Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * heapam.c
4 : : * heap access method code
5 : : *
6 : : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
7 : : * Portions Copyright (c) 1994, Regents of the University of California
8 : : *
9 : : *
10 : : * IDENTIFICATION
11 : : * src/backend/access/heap/heapam.c
12 : : *
13 : : *
14 : : * INTERFACE ROUTINES
15 : : * heap_beginscan - begin relation scan
16 : : * heap_rescan - restart a relation scan
17 : : * heap_endscan - end relation scan
18 : : * heap_getnext - retrieve next tuple in scan
19 : : * heap_fetch - retrieve tuple with given tid
20 : : * heap_insert - insert tuple into a relation
21 : : * heap_multi_insert - insert multiple tuples into a relation
22 : : * heap_delete - delete a tuple from a relation
23 : : * heap_update - replace a tuple in a relation with another tuple
24 : : *
25 : : * NOTES
26 : : * This file contains the heap_ routines which implement
27 : : * the POSTGRES heap access method used for all POSTGRES
28 : : * relations.
29 : : *
30 : : *-------------------------------------------------------------------------
31 : : */
32 : : #include "postgres.h"
33 : :
34 : : #include "access/heapam.h"
35 : : #include "access/heaptoast.h"
36 : : #include "access/hio.h"
37 : : #include "access/multixact.h"
38 : : #include "access/subtrans.h"
39 : : #include "access/syncscan.h"
40 : : #include "access/valid.h"
41 : : #include "access/visibilitymap.h"
42 : : #include "access/xloginsert.h"
43 : : #include "catalog/pg_database.h"
44 : : #include "catalog/pg_database_d.h"
45 : : #include "commands/vacuum.h"
46 : : #include "pgstat.h"
47 : : #include "port/pg_bitutils.h"
48 : : #include "storage/lmgr.h"
49 : : #include "storage/predicate.h"
50 : : #include "storage/procarray.h"
51 : : #include "utils/datum.h"
52 : : #include "utils/injection_point.h"
53 : : #include "utils/inval.h"
54 : : #include "utils/spccache.h"
55 : : #include "utils/syscache.h"
56 : :
57 : :
58 : : static HeapTuple heap_prepare_insert(Relation relation, HeapTuple tup,
59 : : TransactionId xid, CommandId cid, int options);
60 : : static XLogRecPtr log_heap_update(Relation reln, Buffer oldbuf,
61 : : Buffer newbuf, HeapTuple oldtup,
62 : : HeapTuple newtup, HeapTuple old_key_tuple,
63 : : bool all_visible_cleared, bool new_all_visible_cleared);
64 : : #ifdef USE_ASSERT_CHECKING
65 : : static void check_lock_if_inplace_updateable_rel(Relation relation,
66 : : const ItemPointerData *otid,
67 : : HeapTuple newtup);
68 : : static void check_inplace_rel_lock(HeapTuple oldtup);
69 : : #endif
70 : : static Bitmapset *HeapDetermineColumnsInfo(Relation relation,
71 : : Bitmapset *interesting_cols,
72 : : Bitmapset *external_cols,
73 : : HeapTuple oldtup, HeapTuple newtup,
74 : : bool *has_external);
75 : : static bool heap_acquire_tuplock(Relation relation, const ItemPointerData *tid,
76 : : LockTupleMode mode, LockWaitPolicy wait_policy,
77 : : bool *have_tuple_lock);
78 : : static inline BlockNumber heapgettup_advance_block(HeapScanDesc scan,
79 : : BlockNumber block,
80 : : ScanDirection dir);
81 : : static pg_noinline BlockNumber heapgettup_initial_block(HeapScanDesc scan,
82 : : ScanDirection dir);
83 : : static void compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
84 : : uint16 old_infomask2, TransactionId add_to_xmax,
85 : : LockTupleMode mode, bool is_update,
86 : : TransactionId *result_xmax, uint16 *result_infomask,
87 : : uint16 *result_infomask2);
88 : : static TM_Result heap_lock_updated_tuple(Relation rel, HeapTuple tuple,
89 : : const ItemPointerData *ctid, TransactionId xid,
90 : : LockTupleMode mode);
91 : : static void GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
92 : : uint16 *new_infomask2);
93 : : static TransactionId MultiXactIdGetUpdateXid(TransactionId xmax,
94 : : uint16 t_infomask);
95 : : static bool DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
96 : : LockTupleMode lockmode, bool *current_is_member);
97 : : static void MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
98 : : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
99 : : int *remaining);
100 : : static bool ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
101 : : uint16 infomask, Relation rel, int *remaining,
102 : : bool logLockFailure);
103 : : static void index_delete_sort(TM_IndexDeleteOp *delstate);
104 : : static int bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate);
105 : : static XLogRecPtr log_heap_new_cid(Relation relation, HeapTuple tup);
106 : : static HeapTuple ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
107 : : bool *copy);
108 : :
109 : :
110 : : /*
111 : : * Each tuple lock mode has a corresponding heavyweight lock, and one or two
112 : : * corresponding MultiXactStatuses (one to merely lock tuples, another one to
113 : : * update them). This table (and the macros below) helps us determine the
114 : : * heavyweight lock mode and MultiXactStatus values to use for any particular
115 : : * tuple lock strength.
