking the buffer, pin the visibility map page if it
5926 : : * appears to be necessary. Since we haven't got the lock yet,
5927 : : * someone else might be in the middle of changing this, so we'll need
5928 : : * to recheck after we have the lock.
5929 : : */
5930 [ - + ]: 2235 : if (PageIsAllVisible(page))
5931 : : {
5932 : 0 : visibilitymap_pin(rel, block, &vmbuffer);
5933 : 0 : pinned_desired_page = true;
5934 : : }
5935 : : else
5936 : 2235 : pinned_desired_page = false;
5937 : :
5938 : 2235 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5939 : :
5940 : : /*
5941 : : * If we didn't pin the visibility map page and the page has become
5942 : : * all visible while we were busy locking the buffer, we'll have to
5943 : : * unlock and re-lock, to avoid holding the buffer lock across I/O.
5944 : : * That's a bit unfortunate, but hopefully shouldn't happen often.
5945 : : *
5946 : : * Note: in some paths through this function, we will reach here
5947 : : * holding a pin on a vm page that may or may not be the one matching
5948 : : * this page. If this page isn't all-visible, we won't use the vm
5949 : : * page, but we hold onto such a pin till the end of the function.
5950 : : */
5951 [ + - - + ]: 2235 : if (!pinned_desired_page && PageIsAllVisible(page))
5952 : : {
5953 : 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5954 : 0 : visibilitymap_pin(rel, block, &vmbuffer);
5955 : 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5956 : : }
5957 : :
5958 : : /*
5959 : : * Check the tuple XMIN against prior XMAX, if any. If we reached the
5960 : : * end of the chain, we're done, so return success.
5961 : : */
5962 [ + - + + ]: 4470 : if (TransactionIdIsValid(priorXmax) &&
5963 : 2235 : !TransactionIdEquals(HeapTupleHeaderGetXmin(mytup.t_data),
5964 : : priorXmax))
5965 : : {
5966 : 2 : result = TM_Ok;
5967 : 2 : goto out_locked;
5968 : : }
5969 : :
5970 : : /*
5971 : : * Also check Xmin: if this tuple was created by an aborted
5972 : : * (sub)transaction, then we already locked the last live one in the
5973 : : * chain, thus we're done, so return success.
5974 : : */
5975 [ + + ]: 2233 : if (TransactionIdDidAbort(HeapTupleHeaderGetXmin(mytup.t_data)))
5976 : : {
5977 : 25 : result = TM_Ok;
5978 : 25 : goto out_locked;
5979 : : }
5980 : :
5981 : 2208 : old_infomask = mytup.t_data->t_infomask;
5982 : 2208 : old_infomask2 = mytup.t_data->t_infomask2;
5983 : 2208 : xmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5984 : :
5985 : : /*
5986 : : * If this tuple version has been updated or locked by some concurrent
5987 : : * transaction(s), what we do depends on whether our lock mode
5988 : : * conflicts with what those other transactions hold, and also on the
5989 : : * status of them.
5990 : : */
5991 [ + + ]: 2208 : if (!(old_infomask & HEAP_XMAX_INVALID))
5992 : : {
5993 : : TransactionId rawxmax;
5994 : : bool needwait;
5995 : :
5996 : 2145 : rawxmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5997 [ + + ]: 2145 : if (old_infomask & HEAP_XMAX_IS_MULTI)
5998 : : {
5999 : : int nmembers;
6000 : : int i;
6001 : : MultiXactMember *members;
6002 : :
6003 : : /*
6004 : : * We don't need a test for pg_upgrade'd tuples: this is only
6005 : : * applied to tuples after the first in an update chain. Said
6006 : : * first tuple in the chain may well be locked-in-9.2-and-
6007 : : * pg_upgraded, but that one was already locked by our caller,
6008 : : * not us; and any subsequent ones cannot be because our
6009 : : * caller must necessarily have obtained a snapshot later than
6010 : : * the pg_upgrade itself.
6011 : : */
6012 : : Assert(!HEAP_LOCKED_UPGRADED(mytup.t_data->t_infomask));
6013 : :
6014 : 2109 : nmembers = GetMultiXactIdMembers(rawxmax, &members, false,
6015 : 2109 : HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
6016 [ + + ]: 40854 : for (i = 0; i < nmembers; i++)
6017 : : {
6018 : 38745 : result = test_lockmode_for_conflict(members[i].status,
6019 : 38745 : members[i].xid,
6020 : : mode,
6021 : : &mytup,
6022 : : &needwait);
6023 : :
6024 : : /*
6025 : : * If the tuple was already locked by ourselves in a
6026 : : * previous iteration of this (say heap_lock_tuple was
6027 : : * forced to restart the locking loop because of a change
6028 : : * in xmax), then we hold the lock already on this tuple
6029 : : * version and we don't need to do anything; and this is
6030 : : * not an error condition either. We just need to skip
6031 : : * this tuple and continue locking the next version in the
6032 : : * update chain.
6033 : : */
6034 [ - + ]: 38745 : if (result == TM_SelfModified)
6035 : : {
6036 : 0 : pfree(members);
6037 : 0 : goto next;
6038 : : }
6039 : :
6040 [ - + ]: 38745 : if (needwait)
6041 : : {
6042 : 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
6043 : 0 : XactLockTableWait(members[i].xid, rel,
6044 : : &mytup.t_self,
6045 : : XLTW_LockUpdated);
6046 : 0 : pfree(members);
6047 : 0 : goto l4;
6048 : : }
6049 [ - + ]: 38745 : if (result != TM_Ok)
6050 : : {
6051 : 0 : pfree(members);
6052 : 0 : goto out_locked;
6053 : : }
6054 : : }
6055 [ + - ]: 2109 : if (members)
6056 : 2109 : pfree(members);
6057 : : }
6058 : : else
6059 : : {
6060 : : MultiXactStatus status;
6061 : :
6062 : : /*
6063 : : * For a non-multi Xmax, we first need to compute the
6064 : : * corresponding MultiXactStatus by using the infomask bits.
6065 : : */
6066 [ + + ]: 36 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
6067 : : {
6068 [ + - ]: 16 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
6069 : 16 : status = MultiXactStatusForKeyShare;
6070 [ # # ]: 0 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
6071 : 0 : status = MultiXactStatusForShare;
6072 [ # # ]: 0 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
6073 : : {
6074 [ # # ]: 0 : if (old_infomask2 & HEAP_KEYS_UPDATED)
6075 : 0 : status = MultiXactStatusForUpdate;
6076 : : else
6077 : 0 : status = MultiXactStatusForNoKeyUpdate;
6078 : : }
6079 : : else
6080 : : {
6081 : : /*
6082 : : * LOCK_ONLY present alone (a pg_upgraded tuple marked
6083 : : * as share-locked in the old cluster) shouldn't be
6084 : : * seen in the middle of an update chain.
6085 : : */
6086 [ # # ]: 0 : elog(ERROR, "invalid lock status in tuple");
6087 : : }
6088 : : }
6089 : : else
6090 : : {
6091 : : /* it's an update, but which kind? */
6092 [ + + ]: 20 : if (old_infomask2 & HEAP_KEYS_UPDATED)
6093 : 15 : status = MultiXactStatusUpdate;
6094 : : else
6095 : 5 : status = MultiXactStatusNoKeyUpdate;
6096 : : }
6097 : :
6098 : 36 : result = test_lockmode_for_conflict(status, rawxmax, mode,
6099 : : &mytup, &needwait);
6100 : :
6101 : : /*
6102 : : * If the tuple was already locked by ourselves in a previous
6103 : : * iteration of this (say heap_lock_tuple was forced to
6104 : : * restart the locking loop because of a change in xmax), then
6105 : : * we hold the lock already on this tuple version and we don't
6106 : : * need to do anything; and this is not an error condition
6107 : : * either. We just need to skip this tuple and continue
6108 : : * locking the next version in the update chain.
6109 : : */
6110 [ - + ]: 36 : if (result == TM_SelfModified)
6111 : 0 : goto next;
6112 : :
6113 [ + + ]: 36 : if (needwait)
6114 : : {
6115 : 8 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
6116 : 8 : XactLockTableWait(rawxmax, rel, &mytup.t_self,
6117 : : XLTW_LockUpdated);
6118 : 8 : goto l4;
6119 : : }
6120 [ + + ]: 28 : if (result != TM_Ok)
6121 : : {
6122 : 9 : goto out_locked;
6123 : : }
6124 : : }
6125 : : }
6126 : :
6127 : : /* compute the new Xmax and infomask values for the tuple ... */
6128 : 2191 : compute_new_xmax_infomask(xmax, old_infomask, mytup.t_data->t_infomask2,
6129 : : xid, mode, false,
6130 : : &new_xmax, &new_infomask, &new_infomask2);
6131 : :
6132 [ - + ]: 2191 : if (PageIsAllVisible(page))
6133 : : {
6134 : 0 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
6135 : 0 : unlock_vmbuffer = true;
6136 : : }
6137 : :
6138 : 2191 : START_CRIT_SECTION();
6139 : :
6140 : : /* ... and set them */
6141 : 2191 : HeapTupleHeaderSetXmax(mytup.t_data, new_xmax);
6142 : 2191 : mytup.t_data->t_infomask &= ~HEAP_XMAX_BITS;
6143 : 2191 : mytup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6144 : 2191 : mytup.t_data->t_infomask |= new_infomask;
6145 : 2191 : mytup.t_data->t_infomask2 |= new_infomask2;
6146 : :
6147 : 2191 : MarkBufferDirty(buf);
6148 : :
6149 [ - + ]: 2191 : if (PageIsAllVisible(page))
6150 : : {
6151 : : /* It's possible all-frozen was already clear */
6152 [ # # ]: 0 : if (visibilitymap_clear(rel->rd_locator, block, vmbuffer,
6153 : : VISIBILITYMAP_ALL_FROZEN))
6154 : 0 : cleared_all_frozen = true;
6155 : : }
6156 : :
6157 : : /* XLOG stuff */
6158 [ + - + + : 2191 : if (RelationNeedsWAL(rel))
+ - + - ]
6159 : : {
6160 : : xl_heap_lock_updated xlrec;
6161 : : XLogRecPtr recptr;
6162 : :
6163 : 2191 : XLogBeginInsert();
6164 : 2191 : XLogRegisterBuffer(HEAP_LOCK_BLKREF_HEAP, buf, REGBUF_STANDARD);
6165 : :
6166 : 2191 : xlrec.offnum = ItemPointerGetOffsetNumber(&mytup.t_self);
6167 : 2191 : xlrec.xmax = new_xmax;
6168 : 2191 : xlrec.infobits_set = compute_infobits(new_infomask, new_infomask2);
6169 : 2191 : xlrec.flags =
6170 : 2191 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
6171 : :
6172 : 2191 : XLogRegisterData(&xlrec, SizeOfHeapLockUpdated);
6173 : :
6174 [ - + ]: 2191 : if (cleared_all_frozen)
6175 : 0 : XLogRegisterBuffer(HEAP_LOCK_BLKREF_VM, vmbuffer, 0);
6176 : :
6177 : 2191 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_LOCK_UPDATED);
6178 : :
6179 : 2191 : PageSetLSN(page, recptr);
6180 : :
6181 [ - + ]: 2191 : if (cleared_all_frozen)
6182 : 0 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
6183 : : }
6184 : :
6185 : 2191 : END_CRIT_SECTION();
6186 : :
6187 : : /* release VM lock first, since it covers many heap blocks */
6188 [ + - ]: 2191 : if (unlock_vmbuffer)
6189 : : {
6190 : 0 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
6191 : 0 : unlock_vmbuffer = false;
6192 : : }
6193 : :
6194 : 2191 : next:
6195 : : /* if we find the end of update chain, we're done. */
6196 [ + - + - ]: 4382 : if (mytup.t_data->t_infomask & HEAP_XMAX_INVALID ||
6197 [ + + ]: 4382 : HeapTupleHeaderIndicatesMovedPartitions(mytup.t_data) ||
6198 [ + + ]: 2195 : ItemPointerEquals(&mytup.t_self, &mytup.t_data->t_ctid) ||
6199 : 4 : HeapTupleHeaderIsOnlyLocked(mytup.t_data))
6200 : : {
6201 : 2188 : result = TM_Ok;
6202 : 2188 : goto out_locked;
6203 : : }
6204 : :
6205 : : /* tail recursion */
6206 : 3 : priorXmax = HeapTupleHeaderGetUpdateXid(mytup.t_data);
6207 : 3 : ItemPointerCopy(&(mytup.t_data->t_ctid), &tupid);
6208 : 3 : UnlockReleaseBuffer(buf);
6209 : : }
6210 : :
6211 : : result = TM_Ok;
6212 : :
6213 : 2224 : out_locked:
6214 : 2224 : UnlockReleaseBuffer(buf);
6215 : :
6216 : 2224 : out_unlocked:
6217 [ - + ]: 2224 : if (vmbuffer != InvalidBuffer)
6218 : 0 : ReleaseBuffer(vmbuffer);
6219 : : Assert(!unlock_vmbuffer);
6220 : :
6221 : 2224 : return result;
6222 : : }
6223 : :
6224 : : /*
6225 : : * heap_lock_updated_tuple
6226 : : * Follow update chain when locking an updated tuple, acquiring locks (row
6227 : : * marks) on the updated versions.
6228 : : *
6229 : : * 'prior_infomask', 'prior_raw_xmax' and 'prior_ctid' are the corresponding
6230 : : * fields from the initial tuple. We will lock the tuples starting from the
6231 : : * one that 'prior_ctid' points to. Note: This function does not lock the
6232 : : * initial tuple itself.
6233 : : *
6234 : : * This function doesn't check visibility, it just unconditionally marks the
6235 : : * tuple(s) as locked. If any tuple in the updated chain is being deleted
6236 : : * concurrently (or updated with the key being modified), sleep until the
6237 : : * transaction doing it is finished.
6238 : : *
6239 : : * Note that we don't acquire heavyweight tuple locks on the tuples we walk
6240 : : * when we have to wait for other transactions to release them, as opposed to
6241 : : * what heap_lock_tuple does. The reason is that having more than one
6242 : : * transaction walking the chain is probably uncommon enough that risk of
6243 : : * starvation is not likely: one of the preconditions for being here is that
6244 : : * the snapshot in use predates the update that created this tuple (because we
6245 : : * started at an earlier version of the tuple), but at the same time such a
6246 : : * transaction cannot be using repeatable read or serializable isolation
6247 : : * levels, because that would lead to a serializability failure.
6248 : : */
6249 : : static TM_Result
6250 : 2226 : heap_lock_updated_tuple(Relation rel,
6251 : : uint16 prior_infomask,
6252 : : TransactionId prior_raw_xmax,
6253 : : const ItemPointerData *prior_ctid,
6254 : : TransactionId xid, LockTupleMode mode)
6255 : : {
6256 : 2226 : INJECTION_POINT("heap_lock_updated_tuple", NULL);
6257 : :
6258 : : /*
6259 : : * If the tuple has moved into another partition (effectively a delete)
6260 : : * stop here.
6261 : : */
6262 [ + + ]: 2226 : if (!ItemPointerIndicatesMovedPartitions(prior_ctid))
6263 : : {
6264 : : TransactionId prior_xmax;
6265 : :
6266 : : /*
6267 : : * If this is the first possibly-multixact-able operation in the
6268 : : * current transaction, set my per-backend OldestMemberMXactId
6269 : : * setting. We can be certain that the transaction will never become a
6270 : : * member of any older MultiXactIds than that. (We have to do this
6271 : : * even if we end up just using our own TransactionId below, since
6272 : : * some other backend could incorporate our XID into a MultiXact
6273 : : * immediately afterwards.)
6274 : : */
6275 : 2224 : MultiXactIdSetOldestMember();
6276 : :
6277 : 4448 : prior_xmax = (prior_infomask & HEAP_XMAX_IS_MULTI) ?
6278 [ + + ]: 2224 : MultiXactIdGetUpdateXid(prior_raw_xmax, prior_infomask) : prior_raw_xmax;
6279 : 2224 : return heap_lock_updated_tuple_rec(rel, prior_xmax, prior_ctid, xid, mode);
6280 : : }
6281 : :
6282 : : /* nothing to lock */
6283 : 2 : return TM_Ok;
6284 : : }
6285 : :
6286 : : /*
6287 : : * heap_finish_speculative - mark speculative insertion as successful
6288 : : *
6289 : : * To successfully finish a speculative insertion we have to clear speculative
6290 : : * token from tuple. To do so the t_ctid field, which will contain a
6291 : : * speculative token value, is modified in place to point to the tuple itself,
6292 : : * which is characteristic of a newly inserted ordinary tuple.
6293 : : *
6294 : : * NB: It is not ok to commit without either finishing or aborting a
6295 : : * speculative insertion. We could treat speculative tuples of committed
6296 : : * transactions implicitly as completed, but then we would have to be prepared
6297 : : * to deal with speculative tokens on committed tuples. That wouldn't be
6298 : : * difficult - no-one looks at the ctid field of a tuple with invalid xmax -
6299 : : * but clearing the token at completion isn't very expensive either.
6300 : : * An explicit confirmation WAL record also makes logical decoding simpler.
6301 : : */
6302 : : void
6303 : 2216 : heap_finish_speculative(Relation relation, const ItemPointerData *tid)
6304 : : {
6305 : : Buffer buffer;
6306 : : Page page;
6307 : : OffsetNumber offnum;
6308 : : ItemId lp;
6309 : : HeapTupleHeader htup;
6310 : :
6311 : 2216 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
6312 : 2216 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6313 : 2216 : page = BufferGetPage(buffer);
6314 : :
6315 : 2216 : offnum = ItemPointerGetOffsetNumber(tid);
6316 [ + - - + ]: 2216 : if (offnum < 1 || offnum > PageGetMaxOffsetNumber(page))
6317 [ # # ]: 0 : elog(ERROR, "offnum out of range");
6318 : 2216 : lp = PageGetItemId(page, offnum);
6319 [ - + ]: 2216 : if (!ItemIdIsNormal(lp))
6320 [ # # ]: 0 : elog(ERROR, "invalid lp");
6321 : :
6322 : 2216 : htup = (HeapTupleHeader) PageGetItem(page, lp);
6323 : :
6324 : : /* NO EREPORT(ERROR) from here till changes are logged */
6325 : 2216 : START_CRIT_SECTION();
6326 : :
6327 : : Assert(HeapTupleHeaderIsSpeculative(htup));
6328 : :
6329 : 2216 : MarkBufferDirty(buffer);
6330 : :
6331 : : /*
6332 : : * Replace the speculative insertion token with a real t_ctid, pointing to
6333 : : * itself like it does on regular tuples.
6334 : : */
6335 : 2216 : htup->t_ctid = *tid;
6336 : :
6337 : : /* XLOG stuff */
6338 [ + + + + : 2216 : if (RelationNeedsWAL(relation))
+ - + - ]
6339 : : {
6340 : : xl_heap_confirm xlrec;
6341 : : XLogRecPtr recptr;
6342 : :
6343 : 2196 : xlrec.offnum = ItemPointerGetOffsetNumber(tid);
6344 : :
6345 : 2196 : XLogBeginInsert();
6346 : :
6347 : : /* We want the same filtering on this as on a plain insert */
6348 : 2196 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
6349 : :
6350 : 2196 : XLogRegisterData(&xlrec, SizeOfHeapConfirm);
6351 : 2196 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6352 : :
6353 : 2196 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_CONFIRM);
6354 : :
6355 : 2196 : PageSetLSN(page, recptr);
6356 : : }
6357 : :
6358 : 2216 : END_CRIT_SECTION();
6359 : :
6360 : 2216 : UnlockReleaseBuffer(buffer);
6361 : 2216 : }
6362 : :
6363 : : /*
6364 : : * heap_abort_speculative - kill a speculatively inserted tuple
6365 : : *
6366 : : * Marks a tuple that was speculatively inserted in the same command as dead,
6367 : : * by setting its xmin as invalid. That makes it immediately appear as dead
6368 : : * to all transactions, including our own. In particular, it makes
6369 : : * HeapTupleSatisfiesDirty() regard the tuple as dead, so that another backend
6370 : : * inserting a duplicate key value won't unnecessarily wait for our whole
6371 : : * transaction to finish (it'll just wait for our speculative insertion to
6372 : : * finish).
6373 : : *
6374 : : * Killing the tuple prevents "unprincipled deadlocks", which are deadlocks
6375 : : * that arise due to a mutual dependency that is not user visible. By
6376 : : * definition, unprincipled deadlocks cannot be prevented by the user
6377 : : * reordering lock acquisition in client code, because the implementation level
6378 : : * lock acquisitions are not under the user's direct control. If speculative
6379 : : * inserters did not take this precaution, then under high concurrency they
6380 : : * could deadlock with each other, which would not be acceptable.
6381 : : *
6382 : : * This is somewhat redundant with heap_delete, but we prefer to have a
6383 : : * dedicated routine with stripped down requirements. Note that this is also
6384 : : * used to delete the TOAST tuples created during speculative insertion.
6385 : : *
6386 : : * This routine does not affect logical decoding as it only looks at
6387 : : * confirmation records.
6388 : : */
6389 : : void
6390 : 16 : heap_abort_speculative(Relation relation, const ItemPointerData *tid)
6391 : : {
6392 : 16 : TransactionId xid = GetCurrentTransactionId();
6393 : : ItemId lp;
6394 : : HeapTupleData tp;
6395 : : Page page;
6396 : : BlockNumber block;
6397 : : Buffer buffer;
6398 : :
6399 : : Assert(ItemPointerIsValid(tid));
6400 : :
6401 : 16 : block = ItemPointerGetBlockNumber(tid);
6402 : 16 : buffer = ReadBuffer(relation, block);
6403 : 16 : page = BufferGetPage(buffer);
6404 : :
6405 : 16 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6406 : :
6407 : : /*
6408 : : * Page can't be all visible, we just inserted into it, and are still
6409 : : * running.