116 : : *
117 : : * These interact with InplaceUpdateTupleLock, an alias for ExclusiveLock.
118 : : *
119 : : * Don't look at lockstatus/updstatus directly! Use get_mxact_status_for_lock
120 : : * instead.
121 : : */
122 : : static const struct
123 : : {
124 : : LOCKMODE hwlock;
125 : : int lockstatus;
126 : : int updstatus;
127 : : }
128 : :
129 : : tupleLockExtraInfo[MaxLockTupleMode + 1] =
130 : : {
131 : : { /* LockTupleKeyShare */
132 : : AccessShareLock,
133 : : MultiXactStatusForKeyShare,
134 : : -1 /* KeyShare does not allow updating tuples */
135 : : },
136 : : { /* LockTupleShare */
137 : : RowShareLock,
138 : : MultiXactStatusForShare,
139 : : -1 /* Share does not allow updating tuples */
140 : : },
141 : : { /* LockTupleNoKeyExclusive */
142 : : ExclusiveLock,
143 : : MultiXactStatusForNoKeyUpdate,
144 : : MultiXactStatusNoKeyUpdate
145 : : },
146 : : { /* LockTupleExclusive */
147 : : AccessExclusiveLock,
148 : : MultiXactStatusForUpdate,
149 : : MultiXactStatusUpdate
150 : : }
151 : : };
152 : :
153 : : /* Get the LOCKMODE for a given MultiXactStatus */
154 : : #define LOCKMODE_from_mxstatus(status) \
155 : : (tupleLockExtraInfo[TUPLOCK_from_mxstatus((status))].hwlock)
156 : :
157 : : /*
158 : : * Acquire heavyweight locks on tuples, using a LockTupleMode strength value.
159 : : * This is more readable than having every caller translate it to lock.h's
160 : : * LOCKMODE.
161 : : */
162 : : #define LockTupleTuplock(rel, tup, mode) \
163 : : LockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
164 : : #define UnlockTupleTuplock(rel, tup, mode) \
165 : : UnlockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
166 : : #define ConditionalLockTupleTuplock(rel, tup, mode, log) \
167 : : ConditionalLockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock, (log))
168 : :
169 : : #ifdef USE_PREFETCH
170 : : /*
171 : : * heap_index_delete_tuples and index_delete_prefetch_buffer use this
172 : : * structure to coordinate prefetching activity
173 : : */
174 : : typedef struct
175 : : {
176 : : BlockNumber cur_hblkno;
177 : : int next_item;
178 : : int ndeltids;
179 : : TM_IndexDelete *deltids;
180 : : } IndexDeletePrefetchState;
181 : : #endif
182 : :
183 : : /* heap_index_delete_tuples bottom-up index deletion costing constants */
184 : : #define BOTTOMUP_MAX_NBLOCKS 6
185 : : #define BOTTOMUP_TOLERANCE_NBLOCKS 3
186 : :
187 : : /*
188 : : * heap_index_delete_tuples uses this when determining which heap blocks it
189 : : * must visit to help its bottom-up index deletion caller
190 : : */
191 : : typedef struct IndexDeleteCounts
192 : : {
193 : : int16 npromisingtids; /* Number of "promising" TIDs in group */
194 : : int16 ntids; /* Number of TIDs in group */
195 : : int16 ifirsttid; /* Offset to group's first deltid */
196 : : } IndexDeleteCounts;
197 : :
198 : : /*
199 : : * This table maps tuple lock strength values for each particular
200 : : * MultiXactStatus value.
201 : : */
202 : : static const int MultiXactStatusLock[MaxMultiXactStatus + 1] =
203 : : {
204 : : LockTupleKeyShare, /* ForKeyShare */
205 : : LockTupleShare, /* ForShare */
206 : : LockTupleNoKeyExclusive, /* ForNoKeyUpdate */
207 : : LockTupleExclusive, /* ForUpdate */
208 : : LockTupleNoKeyExclusive, /* NoKeyUpdate */
209 : : LockTupleExclusive /* Update */
210 : : };
211 : :
212 : : /* Get the LockTupleMode for a given MultiXactStatus */
213 : : #define TUPLOCK_from_mxstatus(status) \
214 : : (MultiXactStatusLock[(status)])
215 : :
216 : : /*
217 : : * Check that we have a valid snapshot if we might need TOAST access.
218 : : */
219 : : static inline void
201 nathan@postgresql.or 220 :CBC 10583584 : AssertHasSnapshotForToast(Relation rel)
221 : : {
222 : : #ifdef USE_ASSERT_CHECKING
223 : :
224 : : /* bootstrap mode in particular breaks this rule */
225 [ + + ]: 10583584 : if (!IsNormalProcessingMode())
226 : 597516 : return;
227 : :
228 : : /* if the relation doesn't have a TOAST table, we are good */
229 [ + + ]: 9986068 : if (!OidIsValid(rel->rd_rel->reltoastrelid))
230 : 5170316 : return;
231 : :
232 [ - + ]: 4815752 : Assert(HaveRegisteredOrActiveSnapshot());
233 : :
234 : : #endif /* USE_ASSERT_CHECKING */
235 : : }
236 : :
237 : : /* ----------------------------------------------------------------
238 : : * heap support routines
239 : : * ----------------------------------------------------------------
240 : : */
241 : :
242 : : /*
243 : : * Streaming read API callback for parallel sequential scans. Returns the next
244 : : * block the caller wants from the read stream or InvalidBlockNumber when done.