6410 : : */
6411 : : Assert(!PageIsAllVisible(page));
6412 : :
6413 : 16 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
6414 : : Assert(ItemIdIsNormal(lp));
6415 : :
6416 : 16 : tp.t_tableOid = RelationGetRelid(relation);
6417 : 16 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
6418 : 16 : tp.t_len = ItemIdGetLength(lp);
6419 : 16 : tp.t_self = *tid;
6420 : :
6421 : : /*
6422 : : * Sanity check that the tuple really is a speculatively inserted tuple,
6423 : : * inserted by us.
6424 : : */
6425 [ - + ]: 16 : if (tp.t_data->t_choice.t_heap.t_xmin != xid)
6426 [ # # ]: 0 : elog(ERROR, "attempted to kill a tuple inserted by another transaction");
6427 [ + + - + ]: 16 : if (!(IsToastRelation(relation) || HeapTupleHeaderIsSpeculative(tp.t_data)))
6428 [ # # ]: 0 : elog(ERROR, "attempted to kill a non-speculative tuple");
6429 : : Assert(!HeapTupleHeaderIsHeapOnly(tp.t_data));
6430 : :
6431 : : /*
6432 : : * No need to check for serializable conflicts here. There is never a
6433 : : * need for a combo CID, either. No need to extract replica identity, or
6434 : : * do anything special with infomask bits.
6435 : : */
6436 : :
6437 : 16 : START_CRIT_SECTION();
6438 : :
6439 : : /*
6440 : : * The tuple will become DEAD immediately. Flag that this page is a
6441 : : * candidate for pruning by setting xmin to TransactionXmin. While not
6442 : : * immediately prunable, it is the oldest xid we can cheaply determine
6443 : : * that's safe against wraparound / being older than the table's
6444 : : * relfrozenxid. To defend against the unlikely case of a new relation
6445 : : * having a newer relfrozenxid than our TransactionXmin, use relfrozenxid
6446 : : * if so (vacuum can't subsequently move relfrozenxid to beyond
6447 : : * TransactionXmin, so there's no race here).
6448 : : */
6449 : : Assert(TransactionIdIsValid(TransactionXmin));
6450 : : {
6451 : 16 : TransactionId relfrozenxid = relation->rd_rel->relfrozenxid;
6452 : : TransactionId prune_xid;
6453 : :
6454 [ - + ]: 16 : if (TransactionIdPrecedes(TransactionXmin, relfrozenxid))
6455 : 0 : prune_xid = relfrozenxid;
6456 : : else
6457 : 16 : prune_xid = TransactionXmin;
6458 [ + - + + ]: 16 : PageSetPrunable(page, prune_xid);
6459 : : }
6460 : :
6461 : : /* store transaction information of xact deleting the tuple */
6462 : 16 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
6463 : 16 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6464 : :
6465 : : /*
6466 : : * Set the tuple header xmin to InvalidTransactionId. This makes the
6467 : : * tuple immediately invisible everyone. (In particular, to any
6468 : : * transactions waiting on the speculative token, woken up later.)
6469 : : */
6470 : 16 : HeapTupleHeaderSetXmin(tp.t_data, InvalidTransactionId);
6471 : :
6472 : : /* Clear the speculative insertion token too */
6473 : 16 : tp.t_data->t_ctid = tp.t_self;
6474 : :
6475 : 16 : MarkBufferDirty(buffer);
6476 : :
6477 : : /*
6478 : : * XLOG stuff
6479 : : *
6480 : : * The WAL records generated here match heap_delete(). The same recovery
6481 : : * routines are used.
6482 : : */
6483 [ + + + + : 16 : if (RelationNeedsWAL(relation))
+ - + - ]
6484 : : {
6485 : : xl_heap_delete xlrec;
6486 : : XLogRecPtr recptr;
6487 : :
6488 : 12 : xlrec.flags = XLH_DELETE_IS_SUPER;
6489 : 24 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
6490 : 12 : tp.t_data->t_infomask2);
6491 : 12 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
6492 : 12 : xlrec.xmax = xid;
6493 : :
6494 : 12 : XLogBeginInsert();
6495 : 12 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
6496 : 12 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6497 : :
6498 : : /* No replica identity & replication origin logged */
6499 : :
6500 : 12 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
6501 : :
6502 : 12 : PageSetLSN(page, recptr);
6503 : : }
6504 : :
6505 : 16 : END_CRIT_SECTION();
6506 : :
6507 : 16 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6508 : :
6509 [ + + ]: 16 : if (HeapTupleHasExternal(&tp))
6510 : : {
6511 : : Assert(!IsToastRelation(relation));
6512 : 1 : heap_toast_delete(relation, &tp, true);
6513 : : }
6514 : :
6515 : : /*
6516 : : * Never need to mark tuple for invalidation, since catalogs don't support
6517 : : * speculative insertion
6518 : : */
6519 : :
6520 : : /* Now we can release the buffer */
6521 : 16 : ReleaseBuffer(buffer);
6522 : :
6523 : : /* count deletion, as we counted the insertion too */
6524 : 16 : pgstat_count_heap_delete(relation);
6525 : 16 : }
6526 : :
6527 : : /*
6528 : : * heap_inplace_lock - protect inplace update from concurrent heap_update()
6529 : : *
6530 : : * Evaluate whether the tuple's state is compatible with a no-key update.
6531 : : * Current transaction rowmarks are fine, as is KEY SHARE from any
6532 : : * transaction. If compatible, return true with the buffer exclusive-locked,
6533 : : * and the caller must release that by calling
6534 : : * heap_inplace_update_and_unlock(), calling heap_inplace_unlock(), or raising
6535 : : * an error. Otherwise, call release_callback(arg), wait for blocking
6536 : : * transactions to end, and return false.
6537 : : *
6538 : : * Since this is intended for system catalogs and SERIALIZABLE doesn't cover
6539 : : * DDL, this doesn't guarantee any particular predicate locking.
6540 : : *
6541 : : * heap_delete() is a rarer source of blocking transactions (xwait). We'll
6542 : : * wait for such a transaction just like for the normal heap_update() case.
6543 : : * Normal concurrent DROP commands won't cause that, because all inplace
6544 : : * updaters take some lock that conflicts with DROP. An explicit SQL "DELETE
6545 : : * FROM pg_class" can cause it. By waiting, if the concurrent transaction
6546 : : * executed both "DELETE FROM pg_class" and "INSERT INTO pg_class", our caller
6547 : : * can find the successor tuple.
6548 : : *
6549 : : * Readers of inplace-updated fields expect changes to those fields are
6550 : : * durable. For example, vac_truncate_clog() reads datfrozenxid from
6551 : : * pg_database tuples via catalog snapshots. A future snapshot must not
6552 : : * return a lower datfrozenxid for the same database OID (lower in the
6553 : : * FullTransactionIdPrecedes() sense). We achieve that since no update of a
6554 : : * tuple can start while we hold a lock on its buffer. In cases like
6555 : : * BEGIN;GRANT;CREATE INDEX;COMMIT we're inplace-updating a tuple visible only
6556 : : * to this transaction. ROLLBACK then is one case where it's okay to lose
6557 : : * inplace updates. (Restoring relhasindex=false on ROLLBACK is fine, since
6558 : : * any concurrent CREATE INDEX would have blocked, then inplace-updated the
6559 : : * committed tuple.)
6560 : : *
6561 : : * In principle, we could avoid waiting by overwriting every tuple in the
6562 : : * updated tuple chain. Reader expectations permit updating a tuple only if
6563 : : * it's aborted, is the tail of the chain, or we already updated the tuple
6564 : : * referenced in its t_ctid. Hence, we would need to overwrite the tuples in
6565 : : * order from tail to head. That would imply either (a) mutating all tuples
6566 : : * in one critical section or (b) accepting a chance of partial completion.
6567 : : * Partial completion of a relfrozenxid update would have the weird
6568 : : * consequence that the table's next VACUUM could see the table's relfrozenxid
6569 : : * move forward between vacuum_get_cutoffs() and finishing.
6570 : : */
6571 : : bool
6572 : 227460 : heap_inplace_lock(Relation relation,
6573 : : HeapTuple oldtup_ptr, Buffer buffer,
6574 : : void (*release_callback) (void *), void *arg)
6575 : : {
6576 : 227460 : HeapTupleData oldtup = *oldtup_ptr; /* minimize diff vs. heap_update() */
6577 : : TM_Result result;
6578 : : bool ret;
6579 : :
6580 : : #ifdef USE_ASSERT_CHECKING
6581 : : if (RelationGetRelid(relation) == RelationRelationId)
6582 : : check_inplace_rel_lock(oldtup_ptr);
6583 : : #endif
6584 : :
6585 : : Assert(BufferIsValid(buffer));
6586 : :
6587 : : /*
6588 : : * Register shared cache invals if necessary. Other sessions may finish
6589 : : * inplace updates of this tuple between this step and LockTuple(). Since
6590 : : * inplace updates don't change cache keys, that's harmless.
6591 : : *
6592 : : * While it's tempting to register invals only after confirming we can
6593 : : * return true, the following obstacle precludes reordering steps that
6594 : : * way. Registering invals might reach a CatalogCacheInitializeCache()
6595 : : * that locks "buffer". That would hang indefinitely if running after our
6596 : : * own LockBuffer(). Hence, we must register invals before LockBuffer().
6597 : : */
6598 : 227460 : CacheInvalidateHeapTupleInplace(relation, oldtup_ptr);
6599 : :
6600 : 227460 : LockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6601 : 227460 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6602 : :
6603 : : /*----------
6604 : : * Interpret HeapTupleSatisfiesUpdate() like heap_update() does, except:
6605 : : *
6606 : : * - wait unconditionally
6607 : : * - already locked tuple above, since inplace needs that unconditionally
6608 : : * - don't recheck header after wait: simpler to defer to next iteration
6609 : : * - don't try to continue even if the updater aborts: likewise
6610 : : * - no crosscheck
6611 : : */
6612 : 227460 : result = HeapTupleSatisfiesUpdate(&oldtup, GetCurrentCommandId(false),
6613 : : buffer);
6614 : :
6615 [ - + ]: 227460 : if (result == TM_Invisible)
6616 : : {
6617 : : /* no known way this can happen */
6618 [ # # ]: 0 : ereport(ERROR,
6619 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6620 : : errmsg_internal("attempted to overwrite invisible tuple")));
6621 : : }
6622 [ - + ]: 227460 : else if (result == TM_SelfModified)
6623 : : {
6624 : : /*
6625 : : * CREATE INDEX might reach this if an expression is silly enough to
6626 : : * call e.g. SELECT ... FROM pg_class FOR SHARE. C code of other SQL
6627 : : * statements might get here after a heap_update() of the same row, in
6628 : : * the absence of an intervening CommandCounterIncrement().
6629 : : */
6630 [ # # ]: 0 : ereport(ERROR,
6631 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6632 : : errmsg("tuple to be updated was already modified by an operation triggered by the current command")));
6633 : : }
6634 [ + + ]: 227460 : else if (result == TM_BeingModified)
6635 : : {
6636 : : TransactionId xwait;
6637 : : uint16 infomask;
6638 : :
6639 : 38 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
6640 : 38 : infomask = oldtup.t_data->t_infomask;
6641 : :
6642 [ + + ]: 38 : if (infomask & HEAP_XMAX_IS_MULTI)
6643 : : {
6644 : 5 : LockTupleMode lockmode = LockTupleNoKeyExclusive;
6645 : 5 : MultiXactStatus mxact_status = MultiXactStatusNoKeyUpdate;
6646 : : int remain;
6647 : :
6648 [ + + ]: 5 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
6649 : : lockmode, NULL))
6650 : : {
6651 : 2 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6652 : 2 : release_callback(arg);
6653 : 2 : ret = false;
6654 : 2 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
6655 : : relation, &oldtup.t_self, XLTW_Update,
6656 : : &remain);
6657 : : }
6658 : : else
6659 : 3 : ret = true;
6660 : : }
6661 [ + + ]: 33 : else if (TransactionIdIsCurrentTransactionId(xwait))
6662 : 1 : ret = true;
6663 [ + + ]: 32 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
6664 : 1 : ret = true;
6665 : : else
6666 : : {
6667 : 31 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6668 : 31 : release_callback(arg);
6669 : 31 : ret = false;
6670 : 31 : XactLockTableWait(xwait, relation, &oldtup.t_self,
6671 : : XLTW_Update);
6672 : : }
6673 : : }
6674 : : else
6675 : : {
6676 : 227422 : ret = (result == TM_Ok);
6677 [ + + ]: 227422 : if (!ret)
6678 : : {
6679 : 1 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6680 : 1 : release_callback(arg);
6681 : : }
6682 : : }
6683 : :
6684 : : /*
6685 : : * GetCatalogSnapshot() relies on invalidation messages to know when to
6686 : : * take a new snapshot. COMMIT of xwait is responsible for sending the
6687 : : * invalidation. We're not acquiring heavyweight locks sufficient to
6688 : : * block if not yet sent, so we must take a new snapshot to ensure a later
6689 : : * attempt has a fair chance. While we don't need this if xwait aborted,
6690 : : * don't bother optimizing that.
6691 : : */
6692 [ + + ]: 227460 : if (!ret)
6693 : : {
6694 : 34 : UnlockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6695 : 34 : ForgetInplace_Inval();
6696 : 34 : InvalidateCatalogSnapshot();
6697 : : }
6698 : 227460 : return ret;
6699 : : }
6700 : :
6701 : : /*
6702 : : * heap_inplace_update_and_unlock - core of systable_inplace_update_finish
6703 : : *
6704 : : * The tuple cannot change size, and therefore its header fields and null
6705 : : * bitmap (if any) don't change either.
6706 : : *
6707 : : * Since we hold LOCKTAG_TUPLE, no updater has a local copy of this tuple.
6708 : : */
6709 : : void
6710 : 100998 : heap_inplace_update_and_unlock(Relation relation,
6711 : : HeapTuple oldtup, HeapTuple tuple,
6712 : : Buffer buffer)
6713 : : {
6714 : 100998 : HeapTupleHeader htup = oldtup->t_data;
6715 : : uint32 oldlen;
6716 : : uint32 newlen;
6717 : : char *dst;
6718 : : char *src;
6719 : 100998 : int nmsgs = 0;
6720 : 100998 : SharedInvalidationMessage *invalMessages = NULL;
6721 : 100998 : bool RelcacheInitFileInval = false;
6722 : :
6723 : : Assert(ItemPointerEquals(&oldtup->t_self, &tuple->t_self));
6724 : 100998 : oldlen = oldtup->t_len - htup->t_hoff;
6725 : 100998 : newlen = tuple->t_len - tuple->t_data->t_hoff;
6726 [ + - - + ]: 100998 : if (oldlen != newlen || htup->t_hoff != tuple->t_data->t_hoff)
6727 [ # # ]: 0 : elog(ERROR, "wrong tuple length");
6728 : :
6729 : 100998 : dst = (char *) htup + htup->t_hoff;
6730 : 100998 : src = (char *) tuple->t_data + tuple->t_data->t_hoff;
6731 : :
6732 : : /* Like RecordTransactionCommit(), log only if needed */
6733 [ + + ]: 100998 : if (XLogStandbyInfoActive())
6734 : 73789 : nmsgs = inplaceGetInvalidationMessages(&invalMessages,
6735 : : &RelcacheInitFileInval);
6736 : :
6737 : : /*
6738 : : * Unlink relcache init files as needed. If unlinking, acquire
6739 : : * RelCacheInitLock until after associated invalidations. By doing this
6740 : : * in advance, if we checkpoint and then crash between inplace
6741 : : * XLogInsert() and inval, we don't rely on StartupXLOG() ->
6742 : : * RelationCacheInitFileRemove(). That uses elevel==LOG, so replay would
6743 : : * neglect to PANIC on EIO.
6744 : : */
6745 : 100998 : PreInplace_Inval();
6746 : :
6747 : : /*----------
6748 : : * NO EREPORT(ERROR) from here till changes are complete
6749 : : *
6750 : : * Our exclusive buffer lock won't stop a reader having already pinned and
6751 : : * checked visibility for this tuple. With the usual order of changes
6752 : : * (i.e. updating the buffer contents before WAL logging), a reader could
6753 : : * observe our not-yet-persistent update to relfrozenxid and update
6754 : : * datfrozenxid based on that. A crash in that moment could allow
6755 : : * datfrozenxid to overtake relfrozenxid:
6756 : : *
6757 : : * ["D" is a VACUUM (ONLY_DATABASE_STATS)]
6758 : : * ["R" is a VACUUM tbl]
6759 : : * D: vac_update_datfrozenxid() -> systable_beginscan(pg_class)
6760 : : * D: systable_getnext() returns pg_class tuple of tbl
6761 : : * R: memcpy() into pg_class tuple of tbl
6762 : : * D: raise pg_database.datfrozenxid, XLogInsert(), finish
6763 : : * [crash]
6764 : : * [recovery restores datfrozenxid w/o relfrozenxid]
6765 : : *
6766 : : * We avoid that by using a temporary copy of the buffer to hide our
6767 : : * change from other backends until the change has been WAL-logged. We
6768 : : * apply our change to the temporary copy and WAL-log it, before modifying
6769 : : * the real page. That way any action a reader of the in-place-updated
6770 : : * value takes will be WAL logged after this change.
6771 : : */
6772 : 100998 : START_CRIT_SECTION();
6773 : :
6774 : 100998 : MarkBufferDirty(buffer);
6775 : :
6776 : : /* XLOG stuff */
6777 [ + - + + : 100998 : if (RelationNeedsWAL(relation))
+ - + + ]
6778 : : {
6779 : : xl_heap_inplace xlrec;
6780 : : PGAlignedBlock copied_buffer;
6781 : 100994 : char *origdata = (char *) BufferGetBlock(buffer);
6782 : 100994 : Page page = BufferGetPage(buffer);
6783 : 100994 : uint16 lower = ((PageHeader) page)->pd_lower;
6784 : 100994 : uint16 upper = ((PageHeader) page)->pd_upper;
6785 : : uintptr_t dst_offset_in_block;
6786 : : RelFileLocator rlocator;
6787 : : ForkNumber forkno;
6788 : : BlockNumber blkno;
6789 : : XLogRecPtr recptr;
6790 : :
6791 : 100994 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
6792 : 100994 : xlrec.dbId = MyDatabaseId;
6793 : 100994 : xlrec.tsId = MyDatabaseTableSpace;
6794 : 100994 : xlrec.relcacheInitFileInval = RelcacheInitFileInval;
6795 : 100994 : xlrec.nmsgs = nmsgs;
6796 : :
6797 : 100994 : XLogBeginInsert();
6798 : 100994 : XLogRegisterData(&xlrec, MinSizeOfHeapInplace);
6799 [ + + ]: 100994 : if (nmsgs != 0)
6800 : 55834 : XLogRegisterData(invalMessages,
6801 : : nmsgs * sizeof(SharedInvalidationMessage));
6802 : :
6803 : : /* register block matching what buffer will look like after changes */
6804 : 100994 : memcpy(copied_buffer.data, origdata, lower);
6805 : 100994 : memcpy(copied_buffer.data + upper, origdata + upper, BLCKSZ - upper);
6806 : 100994 : dst_offset_in_block = dst - origdata;
6807 : 100994 : memcpy(copied_buffer.data + dst_offset_in_block, src, newlen);
6808 : 100994 : BufferGetTag(buffer, &rlocator, &forkno, &blkno);
6809 : : Assert(forkno == MAIN_FORKNUM);
6810 : 100994 : XLogRegisterBlock(0, &rlocator, forkno, blkno, copied_buffer.data,
6811 : : REGBUF_STANDARD);
6812 : 100994 : XLogRegisterBufData(0, src, newlen);
6813 : :
6814 : : /* inplace updates aren't decoded atm, don't log the origin */
6815 : :
6816 : 100994 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_INPLACE);
6817 : :
6818 : 100994 : PageSetLSN(page, recptr);
6819 : : }
6820 : :
6821 : 100998 : memcpy(dst, src, newlen);
6822 : :
6823 : 100998 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6824 : :
6825 : : /*
6826 : : * Send invalidations to shared queue. SearchSysCacheLocked1() assumes we
6827 : : * do this before UnlockTuple().
6828 : : */
6829 : 100998 : AtInplace_Inval();
6830 : :
6831 : 100998 : END_CRIT_SECTION();
6832 : 100998 : UnlockTuple(relation, &tuple->t_self, InplaceUpdateTupleLock);
6833 : :
6834 : 100998 : AcceptInvalidationMessages(); /* local processing of just-sent inval */
6835 : :
6836 : : /*
6837 : : * Queue a transactional inval, for logical decoding and for third-party
6838 : : * code that might have been relying on it since long before inplace
6839 : : * update adopted immediate invalidation. See README.tuplock section
6840 : : * "Reading inplace-updated columns" for logical decoding details.
6841 : : */
6842 [ + + ]: 100998 : if (!IsBootstrapProcessingMode())
6843 : 83043 : CacheInvalidateHeapTuple(relation, tuple, NULL);
6844 : 100998 : }
6845 : :
6846 : : /*
6847 : : * heap_inplace_unlock - reverse of heap_inplace_lock
6848 : : */
6849 : : void
6850 : 126428 : heap_inplace_unlock(Relation relation,
6851 : : HeapTuple oldtup, Buffer buffer)
6852 : : {
6853 : 126428 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6854 : 126428 : UnlockTuple(relation, &oldtup->t_self, InplaceUpdateTupleLock);
6855 : 126428 : ForgetInplace_Inval();
6856 : 126428 : }
6857 : :
6858 : : #define FRM_NOOP 0x0001
6859 : : #define FRM_INVALIDATE_XMAX 0x0002
6860 : : #define FRM_RETURN_IS_XID 0x0004
6861 : : #define FRM_RETURN_IS_MULTI 0x0008
6862 : : #define FRM_MARK_COMMITTED 0x0010
6863 : :
6864 : : /*
6865 : : * FreezeMultiXactId
6866 : : * Determine what to do during freezing when a tuple is marked by a
6867 : : * MultiXactId.
6868 : : *
6869 : : * "flags" is an output value; it's used to tell caller what to do on return.
6870 : : * "pagefrz" is an input/output value, used to manage page level freezing.
6871 : : *
6872 : : * Possible values that we can set in "flags":
6873 : : * FRM_NOOP
6874 : : * don't do anything -- keep existing Xmax
6875 : : * FRM_INVALIDATE_XMAX
6876 : : * mark Xmax as InvalidTransactionId and set XMAX_INVALID flag.
6877 : : * FRM_RETURN_IS_XID
6878 : : * The Xid return value is a single update Xid to set as xmax.
6879 : : * FRM_MARK_COMMITTED
6880 : : * Xmax can be marked as HEAP_XMAX_COMMITTED
6881 : : * FRM_RETURN_IS_MULTI
6882 : : * The return value is a new MultiXactId to set as new Xmax.
6883 : : * (caller must obtain proper infomask bits using GetMultiXactIdHintBits)
6884 : : *
6885 : : * Caller delegates control of page freezing to us. In practice we always
6886 : : * force freezing of caller's page unless FRM_NOOP processing is indicated.
6887 : : * We help caller ensure that XIDs < FreezeLimit and MXIDs < MultiXactCutoff
6888 : : * can never be left behind. We freely choose when and how to process each
6889 : : * Multi, without ever violating the cutoff postconditions for freezing.
6890 : : *
6891 : : * It's useful to remove Multis on a proactive timeline (relative to freezing
6892 : : * XIDs) to keep MultiXact member SLRU buffer misses to a minimum. It can also
6893 : : * be cheaper in the short run, for us, since we too can avoid SLRU buffer
6894 : : * misses through eager processing.
6895 : : *
6896 : : * NB: Creates a _new_ MultiXactId when FRM_RETURN_IS_MULTI is set, though only
6897 : : * when FreezeLimit and/or MultiXactCutoff cutoffs leave us with no choice.
6898 : : * This can usually be put off, which is usually enough to avoid it altogether.
6899 : : * Allocating new multis during VACUUM should be avoided on general principle;
6900 : : * only VACUUM can advance relminmxid, so allocating new Multis here comes with
6901 : : * its own special risks.
6902 : : *
6903 : : * NB: Caller must maintain "no freeze" NewRelfrozenXid/NewRelminMxid trackers
6904 : : * using heap_tuple_should_freeze when we haven't forced page-level freezing.
6905 : : *
6906 : : * NB: Caller should avoid needlessly calling heap_tuple_should_freeze when we
6907 : : * have already forced page-level freezing, since that might incur the same
6908 : : * SLRU buffer misses that we specifically intended to avoid by freezing.
6909 : : */
6910 : : static TransactionId
6911 : 7 : FreezeMultiXactId(MultiXactId multi, uint16 t_infomask,
6912 : : const struct VacuumCutoffs *cutoffs, uint16 *flags,
6913 : : HeapPageFreeze *pagefrz)
6914 : : {
6915 : : TransactionId newxmax;
6916 : : MultiXactMember *members;
6917 : : int nmembers;
6918 : : bool need_replace;
6919 : : int nnewmembers;
6920 : : MultiXactMember *newmembers;
6921 : : bool has_lockers;
6922 : : TransactionId update_xid;
6923 : : bool update_committed;
6924 : : TransactionId FreezePageRelfrozenXid;
6925 : :
6926 : 7 : *flags = 0;
6927 : :
6928 : : /* We should only be called in Multis */
6929 : : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
6930 : :
6931 [ + - - + ]: 14 : if (!MultiXactIdIsValid(multi) ||
6932 : 7 : HEAP_LOCKED_UPGRADED(t_infomask))
6933 : : {
6934 : 0 : *flags |= FRM_INVALIDATE_XMAX;
6935 : 0 : pagefrz->freeze_required = true;
6936 : 0 : return InvalidTransactionId;
6937 : : }
6938 [ - + ]: 7 : else if (MultiXactIdPrecedes(multi, cutoffs->relminmxid))
6939 [ # # ]: 0 : ereport(ERROR,
6940 : : (errcode(ERRCODE_DATA_CORRUPTED),
6941 : : errmsg_internal("found multixact %u from before relminmxid %u",
6942 : : multi, cutoffs->relminmxid)));
6943 [ + + ]: 7 : else if (MultiXactIdPrecedes(multi, cutoffs->OldestMxact))
6944 : : {
6945 : : TransactionId update_xact;
6946 : :
6947 : : /*
6948 : : * This old multi cannot possibly have members still running, but
6949 : : * verify just in case. If it was a locker only, it can be removed
6950 : : * without any further consideration; but if it contained an update,
6951 : : * we might need to preserve it.
6952 : : */
6953 [ - + ]: 5 : if (MultiXactIdIsRunning(multi,
6954 : 5 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask)))
6955 [ # # ]: 0 : ereport(ERROR,
6956 : : (errcode(ERRCODE_DATA_CORRUPTED),
6957 : : errmsg_internal("multixact %u from before multi freeze cutoff %u found to be still running",
6958 : : multi, cutoffs->OldestMxact)));
6959 : :
6960 [ + - ]: 5 : if (HEAP_XMAX_IS_LOCKED_ONLY(t_infomask))
6961 : : {
6962 : 5 : *flags |= FRM_INVALIDATE_XMAX;
6963 : 5 : pagefrz->freeze_required = true;
6964 : 5 : return InvalidTransactionId;
6965 : : }
6966 : :
6967 : : /* replace multi with single XID for its updater? */
6968 : 0 : update_xact = MultiXactIdGetUpdateXid(multi, t_infomask);
6969 [ # # ]: 0 : if (TransactionIdPrecedes(update_xact, cutoffs->relfrozenxid))
6970 [ # # ]: 0 : ereport(ERROR,
6971 : : (errcode(ERRCODE_DATA_CORRUPTED),
6972 : : errmsg_internal("multixact %u contains update XID %u from before relfrozenxid %u",
6973 : : multi, update_xact,
6974 : : cutoffs->relfrozenxid)));
6975 [ # # ]: 0 : else if (TransactionIdPrecedes(update_xact, cutoffs->OldestXmin))
6976 : : {
6977 : : /*
6978 : : * Updater XID has to have aborted (otherwise the tuple would have
6979 : : * been pruned away instead, since updater XID is < OldestXmin).
6980 : : * Just remove xmax.
6981 : : */
6982 [ # # ]: 0 : if (TransactionIdDidCommit(update_xact))
6983 [ # # ]: 0 : ereport(ERROR,
6984 : : (errcode(ERRCODE_DATA_CORRUPTED),
6985 : : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6986 : : multi, update_xact,
6987 : : cutoffs->OldestXmin)));
6988 : 0 : *flags |= FRM_INVALIDATE_XMAX;
6989 : 0 : pagefrz->freeze_required = true;
6990 : 0 : return InvalidTransactionId;
6991 : : }
6992 : :
6993 : : /* Have to keep updater XID as new xmax */
6994 : 0 : *flags |= FRM_RETURN_IS_XID;
6995 : 0 : pagefrz->freeze_required = true;
6996 : 0 : return update_xact;
6997 : : }
6998 : :
6999 : : /*
7000 : : * Some member(s) of this Multi may be below FreezeLimit xid cutoff, so we
7001 : : * need to walk the whole members array to figure out what to do, if
7002 : : * anything.
7003 : : */
7004 : : nmembers =
7005 : 2 : GetMultiXactIdMembers(multi, &members, false,
7006 : 2 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask));
7007 [ - + ]: 2 : if (nmembers <= 0)
7008 : : {
7009 : : /* Nothing worth keeping */
7010 : 0 : *flags |= FRM_INVALIDATE_XMAX;
7011 : 0 : pagefrz->freeze_required = true;
7012 : 0 : return InvalidTransactionId;
7013 : : }
7014 : :
7015 : : /*
7016 : : * The FRM_NOOP case is the only case where we might need to ratchet back
7017 : : * FreezePageRelfrozenXid or FreezePageRelminMxid. It is also the only
7018 : : * case where our caller might ratchet back its NoFreezePageRelfrozenXid
7019 : : * or NoFreezePageRelminMxid "no freeze" trackers to deal with a multi.
7020 : : * FRM_NOOP handling should result in the NewRelfrozenXid/NewRelminMxid
7021 : : * trackers managed by VACUUM being ratcheting back by xmax to the degree
7022 : : * required to make it safe to leave xmax undisturbed, independent of
7023 : : * whether or not page freezing is triggered somewhere else.
7024 : : *
7025 : : * Our policy is to force freezing in every case other than FRM_NOOP,
7026 : : * which obviates the need to maintain either set of trackers, anywhere.
7027 : : * Every other case will reliably execute a freeze plan for xmax that
7028 : : * either replaces xmax with an XID/MXID >= OldestXmin/OldestMxact, or
7029 : : * sets xmax to an InvalidTransactionId XID, rendering xmax fully frozen.
7030 : : * (VACUUM's NewRelfrozenXid/NewRelminMxid trackers are initialized with
7031 : : * OldestXmin/OldestMxact, so later values never need to be tracked here.)
7032 : : */
7033 : 2 : need_replace = false;
7034 : 2 : FreezePageRelfrozenXid = pagefrz->FreezePageRelfrozenXid;
7035 [ + + ]: 4 : for (int i = 0; i < nmembers; i++)
7036 : : {
7037 : 3 : TransactionId xid = members[i].xid;
7038 : :
7039 : : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
7040 : :
7041 [ + + ]: 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7042 : : {
7043 : : /* Can't violate the FreezeLimit postcondition */
7044 : 1 : need_replace = true;
7045 : 1 : break;
7046 : : }
7047 [ - + ]: 2 : if (TransactionIdPrecedes(xid, FreezePageRelfrozenXid))
7048 : 0 : FreezePageRelfrozenXid = xid;
7049 : : }
7050 : :
7051 : : /* Can't violate the MultiXactCutoff postcondition, either */
7052 [ + + ]: 2 : if (!need_replace)
7053 : 1 : need_replace = MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff);
7054 : :
7055 [ + + ]: 2 : if (!need_replace)
7056 : : {
7057 : : /*
7058 : : * vacuumlazy.c might ratchet back NewRelminMxid, NewRelfrozenXid, or
7059 : : * both together to make it safe to retain this particular multi after
7060 : : * freezing its page
7061 : : */
7062 : 1 : *flags |= FRM_NOOP;
7063 : 1 : pagefrz->FreezePageRelfrozenXid = FreezePageRelfrozenXid;
7064 [ - + ]: 1 : if (MultiXactIdPrecedes(multi, pagefrz->FreezePageRelminMxid))
7065 : 0 : pagefrz->FreezePageRelminMxid = multi;
7066 : 1 : pfree(members);
7067 : 1 : return multi;
7068 : : }
7069 : :
7070 : : /*
7071 : : * Do a more thorough second pass over the multi to figure out which
7072 : : * member XIDs actually need to be kept. Checking the precise status of
7073 : : * individual members might even show that we don't need to keep anything.
7074 : : * That is quite possible even though the Multi must be >= OldestMxact,
7075 : : * since our second pass only keeps member XIDs when it's truly necessary;
7076 : : * even member XIDs >= OldestXmin often won't be kept by second pass.
7077 : : */
7078 : 1 : nnewmembers = 0;
7079 : 1 : newmembers = palloc_array(MultiXactMember, nmembers);
7080 : 1 : has_lockers = false;
7081 : 1 : update_xid = InvalidTransactionId;
7082 : 1 : update_committed = false;
7083 : :
7084 : : /*
7085 : : * Determine whether to keep each member xid, or to ignore it instead
7086 : : */
7087 [ + + ]: 3 : for (int i = 0; i < nmembers; i++)
7088 : : {
7089 : 2 : TransactionId xid = members[i].xid;
7090 : 2 : MultiXactStatus mstatus = members[i].status;
7091 : :
7092 : : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
7093 : :
7094 [ + - ]: 2 : if (!ISUPDATE_from_mxstatus(mstatus))
7095 : : {
7096 : : /*
7097 : : * Locker XID (not updater XID). We only keep lockers that are
7098 : : * still running.
7099 : : */
7100 [ + - + + ]: 4 : if (TransactionIdIsCurrentTransactionId(xid) ||
7101 : 2 : TransactionIdIsInProgress(xid))
7102 : : {
7103 [ - + ]: 1 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
7104 [ # # ]: 0 : ereport(ERROR,
7105 : : (errcode(ERRCODE_DATA_CORRUPTED),
7106 : : errmsg_internal("multixact %u contains running locker XID %u from before removable cutoff %u",
7107 : : multi, xid,
7108 : : cutoffs->OldestXmin)));
7109 : 1 : newmembers[nnewmembers++] = members[i];
7110 : 1 : has_lockers = true;
7111 : : }
7112 : :
7113 : 2 : continue;
7114 : : }
7115 : :
7116 : : /*
7117 : : * Updater XID (not locker XID). Should we keep it?
7118 : : *
7119 : : * Since the tuple wasn't totally removed when vacuum pruned, the
7120 : : * update Xid cannot possibly be older than OldestXmin cutoff unless
7121 : : * the updater XID aborted. If the updater transaction is known
7122 : : * aborted or crashed then it's okay to ignore it, otherwise not.
7123 : : *
7124 : : * In any case the Multi should never contain two updaters, whatever
7125 : : * their individual commit status. Check for that first, in passing.
7126 : : */
7127 [ # # ]: 0 : if (TransactionIdIsValid(update_xid))
7128 [ # # ]: 0 : ereport(ERROR,
7129 : : (errcode(ERRCODE_DATA_CORRUPTED),
7130 : : errmsg_internal("multixact %u has two or more updating members",
7131 : : multi),
7132 : : errdetail_internal("First updater XID=%u second updater XID=%u.",
7133 : : update_xid, xid)));
7134 : :
7135 : : /*
7136 : : * As with all tuple visibility routines, it's critical to test
7137 : : * TransactionIdIsInProgress before TransactionIdDidCommit, because of
7138 : : * race conditions explained in detail in heapam_visibility.c.
7139 : : */
7140 [ # # # # ]: 0 : if (TransactionIdIsCurrentTransactionId(xid) ||
7141 : 0 : TransactionIdIsInProgress(xid))
7142 : 0 : update_xid = xid;
7143 [ # # ]: 0 : else if (TransactionIdDidCommit(xid))
7144 : : {
7145 : : /*
7146 : : * The transaction committed, so we can tell caller to set
7147 : : * HEAP_XMAX_COMMITTED. (We can only do this because we know the
7148 : : * transaction is not running.)
7149 : : */
7150 : 0 : update_committed = true;
7151 : 0 : update_xid = xid;
7152 : : }
7153 : : else
7154 : : {
7155 : : /*
7156 : : * Not in progress, not committed -- must be aborted or crashed;
7157 : : * we can ignore it.
7158 : : */
7159 : 0 : continue;
7160 : : }
7161 : :
7162 : : /*
7163 : : * We determined that updater must be kept -- add it to pending new
7164 : : * members list
7165 : : */
7166 [ # # ]: 0 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
7167 [ # # ]: 0 : ereport(ERROR,
7168 : : (errcode(ERRCODE_DATA_CORRUPTED),
7169 : : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
7170 : : multi, xid, cutoffs->OldestXmin)));
7171 : 0 : newmembers[nnewmembers++] = members[i];
7172 : : }
7173 : :
7174 : 1 : pfree(members);
7175 : :
7176 : : /*
7177 : : * Determine what to do with caller's multi based on information gathered
7178 : : * during our second pass
7179 : : */
7180 [ - + ]: 1 : if (nnewmembers == 0)
7181 : : {
7182 : : /* Nothing worth keeping */
7183 : 0 : *flags |= FRM_INVALIDATE_XMAX;
7184 : 0 : newxmax = InvalidTransactionId;
7185 : : }
7186 [ - + - - ]: 1 : else if (TransactionIdIsValid(update_xid) && !has_lockers)
7187 : : {
7188 : : /*
7189 : : * If there's a single member and it's an update, pass it back alone
7190 : : * without creating a new Multi. (XXX we could do this when there's a
7191 : : * single remaining locker, too, but that would complicate the API too
7192 : : * much; moreover, the case with the single updater is more
7193 : : * interesting, because those are longer-lived.)
7194 : : */
7195 : : Assert(nnewmembers == 1);
7196 : 0 : *flags |= FRM_RETURN_IS_XID;
7197 [ # # ]: 0 : if (update_committed)
7198 : 0 : *flags |= FRM_MARK_COMMITTED;
7199 : 0 : newxmax = update_xid;
7200 : : }
7201 : : else
7202 : : {
7203 : : /*
7204 : : * Create a new multixact with the surviving members of the previous
7205 : : * one, to set as new Xmax in the tuple
7206 : : */
7207 : 1 : newxmax = MultiXactIdCreateFromMembers(nnewmembers, newmembers);
7208 : 1 : *flags |= FRM_RETURN_IS_MULTI;
7209 : : }
7210 : :
7211 : 1 : pfree(newmembers);
7212 : :
7213 : 1 : pagefrz->freeze_required = true;
7214 : 1 : return newxmax;
7215 : : }
7216 : :
7217 : : /*
7218 : : * heap_prepare_freeze_tuple
7219 : : *
7220 : : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7221 : : * are older than the OldestXmin and/or OldestMxact freeze cutoffs. If so,
7222 : : * setup enough state (in the *frz output argument) to enable caller to
7223 : : * process this tuple as part of freezing its page, and return true. Return
7224 : : * false if nothing can be changed about the tuple right now.
7225 : : *
7226 : : * FreezePageConflictXid is advanced only for xmin/xvac freezing, not for xmax
7227 : : * changes. We only remove xmax state here when it is lock-only, or when the
7228 : : * updater XID (including an updater member of a MultiXact) must be aborted;
7229 : : * otherwise, the tuple would already be removable. Neither case affects
7230 : : * visibility on a standby.
7231 : : *
7232 : : * Also sets *totally_frozen to true if the tuple will be totally frozen once
7233 : : * caller executes returned freeze plan (or if the tuple was already totally
7234 : : * frozen by an earlier VACUUM). This indicates that there are no remaining
7235 : : * XIDs or MultiXactIds that will need to be processed by a future VACUUM.
7236 : : *
7237 : : * VACUUM caller must assemble HeapTupleFreeze freeze plan entries for every
7238 : : * tuple that we returned true for, and then execute freezing. Caller must
7239 : : * initialize pagefrz fields for page as a whole before first call here for
7240 : : * each heap page.
7241 : : *
7242 : : * VACUUM caller decides on whether or not to freeze the page as a whole.
7243 : : * We'll often prepare freeze plans for a page that caller just discards.
7244 : : * However, VACUUM doesn't always get to make a choice; it must freeze when
7245 : : * pagefrz.freeze_required is set, to ensure that any XIDs < FreezeLimit (and
7246 : : * MXIDs < MultiXactCutoff) can never be left behind. We help to make sure
7247 : : * that VACUUM always follows that rule.
7248 : : *
7249 : : * We sometimes force freezing of xmax MultiXactId values long before it is
7250 : : * strictly necessary to do so just to ensure the FreezeLimit postcondition.
7251 : : * It's worth processing MultiXactIds proactively when it is cheap to do so,
7252 : : * and it's convenient to make that happen by piggy-backing it on the "force
7253 : : * freezing" mechanism. Conversely, we sometimes delay freezing MultiXactIds
7254 : : * because it is expensive right now (though only when it's still possible to
7255 : : * do so without violating the FreezeLimit/MultiXactCutoff postcondition).
7256 : : *
7257 : : * It is assumed that the caller has checked the tuple with
7258 : : * HeapTupleSatisfiesVacuum() and determined that it is not HEAPTUPLE_DEAD
7259 : : * (else we should be removing the tuple, not freezing it).
7260 : : *
7261 : : * NB: This function has side effects: it might allocate a new MultiXactId.
7262 : : * It will be set as tuple's new xmax when our *frz output is processed within
7263 : : * heap_execute_freeze_tuple later on. If the tuple is in a shared buffer
7264 : : * then caller had better have an exclusive lock on it already.
7265 : : */
7266 : : bool
7267 : 12609254 : heap_prepare_freeze_tuple(HeapTupleHeader tuple,
7268 : : const struct VacuumCutoffs *cutoffs,
7269 : : HeapPageFreeze *pagefrz,
7270 : : HeapTupleFreeze *frz, bool *totally_frozen)
7271 : : {
7272 : 12609254 : bool xmin_already_frozen = false,
7273 : 12609254 : xmax_already_frozen = false;
7274 : 12609254 : bool freeze_xmin = false,
7275 : 12609254 : replace_xvac = false,
7276 : 12609254 : replace_xmax = false,
7277 : 12609254 : freeze_xmax = false;
7278 : : TransactionId xid;
7279 : :
7280 : 12609254 : frz->xmax = HeapTupleHeaderGetRawXmax(tuple);
7281 : 12609254 : frz->t_infomask2 = tuple->t_infomask2;
7282 : 12609254 : frz->t_infomask = tuple->t_infomask;
7283 : 12609254 : frz->frzflags = 0;
7284 : 12609254 : frz->checkflags = 0;
7285 : :
7286 : : /*
7287 : : * Process xmin, while keeping track of whether it's already frozen, or
7288 : : * will become frozen iff our freeze plan is executed by caller (could be
7289 : : * neither).