245 : : */
246 : : static BlockNumber
618 tmunro@postgresql.or 247 : 101342 : heap_scan_stream_read_next_parallel(ReadStream *stream,
248 : : void *callback_private_data,
249 : : void *per_buffer_data)
250 : : {
251 : 101342 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
252 : :
253 [ - + ]: 101342 : Assert(ScanDirectionIsForward(scan->rs_dir));
254 [ - + ]: 101342 : Assert(scan->rs_base.rs_parallel);
255 : :
256 [ + + ]: 101342 : if (unlikely(!scan->rs_inited))
257 : : {
258 : : /* parallel scan */
259 : 1645 : table_block_parallelscan_startblock_init(scan->rs_base.rs_rd,
260 : 1645 : scan->rs_parallelworkerdata,
20 drowley@postgresql.o 261 :GNC 1645 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel,
262 : : scan->rs_startblock,
263 : : scan->rs_numblocks);
264 : :
265 : : /* may return InvalidBlockNumber if there are no more blocks */
618 tmunro@postgresql.or 266 :CBC 3290 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
267 : 1645 : scan->rs_parallelworkerdata,
268 : 1645 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
269 : 1645 : scan->rs_inited = true;
270 : : }
271 : : else
272 : : {
273 : 99697 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
274 : 99697 : scan->rs_parallelworkerdata, (ParallelBlockTableScanDesc)
275 : 99697 : scan->rs_base.rs_parallel);
276 : : }
277 : :
278 : 101342 : return scan->rs_prefetch_block;
279 : : }
280 : :
281 : : /*
282 : : * Streaming read API callback for serial sequential and TID range scans.
283 : : * Returns the next block the caller wants from the read stream or
284 : : * InvalidBlockNumber when done.
285 : : */
286 : : static BlockNumber
287 : 3657485 : heap_scan_stream_read_next_serial(ReadStream *stream,
288 : : void *callback_private_data,
289 : : void *per_buffer_data)
290 : : {
291 : 3657485 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
292 : :
293 [ + + ]: 3657485 : if (unlikely(!scan->rs_inited))
294 : : {
295 : 983976 : scan->rs_prefetch_block = heapgettup_initial_block(scan, scan->rs_dir);
296 : 983976 : scan->rs_inited = true;
297 : : }
298 : : else
299 : 2673509 : scan->rs_prefetch_block = heapgettup_advance_block(scan,
300 : : scan->rs_prefetch_block,
301 : : scan->rs_dir);
302 : :
303 : 3657485 : return scan->rs_prefetch_block;
304 : : }
305 : :
306 : : /*
307 : : * Read stream API callback for bitmap heap scans.
308 : : * Returns the next block the caller wants from the read stream or
309 : : * InvalidBlockNumber when done.
310 : : */
311 : : static BlockNumber
277 melanieplageman@gmai 312 : 214378 : bitmapheap_stream_read_next(ReadStream *pgsr, void *private_data,
313 : : void *per_buffer_data)
314 : : {
315 : 214378 : TBMIterateResult *tbmres = per_buffer_data;
316 : 214378 : BitmapHeapScanDesc bscan = (BitmapHeapScanDesc) private_data;
317 : 214378 : HeapScanDesc hscan = (HeapScanDesc) bscan;
318 : 214378 : TableScanDesc sscan = &hscan->rs_base;
319 : :
320 : : for (;;)
321 : : {
322 [ + + ]: 214378 : CHECK_FOR_INTERRUPTS();
323 : :
324 : : /* no more entries in the bitmap */
325 [ + + ]: 214378 : if (!tbm_iterate(&sscan->st.rs_tbmiterator, tbmres))
326 : 12593 : return InvalidBlockNumber;
327 : :
328 : : /*
329 : : * Ignore any claimed entries past what we think is the end of the
330 : : * relation. It may have been extended after the start of our scan (we
331 : : * only hold an AccessShareLock, and it could be inserts from this
332 : : * backend). We don't take this optimization in SERIALIZABLE
333 : : * isolation though, as we need to examine all invisible tuples
334 : : * reachable by the index.
335 : : */
336 [ + + ]: 201785 : if (!IsolationIsSerializable() &&
337 [ - + ]: 201676 : tbmres->blockno >= hscan->rs_nblocks)
277 melanieplageman@gmai 338 :UBC 0 : continue;
339 : :
277 melanieplageman@gmai 340 :CBC 201785 : return tbmres->blockno;
341 : : }
342 : :
343 : : /* not reachable */
344 : : Assert(false);
345 : : }
346 : :
347 : : /* ----------------
348 : : * initscan - scan code common to heap_beginscan and heap_rescan
349 : : * ----------------
350 : : */
351 : : static void
3798 tgl@sss.pgh.pa.us 352 : 1008385 : initscan(HeapScanDesc scan, ScanKey key, bool keep_startblock)
353 : : {
2473 andres@anarazel.de 354 : 1008385 : ParallelBlockTableScanDesc bpscan = NULL;
355 : : bool allow_strat;
356 : : bool allow_sync;
357 : :
358 : : /*
359 : : * Determine the number of blocks we have to scan.