7290 : : */
7291 : 12609254 : xid = HeapTupleHeaderGetXmin(tuple);
7292 [ + + ]: 12609254 : if (!TransactionIdIsNormal(xid))
7293 : 8491433 : xmin_already_frozen = true;
7294 : : else
7295 : : {
7296 [ - + ]: 4117821 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7297 [ # # ]: 0 : ereport(ERROR,
7298 : : (errcode(ERRCODE_DATA_CORRUPTED),
7299 : : errmsg_internal("found xmin %u from before relfrozenxid %u",
7300 : : xid, cutoffs->relfrozenxid)));
7301 : :
7302 : : /* Will set freeze_xmin flags in freeze plan below */
7303 : 4117821 : freeze_xmin = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7304 : :
7305 : : /* Verify that xmin committed if and when freeze plan is executed */
7306 [ + + ]: 4117821 : if (freeze_xmin)
7307 : : {
7308 : 3405279 : frz->checkflags |= HEAP_FREEZE_CHECK_XMIN_COMMITTED;
7309 [ + + ]: 3405279 : if (TransactionIdFollows(xid, pagefrz->FreezePageConflictXid))
7310 : 514159 : pagefrz->FreezePageConflictXid = xid;
7311 : : }
7312 : : }
7313 : :
7314 : : /*
7315 : : * Old-style VACUUM FULL is gone, but we have to process xvac for as long
7316 : : * as we support having MOVED_OFF/MOVED_IN tuples in the database
7317 : : */
7318 : 12609254 : xid = HeapTupleHeaderGetXvac(tuple);
7319 [ - + ]: 12609254 : if (TransactionIdIsNormal(xid))
7320 : : {
7321 : : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7322 : : Assert(TransactionIdPrecedes(xid, cutoffs->OldestXmin));
7323 : :
7324 : : /*
7325 : : * For Xvac, we always freeze proactively. This allows totally_frozen
7326 : : * tracking to ignore xvac.
7327 : : */
7328 : 0 : replace_xvac = pagefrz->freeze_required = true;
7329 : :
7330 [ # # ]: 0 : if (TransactionIdFollows(xid, pagefrz->FreezePageConflictXid))
7331 : 0 : pagefrz->FreezePageConflictXid = xid;
7332 : :
7333 : : /* Will set replace_xvac flags in freeze plan below */
7334 : : }
7335 : :
7336 : : /* Now process xmax */
7337 : 12609254 : xid = frz->xmax;
7338 [ + + ]: 12609254 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7339 : : {
7340 : : /* Raw xmax is a MultiXactId */
7341 : : TransactionId newxmax;
7342 : : uint16 flags;
7343 : :
7344 : : /*
7345 : : * We will either remove xmax completely (in the "freeze_xmax" path),
7346 : : * process xmax by replacing it (in the "replace_xmax" path), or
7347 : : * perform no-op xmax processing. The only constraint is that the
7348 : : * FreezeLimit/MultiXactCutoff postcondition must never be violated.
7349 : : */
7350 : 7 : newxmax = FreezeMultiXactId(xid, tuple->t_infomask, cutoffs,
7351 : : &flags, pagefrz);
7352 : :
7353 [ + + ]: 7 : if (flags & FRM_NOOP)
7354 : : {
7355 : : /*
7356 : : * xmax is a MultiXactId, and nothing about it changes for now.
7357 : : * This is the only case where 'freeze_required' won't have been
7358 : : * set for us by FreezeMultiXactId, as well as the only case where
7359 : : * neither freeze_xmax nor replace_xmax are set (given a multi).
7360 : : *
7361 : : * This is a no-op, but the call to FreezeMultiXactId might have
7362 : : * ratcheted back NewRelfrozenXid and/or NewRelminMxid trackers
7363 : : * for us (the "freeze page" variants, specifically). That'll
7364 : : * make it safe for our caller to freeze the page later on, while
7365 : : * leaving this particular xmax undisturbed.
7366 : : *
7367 : : * FreezeMultiXactId is _not_ responsible for the "no freeze"
7368 : : * NewRelfrozenXid/NewRelminMxid trackers, though -- that's our
7369 : : * job. A call to heap_tuple_should_freeze for this same tuple
7370 : : * will take place below if 'freeze_required' isn't set already.
7371 : : * (This repeats work from FreezeMultiXactId, but allows "no
7372 : : * freeze" tracker maintenance to happen in only one place.)
7373 : : */
7374 : : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->MultiXactCutoff));
7375 : : Assert(MultiXactIdIsValid(newxmax) && xid == newxmax);
7376 : : }
7377 [ - + ]: 6 : else if (flags & FRM_RETURN_IS_XID)
7378 : : {
7379 : : /*
7380 : : * xmax will become an updater Xid (original MultiXact's updater
7381 : : * member Xid will be carried forward as a simple Xid in Xmax).
7382 : : */
7383 : : Assert(!TransactionIdPrecedes(newxmax, cutoffs->OldestXmin));
7384 : :
7385 : : /*
7386 : : * NB -- some of these transformations are only valid because we
7387 : : * know the return Xid is a tuple updater (i.e. not merely a
7388 : : * locker.) Also note that the only reason we don't explicitly
7389 : : * worry about HEAP_KEYS_UPDATED is because it lives in
7390 : : * t_infomask2 rather than t_infomask.
7391 : : */
7392 : 0 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7393 : 0 : frz->xmax = newxmax;
7394 [ # # ]: 0 : if (flags & FRM_MARK_COMMITTED)
7395 : 0 : frz->t_infomask |= HEAP_XMAX_COMMITTED;
7396 : 0 : replace_xmax = true;
7397 : : }
7398 [ + + ]: 6 : else if (flags & FRM_RETURN_IS_MULTI)
7399 : : {
7400 : : uint16 newbits;
7401 : : uint16 newbits2;
7402 : :
7403 : : /*
7404 : : * xmax is an old MultiXactId that we have to replace with a new
7405 : : * MultiXactId, to carry forward two or more original member XIDs.
7406 : : */
7407 : : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->OldestMxact));
7408 : :
7409 : : /*
7410 : : * We can't use GetMultiXactIdHintBits directly on the new multi
7411 : : * here; that routine initializes the masks to all zeroes, which
7412 : : * would lose other bits we need. Doing it this way ensures all
7413 : : * unrelated bits remain untouched.
7414 : : */
7415 : 1 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7416 : 1 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7417 : 1 : GetMultiXactIdHintBits(newxmax, &newbits, &newbits2);
7418 : 1 : frz->t_infomask |= newbits;
7419 : 1 : frz->t_infomask2 |= newbits2;
7420 : 1 : frz->xmax = newxmax;
7421 : 1 : replace_xmax = true;
7422 : : }
7423 : : else
7424 : : {
7425 : : /*
7426 : : * Freeze plan for tuple "freezes xmax" in the strictest sense:
7427 : : * it'll leave nothing in xmax (neither an Xid nor a MultiXactId).
7428 : : */
7429 : : Assert(flags & FRM_INVALIDATE_XMAX);
7430 : : Assert(!TransactionIdIsValid(newxmax));
7431 : :
7432 : : /* Will set freeze_xmax flags in freeze plan below */
7433 : 5 : freeze_xmax = true;
7434 : : }
7435 : :
7436 : : /* MultiXactId processing forces freezing (barring FRM_NOOP case) */
7437 : : Assert(pagefrz->freeze_required || (!freeze_xmax && !replace_xmax));
7438 : : }
7439 [ + + ]: 12609247 : else if (TransactionIdIsNormal(xid))
7440 : : {
7441 : : /* Raw xmax is normal XID */
7442 [ - + ]: 5841007 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
7443 [ # # ]: 0 : ereport(ERROR,
7444 : : (errcode(ERRCODE_DATA_CORRUPTED),
7445 : : errmsg_internal("found xmax %u from before relfrozenxid %u",
7446 : : xid, cutoffs->relfrozenxid)));
7447 : :
7448 : : /* Will set freeze_xmax flags in freeze plan below */
7449 : 5841007 : freeze_xmax = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7450 : :
7451 : : /*
7452 : : * Verify that xmax aborted if and when freeze plan is executed,
7453 : : * provided it's from an update. (A lock-only xmax can be removed
7454 : : * independent of this, since the lock is released at xact end.)
7455 : : */
7456 [ + + + + ]: 5841007 : if (freeze_xmax && !HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask))
7457 : 1576 : frz->checkflags |= HEAP_FREEZE_CHECK_XMAX_ABORTED;
7458 : : }
7459 [ + - ]: 6768240 : else if (!TransactionIdIsValid(xid))
7460 : : {
7461 : : /* Raw xmax is InvalidTransactionId XID */
7462 : : Assert((tuple->t_infomask & HEAP_XMAX_IS_MULTI) == 0);
7463 : 6768240 : xmax_already_frozen = true;
7464 : : }
7465 : : else
7466 [ # # ]: 0 : ereport(ERROR,
7467 : : (errcode(ERRCODE_DATA_CORRUPTED),
7468 : : errmsg_internal("found raw xmax %u (infomask 0x%04x) not invalid and not multi",
7469 : : xid, tuple->t_infomask)));
7470 : :
7471 [ + + ]: 12609254 : if (freeze_xmin)
7472 : : {
7473 : : Assert(!xmin_already_frozen);
7474 : :
7475 : 3405279 : frz->t_infomask |= HEAP_XMIN_FROZEN;
7476 : : }
7477 [ - + ]: 12609254 : if (replace_xvac)
7478 : : {
7479 : : /*
7480 : : * If a MOVED_OFF tuple is not dead, the xvac transaction must have
7481 : : * failed; whereas a non-dead MOVED_IN tuple must mean the xvac
7482 : : * transaction succeeded.
7483 : : */
7484 : : Assert(pagefrz->freeze_required);
7485 [ # # ]: 0 : if (tuple->t_infomask & HEAP_MOVED_OFF)
7486 : 0 : frz->frzflags |= XLH_INVALID_XVAC;
7487 : : else
7488 : 0 : frz->frzflags |= XLH_FREEZE_XVAC;
7489 : : }
7490 : : if (replace_xmax)
7491 : : {
7492 : : Assert(!xmax_already_frozen && !freeze_xmax);
7493 : : Assert(pagefrz->freeze_required);
7494 : :
7495 : : /* Already set replace_xmax flags in freeze plan earlier */
7496 : : }
7497 [ + + ]: 12609254 : if (freeze_xmax)
7498 : : {
7499 : : Assert(!xmax_already_frozen && !replace_xmax);
7500 : :
7501 : 2466 : frz->xmax = InvalidTransactionId;
7502 : :
7503 : : /*
7504 : : * The tuple might be marked either XMAX_INVALID or XMAX_COMMITTED +
7505 : : * LOCKED. Normalize to INVALID just to be sure no one gets confused.
7506 : : * Also get rid of the HEAP_KEYS_UPDATED bit.
7507 : : */
7508 : 2466 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7509 : 2466 : frz->t_infomask |= HEAP_XMAX_INVALID;
7510 : 2466 : frz->t_infomask2 &= ~HEAP_HOT_UPDATED;
7511 : 2466 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7512 : : }
7513 : :
7514 : : /*
7515 : : * Determine if this tuple is already totally frozen, or will become
7516 : : * totally frozen (provided caller executes freeze plans for the page)
7517 : : */
7518 [ + + + + : 24503500 : *totally_frozen = ((freeze_xmin || xmin_already_frozen) &&
+ + ]
7519 [ + + ]: 11894246 : (freeze_xmax || xmax_already_frozen));
7520 : :
7521 [ + + + + : 12609254 : if (!pagefrz->freeze_required && !(xmin_already_frozen &&
+ + ]
7522 : : xmax_already_frozen))
7523 : : {
7524 : : /*
7525 : : * So far no previous tuple from the page made freezing mandatory.
7526 : : * Does this tuple force caller to freeze the entire page?
7527 : : */
7528 : 8115165 : pagefrz->freeze_required =
7529 : 8115165 : heap_tuple_should_freeze(tuple, cutoffs,
7530 : : &pagefrz->NoFreezePageRelfrozenXid,
7531 : : &pagefrz->NoFreezePageRelminMxid);
7532 : : }
7533 : :
7534 : : /* Tell caller if this tuple has a usable freeze plan set in *frz */
7535 [ + + + - : 12609254 : return freeze_xmin || replace_xvac || replace_xmax || freeze_xmax;
+ - + + ]
7536 : : }
7537 : :
7538 : : /*
7539 : : * Perform xmin/xmax XID status sanity checks before actually executing freeze
7540 : : * plans.
7541 : : *
7542 : : * heap_prepare_freeze_tuple doesn't perform these checks directly because
7543 : : * pg_xact lookups are relatively expensive. They shouldn't be repeated by
7544 : : * successive VACUUMs that each decide against freezing the same page.
7545 : : */
7546 : : void
7547 : 26306 : heap_pre_freeze_checks(Buffer buffer,
7548 : : HeapTupleFreeze *tuples, int ntuples)
7549 : : {
7550 : 26306 : Page page = BufferGetPage(buffer);
7551 : :
7552 [ + + ]: 1240239 : for (int i = 0; i < ntuples; i++)
7553 : : {
7554 : 1213933 : HeapTupleFreeze *frz = tuples + i;
7555 : 1213933 : ItemId itemid = PageGetItemId(page, frz->offset);
7556 : : HeapTupleHeader htup;
7557 : :
7558 : 1213933 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7559 : :
7560 : : /* Deliberately avoid relying on tuple hint bits here */
7561 [ + + ]: 1213933 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMIN_COMMITTED)
7562 : : {
7563 : 1213932 : TransactionId xmin = HeapTupleHeaderGetRawXmin(htup);
7564 : :
7565 : : Assert(!HeapTupleHeaderXminFrozen(htup));
7566 [ - + ]: 1213932 : if (unlikely(!TransactionIdDidCommit(xmin)))
7567 [ # # ]: 0 : ereport(ERROR,
7568 : : (errcode(ERRCODE_DATA_CORRUPTED),
7569 : : errmsg_internal("uncommitted xmin %u needs to be frozen",
7570 : : xmin)));
7571 : : }
7572 : :
7573 : : /*
7574 : : * TransactionIdDidAbort won't work reliably in the presence of XIDs
7575 : : * left behind by transactions that were in progress during a crash,
7576 : : * so we can only check that xmax didn't commit
7577 : : */
7578 [ + + ]: 1213933 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMAX_ABORTED)
7579 : : {
7580 : 392 : TransactionId xmax = HeapTupleHeaderGetRawXmax(htup);
7581 : :
7582 : : Assert(TransactionIdIsNormal(xmax));
7583 [ - + ]: 392 : if (unlikely(TransactionIdDidCommit(xmax)))
7584 [ # # ]: 0 : ereport(ERROR,
7585 : : (errcode(ERRCODE_DATA_CORRUPTED),
7586 : : errmsg_internal("cannot freeze committed xmax %u",
7587 : : xmax)));
7588 : : }
7589 : : }
7590 : 26306 : }
7591 : :
7592 : : /*
7593 : : * Helper which executes freezing of one or more heap tuples on a page on
7594 : : * behalf of caller. Caller passes an array of tuple plans from
7595 : : * heap_prepare_freeze_tuple. Caller must set 'offset' in each plan for us.
7596 : : * Must be called in a critical section that also marks the buffer dirty and,
7597 : : * if needed, emits WAL.
7598 : : */
7599 : : void
7600 : 26306 : heap_freeze_prepared_tuples(Buffer buffer, HeapTupleFreeze *tuples, int ntuples)
7601 : : {
7602 : 26306 : Page page = BufferGetPage(buffer);
7603 : :
7604 [ + + ]: 1240239 : for (int i = 0; i < ntuples; i++)
7605 : : {
7606 : 1213933 : HeapTupleFreeze *frz = tuples + i;
7607 : 1213933 : ItemId itemid = PageGetItemId(page, frz->offset);
7608 : : HeapTupleHeader htup;
7609 : :
7610 : 1213933 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7611 : 1213933 : heap_execute_freeze_tuple(htup, frz);
7612 : : }
7613 : 26306 : }
7614 : :
7615 : : /*
7616 : : * heap_freeze_tuple
7617 : : * Freeze tuple in place, without WAL logging.
7618 : : *
7619 : : * Useful for callers like CLUSTER that perform their own WAL logging.
7620 : : */
7621 : : bool
7622 : 460368 : heap_freeze_tuple(HeapTupleHeader tuple,
7623 : : TransactionId relfrozenxid, TransactionId relminmxid,
7624 : : TransactionId FreezeLimit, TransactionId MultiXactCutoff)
7625 : : {
7626 : : HeapTupleFreeze frz;
7627 : : bool do_freeze;
7628 : : bool totally_frozen;
7629 : : struct VacuumCutoffs cutoffs;
7630 : : HeapPageFreeze pagefrz;
7631 : :
7632 : 460368 : cutoffs.relfrozenxid = relfrozenxid;
7633 : 460368 : cutoffs.relminmxid = relminmxid;
7634 : 460368 : cutoffs.OldestXmin = FreezeLimit;
7635 : 460368 : cutoffs.OldestMxact = MultiXactCutoff;
7636 : 460368 : cutoffs.FreezeLimit = FreezeLimit;
7637 : 460368 : cutoffs.MultiXactCutoff = MultiXactCutoff;
7638 : :
7639 : 460368 : pagefrz.freeze_required = true;
7640 : 460368 : pagefrz.FreezePageRelfrozenXid = FreezeLimit;
7641 : 460368 : pagefrz.FreezePageRelminMxid = MultiXactCutoff;
7642 : 460368 : pagefrz.FreezePageConflictXid = InvalidTransactionId;
7643 : 460368 : pagefrz.NoFreezePageRelfrozenXid = FreezeLimit;
7644 : 460368 : pagefrz.NoFreezePageRelminMxid = MultiXactCutoff;
7645 : :
7646 : 460368 : do_freeze = heap_prepare_freeze_tuple(tuple, &cutoffs,
7647 : : &pagefrz, &frz, &totally_frozen);
7648 : :
7649 : : /*
7650 : : * Note that because this is not a WAL-logged operation, we don't need to
7651 : : * fill in the offset in the freeze record.
7652 : : */
7653 : :
7654 [ + + ]: 460368 : if (do_freeze)
7655 : 377315 : heap_execute_freeze_tuple(tuple, &frz);
7656 : 460368 : return do_freeze;
7657 : : }
7658 : :
7659 : : /*
7660 : : * For a given MultiXactId, return the hint bits that should be set in the
7661 : : * tuple's infomask.
7662 : : *
7663 : : * Normally this should be called for a multixact that was just created, and
7664 : : * so is on our local cache, so the GetMembers call is fast.
7665 : : */
7666 : : static void
7667 : 76878 : GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
7668 : : uint16 *new_infomask2)
7669 : : {
7670 : : int nmembers;
7671 : : MultiXactMember *members;
7672 : : int i;
7673 : 76878 : uint16 bits = HEAP_XMAX_IS_MULTI;
7674 : 76878 : uint16 bits2 = 0;
7675 : 76878 : bool has_update = false;
7676 : 76878 : LockTupleMode strongest = LockTupleKeyShare;
7677 : :
7678 : : /*
7679 : : * We only use this in multis we just created, so they cannot be values
7680 : : * pre-pg_upgrade.
7681 : : */
7682 : 76878 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
7683 : :
7684 [ + + ]: 1472844 : for (i = 0; i < nmembers; i++)
7685 : : {
7686 : : LockTupleMode mode;
7687 : :
7688 : : /*
7689 : : * Remember the strongest lock mode held by any member of the
7690 : : * multixact.
7691 : : */
7692 : 1395966 : mode = TUPLOCK_from_mxstatus(members[i].status);
7693 [ + + ]: 1395966 : if (mode > strongest)
7694 : 2930 : strongest = mode;
7695 : :
7696 : : /* See what other bits we need */
7697 [ + + + + : 1395966 : switch (members[i].status)
- ]
7698 : : {
7699 : 1393544 : case MultiXactStatusForKeyShare:
7700 : : case MultiXactStatusForShare:
7701 : : case MultiXactStatusForNoKeyUpdate:
7702 : 1393544 : break;
7703 : :
7704 : 53 : case MultiXactStatusForUpdate:
7705 : 53 : bits2 |= HEAP_KEYS_UPDATED;
7706 : 53 : break;
7707 : :
7708 : 2359 : case MultiXactStatusNoKeyUpdate:
7709 : 2359 : has_update = true;
7710 : 2359 : break;
7711 : :
7712 : 10 : case MultiXactStatusUpdate:
7713 : 10 : bits2 |= HEAP_KEYS_UPDATED;
7714 : 10 : has_update = true;
7715 : 10 : break;
7716 : : }
7717 : : }
7718 : :
7719 [ + + + + ]: 76878 : if (strongest == LockTupleExclusive ||
7720 : : strongest == LockTupleNoKeyExclusive)
7721 : 2450 : bits |= HEAP_XMAX_EXCL_LOCK;
7722 [ + + ]: 74428 : else if (strongest == LockTupleShare)
7723 : 477 : bits |= HEAP_XMAX_SHR_LOCK;
7724 [ + - ]: 73951 : else if (strongest == LockTupleKeyShare)
7725 : 73951 : bits |= HEAP_XMAX_KEYSHR_LOCK;
7726 : :
7727 [ + + ]: 76878 : if (!has_update)
7728 : 74509 : bits |= HEAP_XMAX_LOCK_ONLY;
7729 : :
7730 [ + - ]: 76878 : if (nmembers > 0)
7731 : 76878 : pfree(members);
7732 : :
7733 : 76878 : *new_infomask = bits;
7734 : 76878 : *new_infomask2 = bits2;
7735 : 76878 : }
7736 : :
7737 : : /*
7738 : : * MultiXactIdGetUpdateXid
7739 : : *
7740 : : * Given a multixact Xmax and corresponding infomask, which does not have the
7741 : : * HEAP_XMAX_LOCK_ONLY bit set, obtain and return the Xid of the updating
7742 : : * transaction.
7743 : : *
7744 : : * Caller is expected to check the status of the updating transaction, if
7745 : : * necessary.
7746 : : */
7747 : : static TransactionId
7748 : 162094 : MultiXactIdGetUpdateXid(TransactionId xmax, uint16 t_infomask)
7749 : : {
7750 : 162094 : TransactionId update_xact = InvalidTransactionId;
7751 : : MultiXactMember *members;
7752 : : int nmembers;
7753 : :
7754 : : Assert(!(t_infomask & HEAP_XMAX_LOCK_ONLY));
7755 : : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
7756 : :
7757 : : /*
7758 : : * Since we know the LOCK_ONLY bit is not set, this cannot be a multi from
7759 : : * pre-pg_upgrade.
7760 : : */
7761 : 162094 : nmembers = GetMultiXactIdMembers(xmax, &members, false, false);
7762 : :
7763 [ + - ]: 162094 : if (nmembers > 0)
7764 : : {
7765 : : int i;
7766 : :
7767 [ + - ]: 245995 : for (i = 0; i < nmembers; i++)
7768 : : {
7769 : : /* Ignore lockers */
7770 [ + + ]: 245995 : if (!ISUPDATE_from_mxstatus(members[i].status))
7771 : 83901 : continue;
7772 : :
7773 : : /* there can be at most one updater */
7774 : : Assert(update_xact == InvalidTransactionId);
7775 : 162094 : update_xact = members[i].xid;
7776 : : #ifndef USE_ASSERT_CHECKING
7777 : :
7778 : : /*
7779 : : * in an assert-enabled build, walk the whole array to ensure
7780 : : * there's no other updater.
7781 : : */
7782 : 162094 : break;
7783 : : #endif
7784 : : }
7785 : :
7786 : 162094 : pfree(members);
7787 : : }
7788 : :
7789 : 162094 : return update_xact;
7790 : : }
7791 : :
7792 : : /*
7793 : : * HeapTupleGetUpdateXid
7794 : : * As above, but use a HeapTupleHeader
7795 : : *
7796 : : * See also HeapTupleHeaderGetUpdateXid, which can be used without previously
7797 : : * checking the hint bits.
7798 : : */
7799 : : TransactionId
7800 : 159953 : HeapTupleGetUpdateXid(const HeapTupleHeaderData *tup)
7801 : : {
7802 : 159953 : return MultiXactIdGetUpdateXid(HeapTupleHeaderGetRawXmax(tup),
7803 : 159953 : tup->t_infomask);
7804 : : }
7805 : :
7806 : : /*
7807 : : * Does the given multixact conflict with the current transaction grabbing a
7808 : : * tuple lock of the given strength?
7809 : : *
7810 : : * The passed infomask pairs up with the given multixact in the tuple header.
7811 : : *
7812 : : * If current_is_member is not NULL, it is set to 'true' if the current
7813 : : * transaction is a member of the given multixact.
7814 : : */
7815 : : static bool
7816 : 218 : DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
7817 : : LockTupleMode lockmode, bool *current_is_member)
7818 : : {
7819 : : int nmembers;
7820 : : MultiXactMember *members;
7821 : 218 : bool result = false;
7822 : 218 : LOCKMODE wanted = tupleLockExtraInfo[lockmode].hwlock;
7823 : :
7824 [ - + ]: 218 : if (HEAP_LOCKED_UPGRADED(infomask))
7825 : 0 : return false;
7826 : :
7827 : 218 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7828 : 218 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7829 [ + - ]: 218 : if (nmembers >= 0)
7830 : : {
7831 : : int i;
7832 : :
7833 [ + + ]: 2682 : for (i = 0; i < nmembers; i++)
7834 : : {
7835 : : TransactionId memxid;
7836 : : LOCKMODE memlockmode;
7837 : :
7838 [ + + + + : 2471 : if (result && (current_is_member == NULL || *current_is_member))
+ - ]
7839 : : break;
7840 : :
7841 : 2464 : memlockmode = LOCKMODE_from_mxstatus(members[i].status);
7842 : :
7843 : : /* ignore members from current xact (but track their presence) */
7844 : 2464 : memxid = members[i].xid;
7845 [ + + ]: 2464 : if (TransactionIdIsCurrentTransactionId(memxid))
7846 : : {
7847 [ + + ]: 92 : if (current_is_member != NULL)
7848 : 78 : *current_is_member = true;
7849 : 92 : continue;
7850 : : }
7851 [ + + ]: 2372 : else if (result)
7852 : 8 : continue;
7853 : :
7854 : : /* ignore members that don't conflict with the lock we want */
7855 [ + + ]: 2364 : if (!DoLockModesConflict(memlockmode, wanted))
7856 : 2325 : continue;
7857 : :
7858 [ + + ]: 39 : if (ISUPDATE_from_mxstatus(members[i].status))
7859 : : {
7860 : : /* ignore aborted updaters */
7861 [ + + ]: 17 : if (TransactionIdDidAbort(memxid))
7862 : 1 : continue;
7863 : : }
7864 : : else
7865 : : {
7866 : : /* ignore lockers-only that are no longer in progress */
7867 [ + + ]: 22 : if (!TransactionIdIsInProgress(memxid))
7868 : 7 : continue;
7869 : : }
7870 : :
7871 : : /*
7872 : : * Whatever remains are either live lockers that conflict with our
7873 : : * wanted lock, and updaters that are not aborted. Those conflict
7874 : : * with what we want. Set up to return true, but keep going to
7875 : : * look for the current transaction among the multixact members,
7876 : : * if needed.
7877 : : */
7878 : 31 : result = true;
7879 : : }
7880 : 218 : pfree(members);
7881 : : }
7882 : :
7883 : 218 : return result;
7884 : : }
7885 : :
7886 : : /*
7887 : : * Do_MultiXactIdWait
7888 : : * Actual implementation for the two functions below.
7889 : : *
7890 : : * 'multi', 'status' and 'infomask' indicate what to sleep on (the status is
7891 : : * needed to ensure we only sleep on conflicting members, and the infomask is
7892 : : * used to optimize multixact access in case it's a lock-only multi); 'nowait'
7893 : : * indicates whether to use conditional lock acquisition, to allow callers to
7894 : : * fail if lock is unavailable. 'rel', 'ctid' and 'oper' are used to set up
7895 : : * context information for error messages. 'remaining', if not NULL, receives
7896 : : * the number of members that are still running, including any (non-aborted)
7897 : : * subtransactions of our own transaction. 'logLockFailure' indicates whether
7898 : : * to log details when a lock acquisition fails with 'nowait' enabled.
7899 : : *
7900 : : * We do this by sleeping on each member using XactLockTableWait. Any
7901 : : * members that belong to the current backend are *not* waited for, however;
7902 : : * this would not merely be useless but would lead to Assert failure inside
7903 : : * XactLockTableWait. By the time this returns, it is certain that all
7904 : : * transactions *of other backends* that were members of the MultiXactId
7905 : : * that conflict with the requested status are dead (and no new ones can have
7906 : : * been added, since it is not legal to add members to an existing
7907 : : * MultiXactId).
7908 : : *
7909 : : * But by the time we finish sleeping, someone else may have changed the Xmax
7910 : : * of the containing tuple, so the caller needs to iterate on us somehow.
7911 : : *
7912 : : * Note that in case we return false, the number of remaining members is
7913 : : * not to be trusted.
7914 : : */
7915 : : static bool
7916 : 61 : Do_MultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7917 : : uint16 infomask, bool nowait,
7918 : : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7919 : : int *remaining, bool logLockFailure)
7920 : : {
7921 : 61 : bool result = true;
7922 : : MultiXactMember *members;
7923 : : int nmembers;
7924 : 61 : int remain = 0;
7925 : :
7926 : : /* for pre-pg_upgrade tuples, no need to sleep at all */
7927 [ + - ]: 61 : nmembers = HEAP_LOCKED_UPGRADED(infomask) ? -1 :
7928 : 61 : GetMultiXactIdMembers(multi, &members, false,
7929 : 61 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7930 : :
7931 [ + - ]: 61 : if (nmembers >= 0)
7932 : : {
7933 : : int i;
7934 : :
7935 [ + + ]: 191 : for (i = 0; i < nmembers; i++)
7936 : : {
7937 : 136 : TransactionId memxid = members[i].xid;
7938 : 136 : MultiXactStatus memstatus = members[i].status;
7939 : :
7940 [ + + ]: 136 : if (TransactionIdIsCurrentTransactionId(memxid))
7941 : : {
7942 : 25 : remain++;
7943 : 25 : continue;
7944 : : }
7945 : :
7946 [ + + ]: 111 : if (!DoLockModesConflict(LOCKMODE_from_mxstatus(memstatus),
7947 : 111 : LOCKMODE_from_mxstatus(status)))
7948 : : {
7949 [ + + + - ]: 22 : if (remaining && TransactionIdIsInProgress(memxid))
7950 : 8 : remain++;
7951 : 22 : continue;
7952 : : }
7953 : :
7954 : : /*
7955 : : * This member conflicts with our multi, so we have to sleep (or
7956 : : * return failure, if asked to avoid waiting.)
7957 : : *
7958 : : * Note that we don't set up an error context callback ourselves,
7959 : : * but instead we pass the info down to XactLockTableWait. This
7960 : : * might seem a bit wasteful because the context is set up and
7961 : : * tore down for each member of the multixact, but in reality it
7962 : : * should be barely noticeable, and it avoids duplicate code.
7963 : : */
7964 [ + + ]: 89 : if (nowait)
7965 : : {
7966 : 6 : result = ConditionalXactLockTableWait(memxid, logLockFailure);
7967 [ + - ]: 6 : if (!result)
7968 : 6 : break;
7969 : : }
7970 : : else
7971 : 83 : XactLockTableWait(memxid, rel, ctid, oper);
7972 : : }
7973 : :
7974 : 61 : pfree(members);
7975 : : }
7976 : :
7977 [ + + ]: 61 : if (remaining)
7978 : 10 : *remaining = remain;
7979 : :
7980 : 61 : return result;
7981 : : }
7982 : :
7983 : : /*
7984 : : * MultiXactIdWait
7985 : : * Sleep on a MultiXactId.
7986 : : *
7987 : : * By the time we finish sleeping, someone else may have changed the Xmax
7988 : : * of the containing tuple, so the caller needs to iterate on us somehow.
7989 : : *
7990 : : * We return (in *remaining, if not NULL) the number of members that are still
7991 : : * running, including any (non-aborted) subtransactions of our own transaction.
7992 : : */
7993 : : static void
7994 : 55 : MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
7995 : : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7996 : : int *remaining)
7997 : : {
7998 : 55 : (void) Do_MultiXactIdWait(multi, status, infomask, false,
7999 : : rel, ctid, oper, remaining, false);
8000 : 55 : }
8001 : :
8002 : : /*
8003 : : * ConditionalMultiXactIdWait
8004 : : * As above, but only lock if we can get the lock without blocking.
8005 : : *
8006 : : * By the time we finish sleeping, someone else may have changed the Xmax
8007 : : * of the containing tuple, so the caller needs to iterate on us somehow.
8008 : : *
8009 : : * If the multixact is now all gone, return true. Returns false if some
8010 : : * transactions might still be running.
8011 : : *
8012 : : * We return (in *remaining, if not NULL) the number of members that are still
8013 : : * running, including any (non-aborted) subtransactions of our own transaction.
8014 : : */
8015 : : static bool
8016 : 6 : ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
8017 : : uint16 infomask, Relation rel, int *remaining,
8018 : : bool logLockFailure)
8019 : : {
8020 : 6 : return Do_MultiXactIdWait(multi, status, infomask, true,
8021 : : rel, NULL, XLTW_None, remaining, logLockFailure);
8022 : : }
8023 : :
8024 : : /*
8025 : : * heap_tuple_needs_eventual_freeze
8026 : : *
8027 : : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
8028 : : * will eventually require freezing (if tuple isn't removed by pruning first).
8029 : : */
8030 : : bool
8031 : 126178 : heap_tuple_needs_eventual_freeze(HeapTupleHeader tuple)
8032 : : {
8033 : : TransactionId xid;
8034 : :
8035 : : /*
8036 : : * If xmin is a normal transaction ID, this tuple is definitely not
8037 : : * frozen.
8038 : : */
8039 : 126178 : xid = HeapTupleHeaderGetXmin(tuple);
8040 [ + + ]: 126178 : if (TransactionIdIsNormal(xid))
8041 : 3468 : return true;
8042 : :
8043 : : /*
8044 : : * If xmax is a valid xact or multixact, this tuple is also not frozen.
8045 : : */
8046 [ - + ]: 122710 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
8047 : : {
8048 : : MultiXactId multi;
8049 : :
8050 : 0 : multi = HeapTupleHeaderGetRawXmax(tuple);
8051 [ # # ]: 0 : if (MultiXactIdIsValid(multi))
8052 : 0 : return true;
8053 : : }
8054 : : else
8055 : : {
8056 : 122710 : xid = HeapTupleHeaderGetRawXmax(tuple);
8057 [ + + ]: 122710 : if (TransactionIdIsNormal(xid))
8058 : 11 : return true;
8059 : : }
8060 : :
8061 [ - + ]: 122699 : if (tuple->t_infomask & HEAP_MOVED)
8062 : : {
8063 : 0 : xid = HeapTupleHeaderGetXvac(tuple);
8064 [ # # ]: 0 : if (TransactionIdIsNormal(xid))
8065 : 0 : return true;
8066 : : }
8067 : :
8068 : 122699 : return false;
8069 : : }
8070 : :
8071 : : /*
8072 : : * heap_tuple_should_freeze
8073 : : *
8074 : : * Return value indicates if heap_prepare_freeze_tuple sibling function would
8075 : : * (or should) force freezing of the heap page that contains caller's tuple.
8076 : : * Tuple header XIDs/MXIDs < FreezeLimit/MultiXactCutoff trigger freezing.
8077 : : * This includes (xmin, xmax, xvac) fields, as well as MultiXact member XIDs.
8078 : : *
8079 : : * The *NoFreezePageRelfrozenXid and *NoFreezePageRelminMxid input/output
8080 : : * arguments help VACUUM track the oldest extant XID/MXID remaining in rel.
8081 : : * Our working assumption is that caller won't decide to freeze this tuple.
8082 : : * It's up to caller to only ratchet back its own top-level trackers after the
8083 : : * point that it fully commits to not freezing the tuple/page in question.
8084 : : */
8085 : : bool
8086 : 8117251 : heap_tuple_should_freeze(HeapTupleHeader tuple,
8087 : : const struct VacuumCutoffs *cutoffs,
8088 : : TransactionId *NoFreezePageRelfrozenXid,
8089 : : MultiXactId *NoFreezePageRelminMxid)
8090 : : {
8091 : : TransactionId xid;
8092 : : MultiXactId multi;
8093 : 8117251 : bool freeze = false;
8094 : :
8095 : : /* First deal with xmin */
8096 : 8117251 : xid = HeapTupleHeaderGetXmin(tuple);
8097 [ + + ]: 8117251 : if (TransactionIdIsNormal(xid))
8098 : : {
8099 : : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
8100 [ + + ]: 2674610 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
8101 : 26031 : *NoFreezePageRelfrozenXid = xid;
8102 [ + + ]: 2674610 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
8103 : 23722 : freeze = true;
8104 : : }
8105 : :
8106 : : /* Now deal with xmax */
8107 : 8117251 : xid = InvalidTransactionId;
8108 : 8117251 : multi = InvalidMultiXactId;
8109 [ + + ]: 8117251 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
8110 : 2 : multi = HeapTupleHeaderGetRawXmax(tuple);
8111 : : else
8112 : 8117249 : xid = HeapTupleHeaderGetRawXmax(tuple);
8113 : :
8114 [ + + ]: 8117251 : if (TransactionIdIsNormal(xid))
8115 : : {
8116 : : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
8117 : : /* xmax is a non-permanent XID */
8118 [ + + ]: 5771006 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
8119 : 4 : *NoFreezePageRelfrozenXid = xid;
8120 [ + + ]: 5771006 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
8121 : 19 : freeze = true;
8122 : : }
8123 [ + + ]: 2346245 : else if (!MultiXactIdIsValid(multi))
8124 : : {
8125 : : /* xmax is a permanent XID or invalid MultiXactId/XID */
8126 : : }
8127 [ - + ]: 2 : else if (HEAP_LOCKED_UPGRADED(tuple->t_infomask))
8128 : : {
8129 : : /* xmax is a pg_upgrade'd MultiXact, which can't have updater XID */
8130 [ # # ]: 0 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
8131 : 0 : *NoFreezePageRelminMxid = multi;
8132 : : /* heap_prepare_freeze_tuple always freezes pg_upgrade'd xmax */
8133 : 0 : freeze = true;
8134 : : }
8135 : : else
8136 : : {
8137 : : /* xmax is a MultiXactId that may have an updater XID */
8138 : : MultiXactMember *members;
8139 : : int nmembers;
8140 : :
8141 : : Assert(MultiXactIdPrecedesOrEquals(cutoffs->relminmxid, multi));
8142 [ + - ]: 2 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
8143 : 2 : *NoFreezePageRelminMxid = multi;
8144 [ + - ]: 2 : if (MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff))
8145 : 2 : freeze = true;
8146 : :
8147 : : /* need to check whether any member of the mxact is old */
8148 : 2 : nmembers = GetMultiXactIdMembers(multi, &members, false,
8149 : 2 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask));
8150 : :
8151 [ + + ]: 5 : for (int i = 0; i < nmembers; i++)
8152 : : {
8153 : 3 : xid = members[i].xid;
8154 : : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
8155 [ - + ]: 3 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
8156 : 0 : *NoFreezePageRelfrozenXid = xid;
8157 [ - + ]: 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
8158 : 0 : freeze = true;
8159 : : }
8160 [ + + ]: 2 : if (nmembers > 0)
8161 : 1 : pfree(members);
8162 : : }
8163 : :
8164 [ - + ]: 8117251 : if (tuple->t_infomask & HEAP_MOVED)
8165 : : {
8166 : 0 : xid = HeapTupleHeaderGetXvac(tuple);
8167 [ # # ]: 0 : if (TransactionIdIsNormal(xid))
8168 : : {
8169 : : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
8170 [ # # ]: 0 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
8171 : 0 : *NoFreezePageRelfrozenXid = xid;
8172 : : /* heap_prepare_freeze_tuple forces xvac freezing */
8173 : 0 : freeze = true;
8174 : : }
8175 : : }
8176 : :
8177 : 8117251 : return freeze;
8178 : : }
8179 : :
8180 : : /*
8181 : : * Maintain snapshotConflictHorizon for caller by ratcheting forward its value
8182 : : * using any committed XIDs contained in 'tuple', an obsolescent heap tuple
8183 : : * that caller is in the process of physically removing, e.g. via HOT pruning
8184 : : * or index deletion.
8185 : : *
8186 : : * Caller must initialize its value to InvalidTransactionId, which is
8187 : : * generally interpreted as "definitely no need for a recovery conflict".
8188 : : * Final value must reflect all heap tuples that caller will physically remove
8189 : : * (or remove TID references to) via its ongoing pruning/deletion operation.
8190 : : * ResolveRecoveryConflictWithSnapshot() is passed the final value (taken from
8191 : : * caller's WAL record) by REDO routine when it replays caller's operation.
8192 : : */
8193 : : void
8194 : 4042147 : HeapTupleHeaderAdvanceConflictHorizon(HeapTupleHeader tuple,
8195 : : TransactionId *snapshotConflictHorizon)
8196 : : {
8197 : 4042147 : TransactionId xmin = HeapTupleHeaderGetXmin(tuple);
8198 : 4042147 : TransactionId xmax = HeapTupleHeaderGetUpdateXid(tuple);
8199 : 4042147 : TransactionId xvac = HeapTupleHeaderGetXvac(tuple);
8200 : :
8201 [ - + ]: 4042147 : if (tuple->t_infomask & HEAP_MOVED)
8202 : : {
8203 [ # # ]: 0 : if (TransactionIdPrecedes(*snapshotConflictHorizon, xvac))
8204 : 0 : *snapshotConflictHorizon = xvac;
8205 : : }
8206 : :
8207 : : /*
8208 : : * Ignore tuples inserted by an aborted transaction or if the tuple was
8209 : : * updated/deleted by the inserting transaction.
8210 : : *
8211 : : * Look for a committed hint bit, or if no xmin bit is set, check clog.
8212 : : */
8213 [ + + ]: 4042147 : if (HeapTupleHeaderXminCommitted(tuple) ||
8214 [ + + + - ]: 131970 : (!HeapTupleHeaderXminInvalid(tuple) && TransactionIdDidCommit(xmin)))
8215 : : {
8216 [ + + + + ]: 7638975 : if (xmax != xmin &&
8217 : 3719110 : TransactionIdFollows(xmax, *snapshotConflictHorizon))
8218 : 135076 : *snapshotConflictHorizon = xmax;
8219 : : }
8220 : 4042147 : }
8221 : :
8222 : : #ifdef USE_PREFETCH
8223 : : /*
8224 : : * Helper function for heap_index_delete_tuples. Issues prefetch requests for
8225 : : * prefetch_count buffers. The prefetch_state keeps track of all the buffers
8226 : : * we can prefetch, and which have already been prefetched; each call to this
8227 : : * function picks up where the previous call left off.