360 : : *
361 : : * It is sufficient to do this once at scan start, since any tuples added
362 : : * while the scan is in progress will be invisible to my snapshot anyway.
363 : : * (That is not true when using a non-MVCC snapshot. However, we couldn't
364 : : * guarantee to return tuples added after scan start anyway, since they
365 : : * might go into pages we already scanned. To guarantee consistent
366 : : * results for a non-MVCC snapshot, the caller must hold some higher-level
367 : : * lock that ensures the interesting tuple(s) won't change.)
368 : : */
369 [ + + ]: 1008385 : if (scan->rs_base.rs_parallel != NULL)
370 : : {
371 : 2172 : bpscan = (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel;
372 : 2172 : scan->rs_nblocks = bpscan->phs_nblocks;
373 : : }
374 : : else
375 : 1006213 : scan->rs_nblocks = RelationGetNumberOfBlocks(scan->rs_base.rs_rd);
376 : :
377 : : /*
378 : : * If the table is large relative to NBuffers, use a bulk-read access
379 : : * strategy and enable synchronized scanning (see syncscan.c). Although
380 : : * the thresholds for these features could be different, we make them the
381 : : * same so that there are only two behaviors to tune rather than four.
382 : : * (However, some callers need to be able to disable one or both of these
383 : : * behaviors, independently of the size of the table; also there is a GUC
384 : : * variable that can disable synchronized scanning.)
385 : : *
386 : : * Note that table_block_parallelscan_initialize has a very similar test;
387 : : * if you change this, consider changing that one, too.
388 : : */
389 [ + + ]: 1008383 : if (!RelationUsesLocalBuffers(scan->rs_base.rs_rd) &&
6766 tgl@sss.pgh.pa.us 390 [ + + ]: 1001057 : scan->rs_nblocks > NBuffers / 4)
391 : : {
2404 andres@anarazel.de 392 : 14313 : allow_strat = (scan->rs_base.rs_flags & SO_ALLOW_STRAT) != 0;
393 : 14313 : allow_sync = (scan->rs_base.rs_flags & SO_ALLOW_SYNC) != 0;
394 : : }
395 : : else
6547 tgl@sss.pgh.pa.us 396 : 994070 : allow_strat = allow_sync = false;
397 : :
398 [ + + ]: 1008383 : if (allow_strat)
399 : : {
400 : : /* During a rescan, keep the previous strategy object. */
6776 401 [ + + ]: 12977 : if (scan->rs_strategy == NULL)
402 : 12806 : scan->rs_strategy = GetAccessStrategy(BAS_BULKREAD);
403 : : }
404 : : else
405 : : {
406 [ - + ]: 995406 : if (scan->rs_strategy != NULL)
6776 tgl@sss.pgh.pa.us 407 :UBC 0 : FreeAccessStrategy(scan->rs_strategy);
6776 tgl@sss.pgh.pa.us 408 :CBC 995406 : scan->rs_strategy = NULL;
409 : : }
410 : :
2473 andres@anarazel.de 411 [ + + ]: 1008383 : if (scan->rs_base.rs_parallel != NULL)
412 : : {
413 : : /* For parallel scan, believe whatever ParallelTableScanDesc says. */
2404 414 [ + + ]: 2172 : if (scan->rs_base.rs_parallel->phs_syncscan)
415 : 2 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
416 : : else
417 : 2170 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
418 : :
419 : : /*
420 : : * If not rescanning, initialize the startblock. Finding the actual
421 : : * start location is done in table_block_parallelscan_startblock_init,
422 : : * based on whether an alternative start location has been set with
423 : : * heap_setscanlimits, or using the syncscan location, when syncscan
424 : : * is enabled.
425 : : */
19 drowley@postgresql.o 426 [ + + ]:GNC 2172 : if (!keep_startblock)
427 : 2058 : scan->rs_startblock = InvalidBlockNumber;
428 : : }
429 : : else
430 : : {
431 [ + + ]: 1006211 : if (keep_startblock)
432 : : {
433 : : /*
434 : : * When rescanning, we want to keep the previous startblock
435 : : * setting, so that rewinding a cursor doesn't generate surprising
436 : : * results. Reset the active syncscan setting, though.
437 : : */
438 [ + + + + ]: 632441 : if (allow_sync && synchronize_seqscans)
439 : 50 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
440 : : else
441 : 632391 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
442 : : }
443 [ + + + + ]: 373770 : else if (allow_sync && synchronize_seqscans)
444 : : {
445 : 72 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
446 : 72 : scan->rs_startblock = ss_get_location(scan->rs_base.rs_rd, scan->rs_nblocks);
447 : : }
448 : : else
449 : : {
450 : 373698 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
451 : 373698 : scan->rs_startblock = 0;
452 : : }
453 : : }
454 : :
4058 alvherre@alvh.no-ip. 455 :CBC 1008383 : scan->rs_numblocks = InvalidBlockNumber;
7326 tgl@sss.pgh.pa.us 456 : 1008383 : scan->rs_inited = false;
8957 457 : 1008383 : scan->rs_ctup.t_data = NULL;
7326 458 : 1008383 : ItemPointerSetInvalid(&scan->rs_ctup.t_self);
8957 459 : 1008383 : scan->rs_cbuf = InvalidBuffer;
7326 460 : 1008383 : scan->rs_cblock = InvalidBlockNumber;
364 melanieplageman@gmai 461 : 1008383 : scan->rs_ntuples = 0;
462 : 1008383 : scan->rs_cindex = 0;
463 : :
464 : : /*
465 : : * Initialize to ForwardScanDirection because it is most common and
466 : : * because heap scans go forward before going backward (e.g. CURSORs).