8228 : : *
8229 : : * Note: we expect the deltids array to be sorted in an order that groups TIDs
8230 : : * by heap block, with all TIDs for each block appearing together in exactly
8231 : : * one group.
8232 : : */
8233 : : static void
8234 : 28213 : index_delete_prefetch_buffer(Relation rel,
8235 : : IndexDeletePrefetchState *prefetch_state,
8236 : : int prefetch_count)
8237 : : {
8238 : 28213 : BlockNumber cur_hblkno = prefetch_state->cur_hblkno;
8239 : 28213 : int count = 0;
8240 : : int i;
8241 : 28213 : int ndeltids = prefetch_state->ndeltids;
8242 : 28213 : TM_IndexDelete *deltids = prefetch_state->deltids;
8243 : :
8244 : 28213 : for (i = prefetch_state->next_item;
8245 [ + + + + ]: 950158 : i < ndeltids && count < prefetch_count;
8246 : 921945 : i++)
8247 : : {
8248 : 921945 : ItemPointer htid = &deltids[i].tid;
8249 : :
8250 [ + + + + ]: 1835392 : if (cur_hblkno == InvalidBlockNumber ||
8251 : 913447 : ItemPointerGetBlockNumber(htid) != cur_hblkno)
8252 : : {
8253 : 24983 : cur_hblkno = ItemPointerGetBlockNumber(htid);
8254 : 24983 : PrefetchBuffer(rel, MAIN_FORKNUM, cur_hblkno);
8255 : 24983 : count++;
8256 : : }
8257 : : }
8258 : :
8259 : : /*
8260 : : * Save the prefetch position so that next time we can continue from that
8261 : : * position.
8262 : : */
8263 : 28213 : prefetch_state->next_item = i;
8264 : 28213 : prefetch_state->cur_hblkno = cur_hblkno;
8265 : 28213 : }
8266 : : #endif
8267 : :
8268 : : /*
8269 : : * Helper function for heap_index_delete_tuples. Checks for index corruption
8270 : : * involving an invalid TID in index AM caller's index page.
8271 : : *
8272 : : * This is an ideal place for these checks. The index AM must hold a buffer
8273 : : * lock on the index page containing the TIDs we examine here, so we don't
8274 : : * have to worry about concurrent VACUUMs at all. We can be sure that the
8275 : : * index is corrupt when htid points directly to an LP_UNUSED item or
8276 : : * heap-only tuple, which is not the case during standard index scans.
8277 : : */
8278 : : static inline void
8279 : 776917 : index_delete_check_htid(TM_IndexDeleteOp *delstate,
8280 : : Page page, OffsetNumber maxoff,
8281 : : const ItemPointerData *htid, TM_IndexStatus *istatus)
8282 : : {
8283 : 776917 : OffsetNumber indexpagehoffnum = ItemPointerGetOffsetNumber(htid);
8284 : : ItemId iid;
8285 : :
8286 : : Assert(OffsetNumberIsValid(istatus->idxoffnum));
8287 : :
8288 [ - + ]: 776917 : if (unlikely(indexpagehoffnum > maxoff))
8289 [ # # ]: 0 : ereport(ERROR,
8290 : : (errcode(ERRCODE_INDEX_CORRUPTED),
8291 : : 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\"",
8292 : : ItemPointerGetBlockNumber(htid),
8293 : : indexpagehoffnum,
8294 : : istatus->idxoffnum, delstate->iblknum,
8295 : : RelationGetRelationName(delstate->irel))));
8296 : :
8297 : 776917 : iid = PageGetItemId(page, indexpagehoffnum);
8298 [ - + ]: 776917 : if (unlikely(!ItemIdIsUsed(iid)))
8299 [ # # ]: 0 : ereport(ERROR,
8300 : : (errcode(ERRCODE_INDEX_CORRUPTED),
8301 : : errmsg_internal("heap tid from index tuple (%u,%u) points to unused heap page item at offset %u of block %u in index \"%s\"",
8302 : : ItemPointerGetBlockNumber(htid),
8303 : : indexpagehoffnum,
8304 : : istatus->idxoffnum, delstate->iblknum,
8305 : : RelationGetRelationName(delstate->irel))));
8306 : :
8307 [ + + ]: 776917 : if (ItemIdHasStorage(iid))
8308 : : {
8309 : : HeapTupleHeader htup;
8310 : :
8311 : : Assert(ItemIdIsNormal(iid));
8312 : 468670 : htup = (HeapTupleHeader) PageGetItem(page, iid);
8313 : :
8314 [ - + ]: 468670 : if (unlikely(HeapTupleHeaderIsHeapOnly(htup)))
8315 [ # # ]: 0 : ereport(ERROR,
8316 : : (errcode(ERRCODE_INDEX_CORRUPTED),
8317 : : errmsg_internal("heap tid from index tuple (%u,%u) points to heap-only tuple at offset %u of block %u in index \"%s\"",
8318 : : ItemPointerGetBlockNumber(htid),
8319 : : indexpagehoffnum,
8320 : : istatus->idxoffnum, delstate->iblknum,
8321 : : RelationGetRelationName(delstate->irel))));
8322 : : }
8323 : 776917 : }
8324 : :
8325 : : /*
8326 : : * heapam implementation of tableam's index_delete_tuples interface.
8327 : : *
8328 : : * This helper function is called by index AMs during index tuple deletion.
8329 : : * See tableam header comments for an explanation of the interface implemented
8330 : : * here and a general theory of operation. Note that each call here is either
8331 : : * a simple index deletion call, or a bottom-up index deletion call.
8332 : : *
8333 : : * It's possible for this to generate a fair amount of I/O, since we may be
8334 : : * deleting hundreds of tuples from a single index block. To amortize that
8335 : : * cost to some degree, this uses prefetching and combines repeat accesses to
8336 : : * the same heap block.
8337 : : */
8338 : : TransactionId
8339 : 8498 : heap_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
8340 : : {
8341 : : /* Initial assumption is that earlier pruning took care of conflict */
8342 : 8498 : TransactionId snapshotConflictHorizon = InvalidTransactionId;
8343 : 8498 : BlockNumber blkno = InvalidBlockNumber;
8344 : 8498 : Buffer buf = InvalidBuffer;
8345 : 8498 : Page page = NULL;
8346 : 8498 : OffsetNumber maxoff = InvalidOffsetNumber;
8347 : : TransactionId priorXmax;
8348 : : #ifdef USE_PREFETCH
8349 : : IndexDeletePrefetchState prefetch_state;
8350 : : int prefetch_distance;
8351 : : #endif
8352 : : SnapshotData SnapshotNonVacuumable;
8353 : 8498 : int finalndeltids = 0,
8354 : 8498 : nblocksaccessed = 0;
8355 : :
8356 : : /* State that's only used in bottom-up index deletion case */
8357 : 8498 : int nblocksfavorable = 0;
8358 : 8498 : int curtargetfreespace = delstate->bottomupfreespace,
8359 : 8498 : lastfreespace = 0,
8360 : 8498 : actualfreespace = 0;
8361 : 8498 : bool bottomup_final_block = false;
8362 : :
8363 : 8498 : InitNonVacuumableSnapshot(SnapshotNonVacuumable, GlobalVisTestFor(rel));
8364 : :
8365 : : /* Sort caller's deltids array by TID for further processing */
8366 : 8498 : index_delete_sort(delstate);
8367 : :
8368 : : /*
8369 : : * Bottom-up case: resort deltids array in an order attuned to where the
8370 : : * greatest number of promising TIDs are to be found, and determine how
8371 : : * many blocks from the start of sorted array should be considered
8372 : : * favorable. This will also shrink the deltids array in order to
8373 : : * eliminate completely unfavorable blocks up front.
8374 : : */
8375 [ + + ]: 8498 : if (delstate->bottomup)
8376 : 2732 : nblocksfavorable = bottomup_sort_and_shrink(delstate);
8377 : :
8378 : : #ifdef USE_PREFETCH
8379 : : /* Initialize prefetch state. */
8380 : 8498 : prefetch_state.cur_hblkno = InvalidBlockNumber;
8381 : 8498 : prefetch_state.next_item = 0;
8382 : 8498 : prefetch_state.ndeltids = delstate->ndeltids;
8383 : 8498 : prefetch_state.deltids = delstate->deltids;
8384 : :
8385 : : /*
8386 : : * Determine the prefetch distance that we will attempt to maintain.
8387 : : *
8388 : : * Since the caller holds a buffer lock somewhere in rel, we'd better make
8389 : : * sure that isn't a catalog relation before we call code that does
8390 : : * syscache lookups, to avoid risk of deadlock.
8391 : : */
8392 [ + + ]: 8498 : if (IsCatalogRelation(rel))
8393 : 6307 : prefetch_distance = maintenance_io_concurrency;
8394 : : else
8395 : : prefetch_distance =
8396 : 2191 : get_tablespace_maintenance_io_concurrency(rel->rd_rel->reltablespace);
8397 : :
8398 : : /* Cap initial prefetch distance for bottom-up deletion caller */
8399 [ + + ]: 8498 : if (delstate->bottomup)
8400 : : {
8401 : : Assert(nblocksfavorable >= 1);
8402 : : Assert(nblocksfavorable <= BOTTOMUP_MAX_NBLOCKS);
8403 : 2732 : prefetch_distance = Min(prefetch_distance, nblocksfavorable);
8404 : : }
8405 : :
8406 : : /* Start prefetching. */
8407 : 8498 : index_delete_prefetch_buffer(rel, &prefetch_state, prefetch_distance);
8408 : : #endif
8409 : :
8410 : : /* Iterate over deltids, determine which to delete, check their horizon */
8411 : : Assert(delstate->ndeltids > 0);
8412 [ + + ]: 785415 : for (int i = 0; i < delstate->ndeltids; i++)
8413 : : {
8414 : 779648 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8415 : 779648 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8416 : 779648 : ItemPointer htid = &ideltid->tid;
8417 : : OffsetNumber offnum;
8418 : :
8419 : : /*
8420 : : * Read buffer, and perform required extra steps each time a new block
8421 : : * is encountered. Avoid refetching if it's the same block as the one
8422 : : * from the last htid.
8423 : : */
8424 [ + + + + ]: 1550798 : if (blkno == InvalidBlockNumber ||
8425 : 771150 : ItemPointerGetBlockNumber(htid) != blkno)
8426 : : {
8427 : : /*
8428 : : * Consider giving up early for bottom-up index deletion caller
8429 : : * first. (Only prefetch next-next block afterwards, when it
8430 : : * becomes clear that we're at least going to access the next
8431 : : * block in line.)
8432 : : *
8433 : : * Sometimes the first block frees so much space for bottom-up
8434 : : * caller that the deletion process can end without accessing any
8435 : : * more blocks. It is usually necessary to access 2 or 3 blocks
8436 : : * per bottom-up deletion operation, though.
8437 : : */
8438 [ + + ]: 22446 : if (delstate->bottomup)
8439 : : {
8440 : : /*
8441 : : * We often allow caller to delete a few additional items
8442 : : * whose entries we reached after the point that space target
8443 : : * from caller was satisfied. The cost of accessing the page
8444 : : * was already paid at that point, so it made sense to finish
8445 : : * it off. When that happened, we finalize everything here
8446 : : * (by finishing off the whole bottom-up deletion operation
8447 : : * without needlessly paying the cost of accessing any more
8448 : : * blocks).
8449 : : */
8450 [ + + ]: 5828 : if (bottomup_final_block)
8451 : 177 : break;
8452 : :
8453 : : /*
8454 : : * Give up when we didn't enable our caller to free any
8455 : : * additional space as a result of processing the page that we
8456 : : * just finished up with. This rule is the main way in which
8457 : : * we keep the cost of bottom-up deletion under control.
8458 : : */
8459 [ + + + + ]: 5651 : if (nblocksaccessed >= 1 && actualfreespace == lastfreespace)
8460 : 2554 : break;
8461 : 3097 : lastfreespace = actualfreespace; /* for next time */
8462 : :
8463 : : /*
8464 : : * Deletion operation (which is bottom-up) will definitely
8465 : : * access the next block in line. Prepare for that now.
8466 : : *
8467 : : * Decay target free space so that we don't hang on for too
8468 : : * long with a marginal case. (Space target is only truly
8469 : : * helpful when it allows us to recognize that we don't need
8470 : : * to access more than 1 or 2 blocks to satisfy caller due to
8471 : : * agreeable workload characteristics.)
8472 : : *
8473 : : * We are a bit more patient when we encounter contiguous
8474 : : * blocks, though: these are treated as favorable blocks. The
8475 : : * decay process is only applied when the next block in line
8476 : : * is not a favorable/contiguous block. This is not an
8477 : : * exception to the general rule; we still insist on finding
8478 : : * at least one deletable item per block accessed. See
8479 : : * bottomup_nblocksfavorable() for full details of the theory
8480 : : * behind favorable blocks and heap block locality in general.
8481 : : *
8482 : : * Note: The first block in line is always treated as a
8483 : : * favorable block, so the earliest possible point that the
8484 : : * decay can be applied is just before we access the second
8485 : : * block in line. The Assert() verifies this for us.
8486 : : */
8487 : : Assert(nblocksaccessed > 0 || nblocksfavorable > 0);
8488 [ + + ]: 3097 : if (nblocksfavorable > 0)
8489 : 2893 : nblocksfavorable--;
8490 : : else
8491 : 204 : curtargetfreespace /= 2;
8492 : : }
8493 : :
8494 : : /* release old buffer */
8495 [ + + ]: 19715 : if (BufferIsValid(buf))
8496 : 11217 : UnlockReleaseBuffer(buf);
8497 : :
8498 : 19715 : blkno = ItemPointerGetBlockNumber(htid);
8499 : 19715 : buf = ReadBuffer(rel, blkno);
8500 : 19715 : nblocksaccessed++;
8501 : : Assert(!delstate->bottomup ||
8502 : : nblocksaccessed <= BOTTOMUP_MAX_NBLOCKS);
8503 : :
8504 : : #ifdef USE_PREFETCH
8505 : :
8506 : : /*
8507 : : * To maintain the prefetch distance, prefetch one more page for
8508 : : * each page we read.
8509 : : */
8510 : 19715 : index_delete_prefetch_buffer(rel, &prefetch_state, 1);
8511 : : #endif
8512 : :
8513 : 19715 : LockBuffer(buf, BUFFER_LOCK_SHARE);
8514 : :
8515 : 19715 : page = BufferGetPage(buf);
8516 : 19715 : maxoff = PageGetMaxOffsetNumber(page);
8517 : : }
8518 : :
8519 : : /*
8520 : : * In passing, detect index corruption involving an index page with a
8521 : : * TID that points to a location in the heap that couldn't possibly be
8522 : : * correct. We only do this with actual TIDs from caller's index page
8523 : : * (not items reached by traversing through a HOT chain).
8524 : : */
8525 : 776917 : index_delete_check_htid(delstate, page, maxoff, htid, istatus);
8526 : :
8527 [ + + ]: 776917 : if (istatus->knowndeletable)
8528 : : Assert(!delstate->bottomup && !istatus->promising);
8529 : : else
8530 : : {
8531 : 605393 : ItemPointerData tmp = *htid;
8532 : : HeapTupleData heapTuple;
8533 : :
8534 : : /* Are any tuples from this HOT chain non-vacuumable? */
8535 [ + + ]: 605393 : if (heap_hot_search_buffer(&tmp, rel, buf, &SnapshotNonVacuumable,
8536 : : &heapTuple, NULL, true))
8537 : 356080 : continue; /* can't delete entry */
8538 : :
8539 : : /* Caller will delete, since whole HOT chain is vacuumable */
8540 : 249313 : istatus->knowndeletable = true;
8541 : :
8542 : : /* Maintain index free space info for bottom-up deletion case */
8543 [ + + ]: 249313 : if (delstate->bottomup)
8544 : : {
8545 : : Assert(istatus->freespace > 0);
8546 : 9379 : actualfreespace += istatus->freespace;
8547 [ + + ]: 9379 : if (actualfreespace >= curtargetfreespace)
8548 : 2702 : bottomup_final_block = true;
8549 : : }
8550 : : }
8551 : :
8552 : : /*
8553 : : * Maintain snapshotConflictHorizon value for deletion operation as a
8554 : : * whole by advancing current value using heap tuple headers. This is
8555 : : * loosely based on the logic for pruning a HOT chain.
8556 : : */
8557 : 420837 : offnum = ItemPointerGetOffsetNumber(htid);
8558 : 420837 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
8559 : : for (;;)
8560 : 22574 : {
8561 : : ItemId lp;
8562 : : HeapTupleHeader htup;
8563 : :
8564 : : /* Sanity check (pure paranoia) */
8565 [ - + ]: 443411 : if (offnum < FirstOffsetNumber)
8566 : 0 : break;
8567 : :
8568 : : /*
8569 : : * An offset past the end of page's line pointer array is possible
8570 : : * when the array was truncated
8571 : : */
8572 [ - + ]: 443411 : if (offnum > maxoff)
8573 : 0 : break;
8574 : :
8575 : 443411 : lp = PageGetItemId(page, offnum);
8576 [ + + ]: 443411 : if (ItemIdIsRedirected(lp))
8577 : : {
8578 : 10202 : offnum = ItemIdGetRedirect(lp);
8579 : 10202 : continue;
8580 : : }
8581 : :
8582 : : /*
8583 : : * We'll often encounter LP_DEAD line pointers (especially with an
8584 : : * entry marked knowndeletable by our caller up front). No heap
8585 : : * tuple headers get examined for an htid that leads us to an
8586 : : * LP_DEAD item. This is okay because the earlier pruning
8587 : : * operation that made the line pointer LP_DEAD in the first place
8588 : : * must have considered the original tuple header as part of
8589 : : * generating its own snapshotConflictHorizon value.
8590 : : *
8591 : : * Relying on XLOG_HEAP2_PRUNE_VACUUM_SCAN records like this is
8592 : : * the same strategy that index vacuuming uses in all cases. Index
8593 : : * VACUUM WAL records don't even have a snapshotConflictHorizon
8594 : : * field of their own for this reason.
8595 : : */
8596 [ + + ]: 433209 : if (!ItemIdIsNormal(lp))
8597 : 279924 : break;
8598 : :
8599 : 153285 : htup = (HeapTupleHeader) PageGetItem(page, lp);
8600 : :
8601 : : /*
8602 : : * Check the tuple XMIN against prior XMAX, if any
8603 : : */
8604 [ + + - + ]: 165657 : if (TransactionIdIsValid(priorXmax) &&
8605 : 12372 : !TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
8606 : 0 : break;
8607 : :
8608 : 153285 : HeapTupleHeaderAdvanceConflictHorizon(htup,
8609 : : &snapshotConflictHorizon);
8610 : :
8611 : : /*
8612 : : * If the tuple is not HOT-updated, then we are at the end of this
8613 : : * HOT-chain. No need to visit later tuples from the same update
8614 : : * chain (they get their own index entries) -- just move on to
8615 : : * next htid from index AM caller.
8616 : : */
8617 [ + + ]: 153285 : if (!HeapTupleHeaderIsHotUpdated(htup))
8618 : 140913 : break;
8619 : :
8620 : : /* Advance to next HOT chain member */
8621 : : Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blkno);
8622 : 12372 : offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
8623 : 12372 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
8624 : : }
8625 : :
8626 : : /* Enable further/final shrinking of deltids for caller */
8627 : 420837 : finalndeltids = i + 1;
8628 : : }
8629 : :
8630 : 8498 : UnlockReleaseBuffer(buf);
8631 : :
8632 : : /*
8633 : : * Shrink deltids array to exclude non-deletable entries at the end. This
8634 : : * is not just a minor optimization. Final deltids array size might be
8635 : : * zero for a bottom-up caller. Index AM is explicitly allowed to rely on
8636 : : * ndeltids being zero in all cases with zero total deletable entries.
8637 : : */
8638 : : Assert(finalndeltids > 0 || delstate->bottomup);
8639 : 8498 : delstate->ndeltids = finalndeltids;
8640 : :
8641 : 8498 : return snapshotConflictHorizon;
8642 : : }
8643 : :
8644 : : /*
8645 : : * Specialized inlineable comparison function for index_delete_sort()
8646 : : */
8647 : : static inline int
8648 : 17574272 : index_delete_sort_cmp(TM_IndexDelete *deltid1, TM_IndexDelete *deltid2)
8649 : : {
8650 : 17574272 : ItemPointer tid1 = &deltid1->tid;
8651 : 17574272 : ItemPointer tid2 = &deltid2->tid;
8652 : :
8653 : : {
8654 : 17574272 : BlockNumber blk1 = ItemPointerGetBlockNumber(tid1);
8655 : 17574272 : BlockNumber blk2 = ItemPointerGetBlockNumber(tid2);
8656 : :
8657 [ + + ]: 17574272 : if (blk1 != blk2)
8658 [ + + ]: 7045057 : return (blk1 < blk2) ? -1 : 1;
8659 : : }
8660 : : {
8661 : 10529215 : OffsetNumber pos1 = ItemPointerGetOffsetNumber(tid1);
8662 : 10529215 : OffsetNumber pos2 = ItemPointerGetOffsetNumber(tid2);
8663 : :
8664 [ + - ]: 10529215 : if (pos1 != pos2)
8665 [ + + ]: 10529215 : return (pos1 < pos2) ? -1 : 1;
8666 : : }
8667 : :
8668 : : Assert(false);
8669 : :
8670 : 0 : return 0;
8671 : : }
8672 : :
8673 : : /*
8674 : : * Sort deltids array from delstate by TID. This prepares it for further
8675 : : * processing by heap_index_delete_tuples().
8676 : : *
8677 : : * This operation becomes a noticeable consumer of CPU cycles with some
8678 : : * workloads, so we go to the trouble of specialization/micro optimization.
8679 : : * We use shellsort for this because it's easy to specialize, compiles to
8680 : : * relatively few instructions, and is adaptive to presorted inputs/subsets
8681 : : * (which are typical here).
8682 : : */
8683 : : static void
8684 : 8498 : index_delete_sort(TM_IndexDeleteOp *delstate)
8685 : : {
8686 : 8498 : TM_IndexDelete *deltids = delstate->deltids;
8687 : 8498 : int ndeltids = delstate->ndeltids;
8688 : :
8689 : : /*
8690 : : * Shellsort gap sequence (taken from Sedgewick-Incerpi paper).
8691 : : *
8692 : : * This implementation is fast with array sizes up to ~4500. This covers
8693 : : * all supported BLCKSZ values.
8694 : : */
8695 : 8498 : const int gaps[9] = {1968, 861, 336, 112, 48, 21, 7, 3, 1};
8696 : :
8697 : : /* Think carefully before changing anything here -- keep swaps cheap */
8698 : : StaticAssertDecl(sizeof(TM_IndexDelete) <= 8,
8699 : : "element size exceeds 8 bytes");
8700 : :
8701 [ + + ]: 84980 : for (size_t g = 0; g < lengthof(gaps); g++)
8702 : : {
8703 [ + + ]: 10655274 : for (int hi = gaps[g], i = hi; i < ndeltids; i++)
8704 : : {
8705 : 10578792 : TM_IndexDelete d = deltids[i];
8706 : 10578792 : int j = i;
8707 : :
8708 [ + + + + ]: 18088658 : while (j >= hi && index_delete_sort_cmp(&deltids[j - hi], &d) >= 0)
8709 : : {
8710 : 7509866 : deltids[j] = deltids[j - hi];
8711 : 7509866 : j -= hi;
8712 : : }
8713 : 10578792 : deltids[j] = d;
8714 : : }
8715 : : }
8716 : 8498 : }
8717 : :
8718 : : /*
8719 : : * Returns how many blocks should be considered favorable/contiguous for a
8720 : : * bottom-up index deletion pass. This is a number of heap blocks that starts
8721 : : * from and includes the first block in line.
8722 : : *
8723 : : * There is always at least one favorable block during bottom-up index
8724 : : * deletion. In the worst case (i.e. with totally random heap blocks) the
8725 : : * first block in line (the only favorable block) can be thought of as a
8726 : : * degenerate array of contiguous blocks that consists of a single block.
8727 : : * heap_index_delete_tuples() will expect this.
8728 : : *
8729 : : * Caller passes blockgroups, a description of the final order that deltids
8730 : : * will be sorted in for heap_index_delete_tuples() bottom-up index deletion
8731 : : * processing. Note that deltids need not actually be sorted just yet (caller
8732 : : * only passes deltids to us so that we can interpret blockgroups).
8733 : : *
8734 : : * You might guess that the existence of contiguous blocks cannot matter much,
8735 : : * since in general the main factor that determines which blocks we visit is
8736 : : * the number of promising TIDs, which is a fixed hint from the index AM.
8737 : : * We're not really targeting the general case, though -- the actual goal is
8738 : : * to adapt our behavior to a wide variety of naturally occurring conditions.
8739 : : * The effects of most of the heuristics we apply are only noticeable in the
8740 : : * aggregate, over time and across many _related_ bottom-up index deletion
8741 : : * passes.
8742 : : *
8743 : : * Deeming certain blocks favorable allows heapam to recognize and adapt to
8744 : : * workloads where heap blocks visited during bottom-up index deletion can be
8745 : : * accessed contiguously, in the sense that each newly visited block is the
8746 : : * neighbor of the block that bottom-up deletion just finished processing (or
8747 : : * close enough to it). It will likely be cheaper to access more favorable
8748 : : * blocks sooner rather than later (e.g. in this pass, not across a series of
8749 : : * related bottom-up passes). Either way it is probably only a matter of time
8750 : : * (or a matter of further correlated version churn) before all blocks that
8751 : : * appear together as a single large batch of favorable blocks get accessed by
8752 : : * _some_ bottom-up pass. Large batches of favorable blocks tend to either
8753 : : * appear almost constantly or not even once (it all depends on per-index
8754 : : * workload characteristics).
8755 : : *
8756 : : * Note that the blockgroups sort order applies a power-of-two bucketing
8757 : : * scheme that creates opportunities for contiguous groups of blocks to get
8758 : : * batched together, at least with workloads that are naturally amenable to
8759 : : * being driven by heap block locality. This doesn't just enhance the spatial
8760 : : * locality of bottom-up heap block processing in the obvious way. It also
8761 : : * enables temporal locality of access, since sorting by heap block number
8762 : : * naturally tends to make the bottom-up processing order deterministic.
8763 : : *
8764 : : * Consider the following example to get a sense of how temporal locality
8765 : : * might matter: There is a heap relation with several indexes, each of which
8766 : : * is low to medium cardinality. It is subject to constant non-HOT updates.
8767 : : * The updates are skewed (in one part of the primary key, perhaps). None of
8768 : : * the indexes are logically modified by the UPDATE statements (if they were
8769 : : * then bottom-up index deletion would not be triggered in the first place).
8770 : : * Naturally, each new round of index tuples (for each heap tuple that gets a
8771 : : * heap_update() call) will have the same heap TID in each and every index.
8772 : : * Since these indexes are low cardinality and never get logically modified,
8773 : : * heapam processing during bottom-up deletion passes will access heap blocks
8774 : : * in approximately sequential order. Temporal locality of access occurs due
8775 : : * to bottom-up deletion passes behaving very similarly across each of the
8776 : : * indexes at any given moment. This keeps the number of buffer misses needed
8777 : : * to visit heap blocks to a minimum.
8778 : : */
8779 : : static int
8780 : 2732 : bottomup_nblocksfavorable(IndexDeleteCounts *blockgroups, int nblockgroups,
8781 : : TM_IndexDelete *deltids)
8782 : : {
8783 : 2732 : int64 lastblock = -1;
8784 : 2732 : int nblocksfavorable = 0;
8785 : :
8786 : : Assert(nblockgroups >= 1);
8787 : : Assert(nblockgroups <= BOTTOMUP_MAX_NBLOCKS);
8788 : :
8789 : : /*
8790 : : * We tolerate heap blocks that will be accessed only slightly out of
8791 : : * physical order. Small blips occur when a pair of almost-contiguous
8792 : : * blocks happen to fall into different buckets (perhaps due only to a
8793 : : * small difference in npromisingtids that the bucketing scheme didn't
8794 : : * quite manage to ignore). We effectively ignore these blips by applying
8795 : : * a small tolerance. The precise tolerance we use is a little arbitrary,
8796 : : * but it works well enough in practice.
8797 : : */
8798 [ + + ]: 8463 : for (int b = 0; b < nblockgroups; b++)
8799 : : {
8800 : 8092 : IndexDeleteCounts *group = blockgroups + b;
8801 : 8092 : TM_IndexDelete *firstdtid = deltids + group->ifirsttid;
8802 : 8092 : BlockNumber block = ItemPointerGetBlockNumber(&firstdtid->tid);
8803 : :
8804 [ + + ]: 8092 : if (lastblock != -1 &&
8805 [ + + ]: 5360 : ((int64) block < lastblock - BOTTOMUP_TOLERANCE_NBLOCKS ||
8806 [ + + ]: 4746 : (int64) block > lastblock + BOTTOMUP_TOLERANCE_NBLOCKS))
8807 : : break;
8808 : :
8809 : 5731 : nblocksfavorable++;
8810 : 5731 : lastblock = block;
8811 : : }
8812 : :
8813 : : /* Always indicate that there is at least 1 favorable block */
8814 : : Assert(nblocksfavorable >= 1);
8815 : :
8816 : 2732 : return nblocksfavorable;
8817 : : }
8818 : :
8819 : : /*
8820 : : * qsort comparison function for bottomup_sort_and_shrink()
8821 : : */
8822 : : static int
8823 : 264893 : bottomup_sort_and_shrink_cmp(const void *arg1, const void *arg2)
8824 : : {
8825 : 264893 : const IndexDeleteCounts *group1 = (const IndexDeleteCounts *) arg1;
8826 : 264893 : const IndexDeleteCounts *group2 = (const IndexDeleteCounts *) arg2;
8827 : :
8828 : : /*
8829 : : * Most significant field is npromisingtids (which we invert the order of
8830 : : * so as to sort in desc order).
8831 : : *
8832 : : * Caller should have already normalized npromisingtids fields into
8833 : : * power-of-two values (buckets).
8834 : : */
8835 [ + + ]: 264893 : if (group1->npromisingtids > group2->npromisingtids)
8836 : 12428 : return -1;
8837 [ + + ]: 252465 : if (group1->npromisingtids < group2->npromisingtids)
8838 : 14578 : return 1;
8839 : :
8840 : : /*
8841 : : * Tiebreak: desc ntids sort order.
8842 : : *
8843 : : * We cannot expect power-of-two values for ntids fields. We should
8844 : : * behave as if they were already rounded up for us instead.
8845 : : */
8846 [ + + ]: 237887 : if (group1->ntids != group2->ntids)
8847 : : {
8848 : 170102 : uint32 ntids1 = pg_nextpower2_32((uint32) group1->ntids);
8849 : 170102 : uint32 ntids2 = pg_nextpower2_32((uint32) group2->ntids);
8850 : :
8851 [ + + ]: 170102 : if (ntids1 > ntids2)
8852 : 26418 : return -1;
8853 [ + + ]: 143684 : if (ntids1 < ntids2)
8854 : 30464 : return 1;
8855 : : }
8856 : :
8857 : : /*
8858 : : * Tiebreak: asc offset-into-deltids-for-block (offset to first TID for
8859 : : * block in deltids array) order.
8860 : : *
8861 : : * This is equivalent to sorting in ascending heap block number order
8862 : : * (among otherwise equal subsets of the array). This approach allows us
8863 : : * to avoid accessing the out-of-line TID. (We rely on the assumption
8864 : : * that the deltids array was sorted in ascending heap TID order when
8865 : : * these offsets to the first TID from each heap block group were formed.)
8866 : : */
8867 [ + + ]: 181005 : if (group1->ifirsttid > group2->ifirsttid)
8868 : 90434 : return 1;
8869 [ + - ]: 90571 : if (group1->ifirsttid < group2->ifirsttid)
8870 : 90571 : return -1;
8871 : :
8872 : 0 : pg_unreachable();
8873 : :
8874 : : return 0;
8875 : : }
8876 : :
8877 : : /*
8878 : : * heap_index_delete_tuples() helper function for bottom-up deletion callers.
8879 : : *
8880 : : * Sorts deltids array in the order needed for useful processing by bottom-up
8881 : : * deletion. The array should already be sorted in TID order when we're
8882 : : * called. The sort process groups heap TIDs from deltids into heap block
8883 : : * groupings. Earlier/more-promising groups/blocks are usually those that are
8884 : : * known to have the most "promising" TIDs.
8885 : : *
8886 : : * Sets new size of deltids array (ndeltids) in state. deltids will only have
8887 : : * TIDs from the BOTTOMUP_MAX_NBLOCKS most promising heap blocks when we
8888 : : * return. This often means that deltids will be shrunk to a small fraction
8889 : : * of its original size (we eliminate many heap blocks from consideration for
8890 : : * caller up front).
8891 : : *
8892 : : * Returns the number of "favorable" blocks. See bottomup_nblocksfavorable()
8893 : : * for a definition and full details.
8894 : : */
8895 : : static int
8896 : 2732 : bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate)
8897 : : {
8898 : : IndexDeleteCounts *blockgroups;
8899 : : TM_IndexDelete *reordereddeltids;
8900 : 2732 : BlockNumber curblock = InvalidBlockNumber;
8901 : 2732 : int nblockgroups = 0;
8902 : 2732 : int ncopied = 0;
8903 : 2732 : int nblocksfavorable = 0;
8904 : :
8905 : : Assert(delstate->bottomup);
8906 : : Assert(delstate->ndeltids > 0);
8907 : :
8908 : : /* Calculate per-heap-block count of TIDs */
8909 : 2732 : blockgroups = palloc_array(IndexDeleteCounts, delstate->ndeltids);
8910 [ + + ]: 1319086 : for (int i = 0; i < delstate->ndeltids; i++)
8911 : : {
8912 : 1316354 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8913 : 1316354 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8914 : 1316354 : ItemPointer htid = &ideltid->tid;
8915 : 1316354 : bool promising = istatus->promising;
8916 : :
8917 [ + + ]: 1316354 : if (curblock != ItemPointerGetBlockNumber(htid))
8918 : : {
8919 : : /* New block group */
8920 : 51323 : nblockgroups++;
8921 : :
8922 : : Assert(curblock < ItemPointerGetBlockNumber(htid) ||
8923 : : !BlockNumberIsValid(curblock));
8924 : :
8925 : 51323 : curblock = ItemPointerGetBlockNumber(htid);
8926 : 51323 : blockgroups[nblockgroups - 1].ifirsttid = i;
8927 : 51323 : blockgroups[nblockgroups - 1].ntids = 1;
8928 : 51323 : blockgroups[nblockgroups - 1].npromisingtids = 0;
8929 : : }
8930 : : else
8931 : : {
8932 : 1265031 : blockgroups[nblockgroups - 1].ntids++;
8933 : : }
8934 : :
8935 [ + + ]: 1316354 : if (promising)
8936 : 211034 : blockgroups[nblockgroups - 1].npromisingtids++;
8937 : : }
8938 : :
8939 : : /*
8940 : : * We're about ready to sort block groups to determine the optimal order
8941 : : * for visiting heap blocks. But before we do, round the number of
8942 : : * promising tuples for each block group up to the next power-of-two,
8943 : : * unless it is very low (less than 4), in which case we round up to 4.
8944 : : * npromisingtids is far too noisy to trust when choosing between a pair
8945 : : * of block groups that both have very low values.
8946 : : *
8947 : : * This scheme divides heap blocks/block groups into buckets. Each bucket
8948 : : * contains blocks that have _approximately_ the same number of promising
8949 : : * TIDs as each other. The goal is to ignore relatively small differences
8950 : : * in the total number of promising entries, so that the whole process can
8951 : : * give a little weight to heapam factors (like heap block locality)
8952 : : * instead. This isn't a trade-off, really -- we have nothing to lose. It
8953 : : * would be foolish to interpret small differences in npromisingtids
8954 : : * values as anything more than noise.
8955 : : *
8956 : : * We tiebreak on nhtids when sorting block group subsets that have the
8957 : : * same npromisingtids, but this has the same issues as npromisingtids,
8958 : : * and so nhtids is subject to the same power-of-two bucketing scheme. The
8959 : : * only reason that we don't fix nhtids in the same way here too is that
8960 : : * we'll need accurate nhtids values after the sort. We handle nhtids
8961 : : * bucketization dynamically instead (in the sort comparator).
8962 : : *
8963 : : * See bottomup_nblocksfavorable() for a full explanation of when and how
8964 : : * heap locality/favorable blocks can significantly influence when and how
8965 : : * heap blocks are accessed.
8966 : : */
8967 [ + + ]: 54055 : for (int b = 0; b < nblockgroups; b++)
8968 : : {
8969 : 51323 : IndexDeleteCounts *group = blockgroups + b;
8970 : :
8971 : : /* Better off falling back on nhtids with low npromisingtids */
8972 [ + + ]: 51323 : if (group->npromisingtids <= 4)
8973 : 43615 : group->npromisingtids = 4;
8974 : : else
8975 : 7708 : group->npromisingtids =
8976 : 7708 : pg_nextpower2_32((uint32) group->npromisingtids);
8977 : : }
8978 : :
8979 : : /* Sort groups and rearrange caller's deltids array */
8980 : 2732 : qsort(blockgroups, nblockgroups, sizeof(IndexDeleteCounts),
8981 : : bottomup_sort_and_shrink_cmp);
8982 : 2732 : reordereddeltids = palloc(delstate->ndeltids * sizeof(TM_IndexDelete));
8983 : :
8984 : 2732 : nblockgroups = Min(BOTTOMUP_MAX_NBLOCKS, nblockgroups);
8985 : : /* Determine number of favorable blocks at the start of final deltids */
8986 : 2732 : nblocksfavorable = bottomup_nblocksfavorable(blockgroups, nblockgroups,
8987 : : delstate->deltids);
8988 : :
8989 [ + + ]: 17785 : for (int b = 0; b < nblockgroups; b++)
8990 : : {
8991 : 15053 : IndexDeleteCounts *group = blockgroups + b;
8992 : 15053 : TM_IndexDelete *firstdtid = delstate->deltids + group->ifirsttid;
8993 : :
8994 : 15053 : memcpy(reordereddeltids + ncopied, firstdtid,
8995 : 15053 : sizeof(TM_IndexDelete) * group->ntids);
8996 : 15053 : ncopied += group->ntids;
8997 : : }
8998 : :
8999 : : /* Copy final grouped and sorted TIDs back into start of caller's array */
9000 : 2732 : memcpy(delstate->deltids, reordereddeltids,
9001 : : sizeof(TM_IndexDelete) * ncopied);
9002 : 2732 : delstate->ndeltids = ncopied;
9003 : :
9004 : 2732 : pfree(reordereddeltids);
9005 : 2732 : pfree(blockgroups);
9006 : :
9007 : 2732 : return nblocksfavorable;
9008 : : }
9009 : :
9010 : : /*
9011 : : * Perform XLogInsert for a heap-update operation. Caller must already
9012 : : * have modified the buffer(s) and marked them dirty.
9013 : : */
9014 : : static XLogRecPtr
9015 : 2374838 : log_heap_update(Relation reln, Buffer oldbuf, Buffer vmbuffer_old,
9016 : : Buffer newbuf, Buffer vmbuffer_new,
9017 : : HeapTuple oldtup, HeapTuple newtup,
9018 : : HeapTuple old_key_tuple,
9019 : : bool all_visible_cleared, bool new_all_visible_cleared,
9020 : : bool walLogical)
9021 : : {
9022 : : xl_heap_update xlrec;
9023 : : xl_heap_header xlhdr;
9024 : : xl_heap_header xlhdr_idx;
9025 : : uint8 info;
9026 : : uint16 prefix_suffix[2];
9027 : 2374838 : uint16 prefixlen = 0,
9028 : 2374838 : suffixlen = 0;
9029 : : XLogRecPtr recptr;
9030 : 2374838 : Page page = BufferGetPage(newbuf);
9031 [ + + + + : 2374838 : bool need_tuple_data = walLogical && RelationIsLogicallyLogged(reln);
+ + + - -
+ - - - -
+ - + + ]
9032 : : bool init;
9033 : : int bufflags;
9034 : :
9035 : : /* Caller should not call me on a non-WAL-logged relation */
9036 : : Assert(RelationNeedsWAL(reln));
9037 : :
9038 : 2374838 : XLogBeginInsert();
9039 : :
9040 [ + + ]: 2374838 : if (HeapTupleIsHeapOnly(newtup))
9041 : 170497 : info = XLOG_HEAP_HOT_UPDATE;
9042 : : else
9043 : 2204341 : info = XLOG_HEAP_UPDATE;
9044 : :
9045 : : /*
9046 : : * If the old and new tuple are on the same page, we only need to log the
9047 : : * parts of the new tuple that were changed. That saves on the amount of
9048 : : * WAL we need to write. Currently, we just count any unchanged bytes in
9049 : : * the beginning and end of the tuple. That's quick to check, and
9050 : : * perfectly covers the common case that only one field is updated.
9051 : : *
9052 : : * We could do this even if the old and new tuple are on different pages,
9053 : : * but only if we don't make a full-page image of the old page, which is
9054 : : * difficult to know in advance. Also, if the old tuple is corrupt for
9055 : : * some reason, it would allow the corruption to propagate the new page,
9056 : : * so it seems best to avoid. Under the general assumption that most
9057 : : * updates tend to create the new tuple version on the same page, there
9058 : : * isn't much to be gained by doing this across pages anyway.
9059 : : *
9060 : : * Skip this if we're taking a full-page image of the new page, as we
9061 : : * don't include the new tuple in the WAL record in that case. Also
9062 : : * disable if effective_wal_level='logical', as logical decoding needs to
9063 : : * be able to read the new tuple in whole from the WAL record alone.
9064 : : */
9065 [ + + + + ]: 2374838 : if (oldbuf == newbuf && !need_tuple_data &&
9066 [ + + ]: 175548 : !XLogCheckBufferNeedsBackup(newbuf))
9067 : : {
9068 : 175087 : char *oldp = (char *) oldtup->t_data + oldtup->t_data->t_hoff;
9069 : 175087 : char *newp = (char *) newtup->t_data + newtup->t_data->t_hoff;
9070 : 175087 : int oldlen = oldtup->t_len - oldtup->t_data->t_hoff;
9071 : 175087 : int newlen = newtup->t_len - newtup->t_data->t_hoff;
9072 : :
9073 : : /* Check for common prefix between old and new tuple */
9074 [ + + ]: 14853839 : for (prefixlen = 0; prefixlen < Min(oldlen, newlen); prefixlen++)
9075 : : {
9076 [ + + ]: 14825162 : if (newp[prefixlen] != oldp[prefixlen])
9077 : 146410 : break;
9078 : : }
9079 : :
9080 : : /*
9081 : : * Storing the length of the prefix takes 2 bytes, so we need to save
9082 : : * at least 3 bytes or there's no point.