467 : : */
618 tmunro@postgresql.or 468 : 1008383 : scan->rs_dir = ForwardScanDirection;
469 : 1008383 : scan->rs_prefetch_block = InvalidBlockNumber;
470 : :
471 : : /* page-at-a-time fields are always invalid when not rs_inited */
472 : :
473 : : /*
474 : : * copy the scan key, if appropriate
475 : : */
1385 tgl@sss.pgh.pa.us 476 [ + + + + ]: 1008383 : if (key != NULL && scan->rs_base.rs_nkeys > 0)
2473 andres@anarazel.de 477 : 214084 : memcpy(scan->rs_base.rs_key, key, scan->rs_base.rs_nkeys * sizeof(ScanKeyData));
478 : :
479 : : /*
480 : : * Currently, we only have a stats counter for sequential heap scans (but
481 : : * e.g for bitmap scans the underlying bitmap index scans will be counted,
482 : : * and for sample scans we update stats for tuple fetches).
483 : : */
2404 484 [ + + ]: 1008383 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN)
2473 485 [ + + + + : 985179 : pgstat_count_heap_scan(scan->rs_base.rs_rd);
+ + ]
10753 scrappy@hub.org 486 : 1008383 : }
487 : :
488 : : /*
489 : : * heap_setscanlimits - restrict range of a heapscan
490 : : *
491 : : * startBlk is the page to start at
492 : : * numBlks is number of pages to scan (InvalidBlockNumber means "all")
493 : : */
494 : : void
2473 andres@anarazel.de 495 : 2878 : heap_setscanlimits(TableScanDesc sscan, BlockNumber startBlk, BlockNumber numBlks)
496 : : {
497 : 2878 : HeapScanDesc scan = (HeapScanDesc) sscan;
498 : :
3802 tgl@sss.pgh.pa.us 499 [ - + ]: 2878 : Assert(!scan->rs_inited); /* else too late to change */
500 : : /* else rs_startblock is significant */
2404 andres@anarazel.de 501 [ - + ]: 2878 : Assert(!(scan->rs_base.rs_flags & SO_ALLOW_SYNC));
502 : :
503 : : /* Check startBlk is valid (but allow case of zero blocks...) */
3802 tgl@sss.pgh.pa.us 504 [ + + - + ]: 2878 : Assert(startBlk == 0 || startBlk < scan->rs_nblocks);
505 : :
4058 alvherre@alvh.no-ip. 506 : 2878 : scan->rs_startblock = startBlk;
507 : 2878 : scan->rs_numblocks = numBlks;
508 : 2878 : }
509 : :
510 : : /*
511 : : * Per-tuple loop for heap_prepare_pagescan(). Pulled out so it can be called
512 : : * multiple times, with constant arguments for all_visible,
513 : : * check_serializable.
514 : : */
515 : : pg_attribute_always_inline
516 : : static int
619 andres@anarazel.de 517 : 2664157 : page_collect_tuples(HeapScanDesc scan, Snapshot snapshot,
518 : : Page page, Buffer buffer,
519 : : BlockNumber block, int lines,
520 : : bool all_visible, bool check_serializable)
521 : : {
620 522 : 2664157 : int ntup = 0;
523 : : OffsetNumber lineoff;
524 : :
525 [ + + ]: 134047296 : for (lineoff = FirstOffsetNumber; lineoff <= lines; lineoff++)
526 : : {
527 : 131383147 : ItemId lpp = PageGetItemId(page, lineoff);
528 : : HeapTupleData loctup;
529 : : bool valid;
530 : :
531 [ + + ]: 131383147 : if (!ItemIdIsNormal(lpp))
532 : 25525295 : continue;
533 : :
534 : 105857852 : loctup.t_data = (HeapTupleHeader) PageGetItem(page, lpp);
535 : 105857852 : loctup.t_len = ItemIdGetLength(lpp);
536 : 105857852 : loctup.t_tableOid = RelationGetRelid(scan->rs_base.rs_rd);
537 : 105857852 : ItemPointerSet(&(loctup.t_self), block, lineoff);
538 : :
539 [ + + ]: 105857852 : if (all_visible)
540 : 40576159 : valid = true;
541 : : else
542 : 65281693 : valid = HeapTupleSatisfiesVisibility(&loctup, snapshot, buffer);
543 : :
544 [ + + ]: 105857852 : if (check_serializable)
545 : 1421 : HeapCheckForSerializableConflictOut(valid, scan->rs_base.rs_rd,
546 : : &loctup, buffer, snapshot);
547 : :
548 [ + + ]: 105857844 : if (valid)
549 : : {
550 : 98452210 : scan->rs_vistuples[ntup] = lineoff;
551 : 98452210 : ntup++;
552 : : }
553 : : }
554 : :
555 [ - + ]: 2664149 : Assert(ntup <= MaxHeapTuplesPerPage);
556 : :
557 : 2664149 : return ntup;
558 : : }
559 : :
560 : : /*
561 : : * heap_prepare_pagescan - Prepare current scan page to be scanned in pagemode
562 : : *
563 : : * Preparation currently consists of 1. prune the scan's rs_cbuf page, and 2.
564 : : * fill the rs_vistuples[] array with the OffsetNumbers of visible tuples.
565 : : */
566 : : void
622 drowley@postgresql.o 567 : 2664157 : heap_prepare_pagescan(TableScanDesc sscan)
568 : : {
2473 andres@anarazel.de 569 : 2664157 : HeapScanDesc scan = (HeapScanDesc) sscan;
622 drowley@postgresql.o 570 : 2664157 : Buffer buffer = scan->rs_cbuf;
571 : 2664157 : BlockNumber block = scan->rs_cblock;
572 : : Snapshot snapshot;
573 : : Page page;
574 : : int lines;
575 : : bool all_visible;
576 : : bool check_serializable;
577 : :
578 [ - + ]: 2664157 : Assert(BufferGetBlockNumber(buffer) == block);
579 : :
580 : : /* ensure we're not accidentally being used when not in pagemode */
581 [ - + ]: 2664157 : Assert(scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE);
2473 andres@anarazel.de 582 : 2664157 : snapshot = scan->rs_base.rs_snapshot;
583 : :
584 : : /*
585 : : * Prune and repair fragmentation for the whole page, if possible.
586 : : */
587 : 2664157 : heap_page_prune_opt(scan->rs_base.rs_rd, buffer);
588 : :
589 : : /*
590 : : * We must hold share lock on the buffer content while examining tuple
591 : : * visibility. Afterwards, however, the tuples we have found to be
592 : : * visible are guaranteed good as long as we hold the buffer pin.
593 : : */
7326 tgl@sss.pgh.pa.us 594 : 2664157 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
595 : :
1127 peter@eisentraut.org 596 : 2664157 : page = BufferGetPage(buffer);
597 : 2664157 : lines = PageGetMaxOffsetNumber(page);
598 : :
599 : : /*
600 : : * If the all-visible flag indicates that all tuples on the page are
601 : : * visible to everyone, we can skip the per-tuple visibility tests.
602 : : *
603 : : * Note: In hot standby, a tuple that's already visible to all
604 : : * transactions on the primary might still be invisible to a read-only
605 : : * transaction in the standby. We partly handle this problem by tracking
606 : : * the minimum xmin of visible tuples as the cut-off XID while marking a
607 : : * page all-visible on the primary and WAL log that along with the
608 : : * visibility map SET operation. In hot standby, we wait for (or abort)
609 : : * all transactions that can potentially may not see one or more tuples on
610 : : * the page. That's how index-only scans work fine in hot standby. A
611 : : * crucial difference between index-only scans and heap scans is that the
612 : : * index-only scan completely relies on the visibility map where as heap
613 : : * scan looks at the page-level PD_ALL_VISIBLE flag. We are not sure if
614 : : * the page-level flag can be trusted in the same way, because it might
615 : : * get propagated somehow without being explicitly WAL-logged, e.g. via a
616 : : * full page write. Until we can prove that beyond doubt, let's check each
617 : : * tuple for visibility the hard way.
618 : : */
619 [ + + + + ]: 2664157 : all_visible = PageIsAllVisible(page) && !snapshot->takenDuringRecovery;
620 : : check_serializable =
620 andres@anarazel.de 621 : 2664157 : CheckForSerializableConflictOutNeeded(scan->rs_base.rs_rd, snapshot);
622 : :
623 : : /*
624 : : * We call page_collect_tuples() with constant arguments, to get the
625 : : * compiler to constant fold the constant arguments. Separate calls with
626 : : * constant arguments, rather than variables, are needed on several
627 : : * compilers to actually perform constant folding.
628 : : */
629 [ + + ]: 2664157 : if (likely(all_visible))
630 : : {
631 [ + - ]: 988129 : if (likely(!check_serializable))
619 632 : 988129 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
633 : : block, lines, true, false);
634 : : else
619 andres@anarazel.de 635 :UBC 0 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
636 : : block, lines, true, true);
637 : : }
638 : : else
639 : : {
620 andres@anarazel.de 640 [ + + ]:CBC 1676028 : if (likely(!check_serializable))
619 641 : 1675396 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
642 : : block, lines, false, false);
643 : : else
644 : 632 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
645 : : block, lines, false, true);
646 : : }
647 : :
7326 tgl@sss.pgh.pa.us 648 : 2664149 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
649 : 2664149 : }
650 : :
651 : : /*
652 : : * heap_fetch_next_buffer - read and pin the next block from MAIN_FORKNUM.
653 : : *
654 : : * Read the next block of the scan relation from the read stream and save it
655 : : * in the scan descriptor. It is already pinned.