9083 : : */
9084 [ + + ]: 175087 : if (prefixlen < 3)
9085 : 22844 : prefixlen = 0;
9086 : :
9087 : : /* Same for suffix */
9088 [ + + ]: 6434923 : for (suffixlen = 0; suffixlen < Min(oldlen, newlen) - prefixlen; suffixlen++)
9089 : : {
9090 [ + + ]: 6405896 : if (newp[newlen - suffixlen - 1] != oldp[oldlen - suffixlen - 1])
9091 : 146060 : break;
9092 : : }
9093 [ + + ]: 175087 : if (suffixlen < 3)
9094 : 43281 : suffixlen = 0;
9095 : : }
9096 : :
9097 : : /* Prepare main WAL data chain */
9098 : 2374838 : xlrec.flags = 0;
9099 [ + + ]: 2374838 : if (all_visible_cleared)
9100 : 2170 : xlrec.flags |= XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED;
9101 [ + + ]: 2374838 : if (new_all_visible_cleared)
9102 : 825 : xlrec.flags |= XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED;
9103 [ + + ]: 2374838 : if (prefixlen > 0)
9104 : 152243 : xlrec.flags |= XLH_UPDATE_PREFIX_FROM_OLD;
9105 [ + + ]: 2374838 : if (suffixlen > 0)
9106 : 131806 : xlrec.flags |= XLH_UPDATE_SUFFIX_FROM_OLD;
9107 [ + + ]: 2374838 : if (need_tuple_data)
9108 : : {
9109 : 47053 : xlrec.flags |= XLH_UPDATE_CONTAINS_NEW_TUPLE;
9110 [ + + ]: 47053 : if (old_key_tuple)
9111 : : {
9112 [ + + ]: 161 : if (reln->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
9113 : 68 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_TUPLE;
9114 : : else
9115 : 93 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_KEY;
9116 : : }
9117 : : }
9118 : :
9119 : : /* If new tuple is the single and first tuple on page... */
9120 [ + + + + ]: 2388057 : if (ItemPointerGetOffsetNumber(&(newtup->t_self)) == FirstOffsetNumber &&
9121 : 13219 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
9122 : : {
9123 : 12972 : info |= XLOG_HEAP_INIT_PAGE;
9124 : 12972 : init = true;
9125 : : }
9126 : : else
9127 : 2361866 : init = false;
9128 : :
9129 : : /* Prepare WAL data for the old page */
9130 : 2374838 : xlrec.old_offnum = ItemPointerGetOffsetNumber(&oldtup->t_self);
9131 : 2374838 : xlrec.old_xmax = HeapTupleHeaderGetRawXmax(oldtup->t_data);
9132 : 4749676 : xlrec.old_infobits_set = compute_infobits(oldtup->t_data->t_infomask,
9133 : 2374838 : oldtup->t_data->t_infomask2);
9134 : :
9135 : : /* Prepare WAL data for the new page */
9136 : 2374838 : xlrec.new_offnum = ItemPointerGetOffsetNumber(&newtup->t_self);
9137 : 2374838 : xlrec.new_xmax = HeapTupleHeaderGetRawXmax(newtup->t_data);
9138 : :
9139 : 2374838 : bufflags = REGBUF_STANDARD;
9140 [ + + ]: 2374838 : if (init)
9141 : 12972 : bufflags |= REGBUF_WILL_INIT;
9142 [ + + ]: 2374838 : if (need_tuple_data)
9143 : 47053 : bufflags |= REGBUF_KEEP_DATA;
9144 : :
9145 : 2374838 : XLogRegisterBuffer(HEAP_UPDATE_BLKREF_HEAP_NEW, newbuf, bufflags);
9146 [ + + ]: 2374838 : if (oldbuf != newbuf)
9147 : 2187317 : XLogRegisterBuffer(HEAP_UPDATE_BLKREF_HEAP_OLD, oldbuf, REGBUF_STANDARD);
9148 : :
9149 : 2374838 : XLogRegisterData(&xlrec, SizeOfHeapUpdate);
9150 : :
9151 : : /*
9152 : : * Prepare WAL data for the new tuple.
9153 : : */
9154 [ + + + + ]: 2374838 : if (prefixlen > 0 || suffixlen > 0)
9155 : : {
9156 [ + + + + ]: 174511 : if (prefixlen > 0 && suffixlen > 0)
9157 : : {
9158 : 109538 : prefix_suffix[0] = prefixlen;
9159 : 109538 : prefix_suffix[1] = suffixlen;
9160 : 109538 : XLogRegisterBufData(HEAP_UPDATE_BLKREF_HEAP_NEW, &prefix_suffix,
9161 : : sizeof(uint16) * 2);
9162 : : }
9163 [ + + ]: 64973 : else if (prefixlen > 0)
9164 : : {
9165 : 42705 : XLogRegisterBufData(HEAP_UPDATE_BLKREF_HEAP_NEW, &prefixlen,
9166 : : sizeof(uint16));
9167 : : }
9168 : : else
9169 : : {
9170 : 22268 : XLogRegisterBufData(HEAP_UPDATE_BLKREF_HEAP_NEW, &suffixlen,
9171 : : sizeof(uint16));
9172 : : }
9173 : : }
9174 : :
9175 : 2374838 : xlhdr.t_infomask2 = newtup->t_data->t_infomask2;
9176 : 2374838 : xlhdr.t_infomask = newtup->t_data->t_infomask;
9177 : 2374838 : xlhdr.t_hoff = newtup->t_data->t_hoff;
9178 : : Assert(SizeofHeapTupleHeader + prefixlen + suffixlen <= newtup->t_len);
9179 : :
9180 : : /*
9181 : : * PG73FORMAT: write bitmap [+ padding] [+ oid] + data
9182 : : *
9183 : : * The 'data' doesn't include the common prefix or suffix.
9184 : : */
9185 : 2374838 : XLogRegisterBufData(HEAP_UPDATE_BLKREF_HEAP_NEW, &xlhdr, SizeOfHeapHeader);
9186 [ + + ]: 2374838 : if (prefixlen == 0)
9187 : : {
9188 : 2222595 : XLogRegisterBufData(HEAP_UPDATE_BLKREF_HEAP_NEW,
9189 : 2222595 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9190 : 2222595 : newtup->t_len - SizeofHeapTupleHeader - suffixlen);
9191 : : }
9192 : : else
9193 : : {
9194 : : /*
9195 : : * Have to write the null bitmap and data after the common prefix as
9196 : : * two separate rdata entries.
9197 : : */
9198 : : /* bitmap [+ padding] [+ oid] */
9199 [ + - ]: 152243 : if (newtup->t_data->t_hoff - SizeofHeapTupleHeader > 0)
9200 : : {
9201 : 152243 : XLogRegisterBufData(HEAP_UPDATE_BLKREF_HEAP_NEW,
9202 : 152243 : (char *) newtup->t_data + SizeofHeapTupleHeader,
9203 : 152243 : newtup->t_data->t_hoff - SizeofHeapTupleHeader);
9204 : : }
9205 : :
9206 : : /* data after common prefix */
9207 : 152243 : XLogRegisterBufData(HEAP_UPDATE_BLKREF_HEAP_NEW,
9208 : 152243 : (char *) newtup->t_data + newtup->t_data->t_hoff + prefixlen,
9209 : 152243 : newtup->t_len - newtup->t_data->t_hoff - prefixlen - suffixlen);
9210 : : }
9211 : :
9212 : : /* We need to log a tuple identity */
9213 [ + + + + ]: 2374838 : if (need_tuple_data && old_key_tuple)
9214 : : {
9215 : : /* don't really need this, but its more comfy to decode */
9216 : 161 : xlhdr_idx.t_infomask2 = old_key_tuple->t_data->t_infomask2;
9217 : 161 : xlhdr_idx.t_infomask = old_key_tuple->t_data->t_infomask;
9218 : 161 : xlhdr_idx.t_hoff = old_key_tuple->t_data->t_hoff;
9219 : :
9220 : 161 : XLogRegisterData(&xlhdr_idx, SizeOfHeapHeader);
9221 : :
9222 : : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
9223 : 161 : XLogRegisterData((char *) old_key_tuple->t_data + SizeofHeapTupleHeader,
9224 : 161 : old_key_tuple->t_len - SizeofHeapTupleHeader);
9225 : : }
9226 : :
9227 : : /*
9228 : : * Register VM buffers. If the old and new heap pages' VM bits are on the
9229 : : * same VM page and both their VM bits were cleared, the caller passes
9230 : : * only vmbuffer_new (mirroring the heap page convention where block 0 =
9231 : : * new is always registered).
9232 : : */
9233 : : Assert((BufferIsInvalid(vmbuffer_old) && BufferIsInvalid(vmbuffer_new)) ||
9234 : : (vmbuffer_old != vmbuffer_new));
9235 : :
9236 [ + + ]: 2374838 : if (BufferIsValid(vmbuffer_new))
9237 : 825 : XLogRegisterBuffer(HEAP_UPDATE_BLKREF_VM_NEW, vmbuffer_new, 0);
9238 : :
9239 [ + + ]: 2374838 : if (BufferIsValid(vmbuffer_old))
9240 : 2026 : XLogRegisterBuffer(HEAP_UPDATE_BLKREF_VM_OLD, vmbuffer_old, 0);
9241 : :
9242 : : /* filtering by origin on a row level is much more efficient */
9243 : 2374838 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
9244 : :
9245 : 2374838 : recptr = XLogInsert(RM_HEAP_ID, info);
9246 : :
9247 : 2374838 : return recptr;
9248 : : }
9249 : :
9250 : : /*
9251 : : * Perform XLogInsert of an XLOG_HEAP2_NEW_CID record
9252 : : *
9253 : : * This is only used when effective_wal_level is logical, and only for
9254 : : * catalog tuples.
9255 : : */
9256 : : static XLogRecPtr
9257 : 26610 : log_heap_new_cid(Relation relation, HeapTuple tup)
9258 : : {
9259 : : xl_heap_new_cid xlrec;
9260 : :
9261 : : XLogRecPtr recptr;
9262 : 26610 : HeapTupleHeader hdr = tup->t_data;
9263 : :
9264 : : Assert(ItemPointerIsValid(&tup->t_self));
9265 : : Assert(tup->t_tableOid != InvalidOid);
9266 : :
9267 : 26610 : xlrec.top_xid = GetTopTransactionId();
9268 : 26610 : xlrec.target_locator = relation->rd_locator;
9269 : 26610 : xlrec.target_tid = tup->t_self;
9270 : :
9271 : : /*
9272 : : * If the tuple got inserted & deleted in the same TX we definitely have a
9273 : : * combo CID, set cmin and cmax.
9274 : : */
9275 [ + + ]: 26610 : if (hdr->t_infomask & HEAP_COMBOCID)
9276 : : {
9277 : : Assert(!(hdr->t_infomask & HEAP_XMAX_INVALID));
9278 : : Assert(!HeapTupleHeaderXminInvalid(hdr));
9279 : 2237 : xlrec.cmin = HeapTupleHeaderGetCmin(hdr);
9280 : 2237 : xlrec.cmax = HeapTupleHeaderGetCmax(hdr);
9281 : 2237 : xlrec.combocid = HeapTupleHeaderGetRawCommandId(hdr);
9282 : : }
9283 : : /* No combo CID, so only cmin or cmax can be set by this TX */
9284 : : else
9285 : : {
9286 : : /*
9287 : : * Tuple inserted.
9288 : : *
9289 : : * We need to check for LOCK ONLY because multixacts might be
9290 : : * transferred to the new tuple in case of FOR KEY SHARE updates in
9291 : : * which case there will be an xmax, although the tuple just got
9292 : : * inserted.
9293 : : */
9294 [ + + + + ]: 31836 : if (hdr->t_infomask & HEAP_XMAX_INVALID ||
9295 : 7463 : HEAP_XMAX_IS_LOCKED_ONLY(hdr->t_infomask))
9296 : : {
9297 : 16911 : xlrec.cmin = HeapTupleHeaderGetRawCommandId(hdr);
9298 : 16911 : xlrec.cmax = InvalidCommandId;
9299 : : }
9300 : : /* Tuple from a different tx updated or deleted. */
9301 : : else
9302 : : {
9303 : 7462 : xlrec.cmin = InvalidCommandId;
9304 : 7462 : xlrec.cmax = HeapTupleHeaderGetRawCommandId(hdr);
9305 : : }
9306 : 24373 : xlrec.combocid = InvalidCommandId;
9307 : : }
9308 : :
9309 : : /*
9310 : : * Note that we don't need to register the buffer here, because this
9311 : : * operation does not modify the page. The insert/update/delete that
9312 : : * called us certainly did, but that's WAL-logged separately.
9313 : : */
9314 : 26610 : XLogBeginInsert();
9315 : 26610 : XLogRegisterData(&xlrec, SizeOfHeapNewCid);
9316 : :
9317 : : /* will be looked at irrespective of origin */
9318 : :
9319 : 26610 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_NEW_CID);
9320 : :
9321 : 26610 : return recptr;
9322 : : }
9323 : :
9324 : : /*
9325 : : * Build a heap tuple representing the configured REPLICA IDENTITY to represent
9326 : : * the old tuple in an UPDATE or DELETE.
9327 : : *
9328 : : * Returns NULL if there's no need to log an identity or if there's no suitable
9329 : : * key defined.
9330 : : *
9331 : : * Pass key_required true if any replica identity columns changed value, or if
9332 : : * any of them have any external data. Delete must always pass true.
9333 : : *
9334 : : * *copy is set to true if the returned tuple is a modified copy rather than
9335 : : * the same tuple that was passed in.
9336 : : */
9337 : : static HeapTuple
9338 : 4310221 : ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
9339 : : bool *copy)
9340 : : {
9341 : 4310221 : TupleDesc desc = RelationGetDescr(relation);
9342 : 4310221 : char replident = relation->rd_rel->relreplident;
9343 : : Bitmapset *idattrs;
9344 : : HeapTuple key_tuple;
9345 : : bool nulls[MaxHeapAttributeNumber];
9346 : : Datum values[MaxHeapAttributeNumber];
9347 : :
9348 : 4310221 : *copy = false;
9349 : :
9350 [ + + + + : 4310221 : if (!RelationIsLogicallyLogged(relation))
+ + - + -
- - - + -
+ + ]
9351 : 4209824 : return NULL;
9352 : :
9353 [ + + ]: 100397 : if (replident == REPLICA_IDENTITY_NOTHING)
9354 : 272 : return NULL;
9355 : :
9356 [ + + ]: 100125 : if (replident == REPLICA_IDENTITY_FULL)
9357 : : {
9358 : : /*
9359 : : * When logging the entire old tuple, it very well could contain
9360 : : * toasted columns. If so, force them to be inlined.
9361 : : */
9362 [ + + ]: 203 : if (HeapTupleHasExternal(tp))
9363 : : {
9364 : 4 : *copy = true;
9365 : 4 : tp = toast_flatten_tuple(tp, desc);
9366 : : }
9367 : 203 : return tp;
9368 : : }
9369 : :
9370 : : /* if the key isn't required and we're only logging the key, we're done */
9371 [ + + ]: 99922 : if (!key_required)
9372 : 46914 : return NULL;
9373 : :
9374 : : /* find out the replica identity columns */
9375 : 53008 : idattrs = RelationGetIndexAttrBitmap(relation,
9376 : : INDEX_ATTR_BITMAP_IDENTITY_KEY);
9377 : :
9378 : : /*
9379 : : * If there's no defined replica identity columns, treat as !key_required.
9380 : : * (This case should not be reachable from heap_update, since that should
9381 : : * calculate key_required accurately. But heap_delete just passes
9382 : : * constant true for key_required, so we can hit this case in deletes.)
9383 : : */
9384 [ + + ]: 53008 : if (bms_is_empty(idattrs))
9385 : 6021 : return NULL;
9386 : :
9387 : : /*
9388 : : * Construct a new tuple containing only the replica identity columns,
9389 : : * with nulls elsewhere. While we're at it, assert that the replica
9390 : : * identity columns aren't null.
9391 : : */
9392 : 46987 : heap_deform_tuple(tp, desc, values, nulls);
9393 : :
9394 [ + + ]: 150982 : for (int i = 0; i < desc->natts; i++)
9395 : : {
9396 [ + + ]: 103995 : if (bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
9397 : : idattrs))
9398 : : Assert(!nulls[i]);
9399 : : else
9400 : 56988 : nulls[i] = true;
9401 : : }
9402 : :
9403 : 46987 : key_tuple = heap_form_tuple(desc, values, nulls);
9404 : 46987 : *copy = true;
9405 : :
9406 : 46987 : bms_free(idattrs);
9407 : :
9408 : : /*
9409 : : * If the tuple, which by here only contains indexed columns, still has
9410 : : * toasted columns, force them to be inlined. This is somewhat unlikely
9411 : : * since there's limits on the size of indexed columns, so we don't
9412 : : * duplicate toast_flatten_tuple()s functionality in the above loop over
9413 : : * the indexed columns, even if it would be more efficient.
9414 : : */
9415 [ + + ]: 46987 : if (HeapTupleHasExternal(key_tuple))
9416 : : {
9417 : 4 : HeapTuple oldtup = key_tuple;
9418 : :
9419 : 4 : key_tuple = toast_flatten_tuple(oldtup, desc);
9420 : 4 : heap_freetuple(oldtup);
9421 : : }
9422 : :
9423 : 46987 : return key_tuple;
9424 : : }
9425 : :
9426 : : /*
9427 : : * HeapCheckForSerializableConflictOut
9428 : : * We are reading a tuple. If it's not visible, there may be a
9429 : : * rw-conflict out with the inserter. Otherwise, if it is visible to us
9430 : : * but has been deleted, there may be a rw-conflict out with the deleter.
9431 : : *
9432 : : * We will determine the top level xid of the writing transaction with which
9433 : : * we may be in conflict, and ask CheckForSerializableConflictOut() to check
9434 : : * for overlap with our own transaction.
9435 : : *
9436 : : * This function should be called just about anywhere in heapam.c where a
9437 : : * tuple has been read. The caller must hold at least a shared lock on the
9438 : : * buffer, because this function might set hint bits on the tuple. There is
9439 : : * currently no known reason to call this function from an index AM.
9440 : : */
9441 : : void
9442 : 42406854 : HeapCheckForSerializableConflictOut(bool visible, Relation relation,
9443 : : HeapTuple tuple, Buffer buffer,
9444 : : Snapshot snapshot)
9445 : : {
9446 : : TransactionId xid;
9447 : : HTSV_Result htsvResult;
9448 : :
9449 [ + + ]: 42406854 : if (!CheckForSerializableConflictOutNeeded(relation, snapshot))
9450 : 42398907 : return;
9451 : :
9452 : : /*
9453 : : * Check to see whether the tuple has been written to by a concurrent
9454 : : * transaction, either to create it not visible to us, or to delete it
9455 : : * while it is visible to us. The "visible" bool indicates whether the
9456 : : * tuple is visible to us, while HeapTupleSatisfiesVacuum checks what else
9457 : : * is going on with it.
9458 : : *
9459 : : * In the event of a concurrently inserted tuple that also happens to have
9460 : : * been concurrently updated (by a separate transaction), the xmin of the
9461 : : * tuple will be used -- not the updater's xid.
9462 : : */
9463 : 7947 : htsvResult = HeapTupleSatisfiesVacuum(tuple, TransactionXmin, buffer);
9464 [ + + + + : 7947 : switch (htsvResult)
- ]
9465 : : {
9466 : 7139 : case HEAPTUPLE_LIVE:
9467 [ + + ]: 7139 : if (visible)
9468 : 7124 : return;
9469 : 15 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9470 : 15 : break;
9471 : 361 : case HEAPTUPLE_RECENTLY_DEAD:
9472 : : case HEAPTUPLE_DELETE_IN_PROGRESS:
9473 [ + + ]: 361 : if (visible)
9474 : 286 : xid = HeapTupleHeaderGetUpdateXid(tuple->t_data);
9475 : : else
9476 : 75 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9477 : :
9478 [ + + ]: 361 : if (TransactionIdPrecedes(xid, TransactionXmin))
9479 : : {
9480 : : /* This is like the HEAPTUPLE_DEAD case */
9481 : : Assert(!visible);
9482 : 71 : return;
9483 : : }
9484 : 290 : break;
9485 : 327 : case HEAPTUPLE_INSERT_IN_PROGRESS:
9486 : 327 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9487 : 327 : break;
9488 : 120 : case HEAPTUPLE_DEAD:
9489 : : Assert(!visible);
9490 : 120 : return;
9491 : 0 : default:
9492 : :
9493 : : /*
9494 : : * The only way to get to this default clause is if a new value is
9495 : : * added to the enum type without adding it to this switch
9496 : : * statement. That's a bug, so elog.
9497 : : */
9498 [ # # ]: 0 : elog(ERROR, "unrecognized return value from HeapTupleSatisfiesVacuum: %u", htsvResult);
9499 : :
9500 : : /*
9501 : : * In spite of having all enum values covered and calling elog on
9502 : : * this default, some compilers think this is a code path which
9503 : : * allows xid to be used below without initialization. Silence
9504 : : * that warning.
9505 : : */
9506 : : xid = InvalidTransactionId;
9507 : : }
9508 : :
9509 : : Assert(TransactionIdIsValid(xid));
9510 : : Assert(TransactionIdFollowsOrEquals(xid, TransactionXmin));
9511 : :
9512 : : /*
9513 : : * Find top level xid. Bail out if xid is too early to be a conflict, or
9514 : : * if it's our own xid.
9515 : : */
9516 [ + + ]: 632 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
9517 : 64 : return;
9518 : 568 : xid = SubTransGetTopmostTransaction(xid);
9519 [ - + ]: 568 : if (TransactionIdPrecedes(xid, TransactionXmin))
9520 : 0 : return;
9521 : :
9522 : 568 : CheckForSerializableConflictOut(relation, xid, snapshot);
9523 : : }
|