656 : : */
657 : : static inline void
622 drowley@postgresql.o 658 : 3579175 : heap_fetch_next_buffer(HeapScanDesc scan, ScanDirection dir)
659 : : {
618 tmunro@postgresql.or 660 [ - + ]: 3579175 : Assert(scan->rs_read_stream);
661 : :
662 : : /* release previous scan buffer, if any */
622 drowley@postgresql.o 663 [ + + ]: 3579175 : if (BufferIsValid(scan->rs_cbuf))
664 : : {
665 : 2593553 : ReleaseBuffer(scan->rs_cbuf);
666 : 2593553 : scan->rs_cbuf = InvalidBuffer;
667 : : }
668 : :
669 : : /*
670 : : * Be sure to check for interrupts at least once per page. Checks at
671 : : * higher code levels won't be able to stop a seqscan that encounters many
672 : : * pages' worth of consecutive dead tuples.
673 : : */
674 [ + + ]: 3579175 : CHECK_FOR_INTERRUPTS();
675 : :
676 : : /*
677 : : * If the scan direction is changing, reset the prefetch block to the
678 : : * current block. Otherwise, we will incorrectly prefetch the blocks
679 : : * between the prefetch block and the current block again before
680 : : * prefetching blocks in the new, correct scan direction.
681 : : */
618 tmunro@postgresql.or 682 [ + + ]: 3579172 : if (unlikely(scan->rs_dir != dir))
683 : : {
684 : 76 : scan->rs_prefetch_block = scan->rs_cblock;
685 : 76 : read_stream_reset(scan->rs_read_stream);
686 : : }
687 : :
688 : 3579172 : scan->rs_dir = dir;
689 : :
690 : 3579172 : scan->rs_cbuf = read_stream_next_buffer(scan->rs_read_stream, NULL);
691 [ + + ]: 3579150 : if (BufferIsValid(scan->rs_cbuf))
692 : 2751862 : scan->rs_cblock = BufferGetBlockNumber(scan->rs_cbuf);
622 drowley@postgresql.o 693 : 3579150 : }
694 : :
695 : : /*
696 : : * heapgettup_initial_block - return the first BlockNumber to scan
697 : : *
698 : : * Returns InvalidBlockNumber when there are no blocks to scan. This can
699 : : * occur with empty tables and in parallel scans when parallel workers get all
700 : : * of the pages before we can get a chance to get our first page.
701 : : */
702 : : static pg_noinline BlockNumber
1049 703 : 983976 : heapgettup_initial_block(HeapScanDesc scan, ScanDirection dir)
704 : : {
705 [ - + ]: 983976 : Assert(!scan->rs_inited);
618 tmunro@postgresql.or 706 [ - + ]: 983976 : Assert(scan->rs_base.rs_parallel == NULL);
707 : :
708 : : /* When there are no pages to scan, return InvalidBlockNumber */
1049 drowley@postgresql.o 709 [ + + + + ]: 983976 : if (scan->rs_nblocks == 0 || scan->rs_numblocks == 0)
710 : 512723 : return InvalidBlockNumber;
711 : :
712 [ + + ]: 471253 : if (ScanDirectionIsForward(dir))
713 : : {
618 tmunro@postgresql.or 714 : 471222 : return scan->rs_startblock;
715 : : }
716 : : else
717 : : {
718 : : /*
719 : : * Disable reporting to syncscan logic in a backwards scan; it's not
720 : : * very likely anyone else is doing the same thing at the same time,
721 : : * and much more likely that we'll just bollix things for forward
722 : : * scanners.
723 : : */
1049 drowley@postgresql.o 724 : 31 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
725 : :
726 : : /*
727 : : * Start from last page of the scan. Ensure we take into account
728 : : * rs_numblocks if it's been adjusted by heap_setscanlimits().
729 : : */
730 [ + + ]: 31 : if (scan->rs_numblocks != InvalidBlockNumber)
731 : 3 : return (scan->rs_startblock + scan->rs_numblocks - 1) % scan->rs_nblocks;
732 : :
733 [ - + ]: 28 : if (scan->rs_startblock > 0)
1049 drowley@postgresql.o 734 :UBC 0 : return scan->rs_startblock - 1;
735 : :
1049 drowley@postgresql.o 736 :CBC 28 : return scan->rs_nblocks - 1;
737 : : }
738 : : }
739 : :
740 : :
741 : : /*
742 : : * heapgettup_start_page - helper function for heapgettup()
743 : : *
744 : : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
745 : : * to the number of tuples on this page. Also set *lineoff to the first
746 : : * offset to scan with forward scans getting the first offset and backward
747 : : * getting the final offset on the page.
748 : : */
749 : : static Page
1048 750 : 91982 : heapgettup_start_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
751 : : OffsetNumber *lineoff)
752 : : {
753 : : Page page;
754 : :
755 [ - + ]: 91982 : Assert(scan->rs_inited);
756 [ - + ]: 91982 : Assert(BufferIsValid(scan->rs_cbuf));
757 : :
758 : : /* Caller is responsible for ensuring buffer is locked if needed */
759 : 91982 : page = BufferGetPage(scan->rs_cbuf);
760 : :
1044 761 : 91982 : *linesleft = PageGetMaxOffsetNumber(page) - FirstOffsetNumber + 1;
762 : :
1048 763 [ + - ]: 91982 : if (ScanDirectionIsForward(dir))
764 : 91982 : *lineoff = FirstOffsetNumber;
765 : : else
1048 drowley@postgresql.o 766 :UBC 0 : *lineoff = (OffsetNumber) (*linesleft);
767 : :
768 : : /* lineoff now references the physically previous or next tid */
1048 drowley@postgresql.o 769 :CBC 91982 : return page;
770 : : }
771 : :
772 : :
773 : : /*
774 : : * heapgettup_continue_page - helper function for heapgettup()
775 : : *
776 : : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
777 : : * to the number of tuples left to scan on this page. Also set *lineoff to
778 : : * the next offset to scan according to the ScanDirection in 'dir'.
779 : : */
780 : : static inline Page
781 : 7465663 : heapgettup_continue_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
782 : : OffsetNumber *lineoff)
783 : : {
784 : : Page page;
785 : :
786 [ - + ]: 7465663 : Assert(scan->rs_inited);
787 [ - + ]: 7465663 : Assert(BufferIsValid(scan->rs_cbuf));
788 : :
789 : : /* Caller is responsible for ensuring buffer is locked if needed */
790 : 7465663 : page = BufferGetPage(scan->rs_cbuf);
791 : :
792 [ + - ]: 7465663 : if (ScanDirectionIsForward(dir))
793 : : {
794 : 7465663 : *lineoff = OffsetNumberNext(scan->rs_coffset);
795 : 7465663 : *linesleft = PageGetMaxOffsetNumber(page) - (*lineoff) + 1;
796 : : }
797 : : else
798 : : {
799 : : /*
800 : : * The previous returned tuple may have been vacuumed since the
801 : : * previous scan when we use a non-MVCC snapshot, so we must
802 : : * re-establish the lineoff <= PageGetMaxOffsetNumber(page) invariant
803 : : */
1048 drowley@postgresql.o 804 [ # # ]:UBC 0 : *lineoff = Min(PageGetMaxOffsetNumber(page), OffsetNumberPrev(scan->rs_coffset));
805 : 0 : *linesleft = *lineoff;
806 : : }
807 : :
808 : : /* lineoff now references the physically previous or next tid */
1048 drowley@postgresql.o 809 :CBC 7465663 : return page;
810 : : }
811 : :
812 : : /*
813 : : * heapgettup_advance_block - helper for heap_fetch_next_buffer()
814 : : *
815 : : * Given the current block number, the scan direction, and various information
816 : : * contained in the scan descriptor, calculate the BlockNumber to scan next
817 : : * and return it. If there are no further blocks to scan, return
818 : : * InvalidBlockNumber to indicate this fact to the caller.
819 : : *
820 : : * This should not be called to determine the initial block number -- only for
821 : : * subsequent blocks.
822 : : *
823 : : * This also adjusts rs_numblocks when a limit has been imposed by
824 : : * heap_setscanlimits().
825 : : */
826 : : static inline BlockNumber
827 : 2673509 : heapgettup_advance_block(HeapScanDesc scan, BlockNumber block, ScanDirection dir)
828 : : {
618 tmunro@postgresql.or 829 [ - + ]: 2673509 : Assert(scan->rs_base.rs_parallel == NULL);
830 : :
831 [ + + ]: 2673509 : if (likely(ScanDirectionIsForward(dir)))
832 : : {
833 : 2673451 : block++;
834 : :
835 : : /* wrap back to the start of the heap */
836 [ + + ]: 2673451 : if (block >= scan->rs_nblocks)
837 : 372111 : block = 0;
838 : :
839 : : /*
840 : : * Report our new scan position for synchronization purposes. We don't
841 : : * do that when moving backwards, however. That would just mess up any
842 : : * other forward-moving scanners.
843 : : *
844 : : * Note: we do this before checking for end of scan so that the final
845 : : * state of the position hint is back at the start of the rel. That's
846 : : * not strictly necessary, but otherwise when you run the same query
847 : : * multiple times the starting position would shift a little bit
848 : : * backwards on every invocation, which is confusing. We don't
849 : : * guarantee any specific ordering in general, though.
850 : : */
851 [ + + ]: 2673451 : if (scan->rs_base.rs_flags & SO_ALLOW_SYNC)
852 : 11265 : ss_report_location(scan->rs_base.rs_rd, block);
853 : :
854 : : /* we're done if we're back at where we started */
855 [ + + ]: 2673451 : if (block == scan->rs_startblock)
856 : 372070 : return InvalidBlockNumber;
857 : :
858 : : /* check if the limit imposed by heap_setscanlimits() is met */
859 [ + + ]: 2301381 : if (scan->rs_numblocks != InvalidBlockNumber)
860 : : {
861 [ + + ]: 2484 : if (--scan->rs_numblocks == 0)
862 : 1550 : return InvalidBlockNumber;
863 : : }
864 : :
865 : 2299831 : return block;
866 : : }
867 : : else
868 : : {
869 : : /* we're done if the last block is the start position */
1048 drowley@postgresql.o 870 [ + - ]: 58 : if (block == scan->rs_startblock)
871 : 58 : return InvalidBlockNumber;
872 : :
873 : : /* check if the limit imposed by heap_setscanlimits() is met */
1048 drowley@postgresql.o 874 [ # # ]:UBC 0 : if (scan->rs_numblocks != InvalidBlockNumber)
875 : : {
876 [ # # ]: 0 : if (--scan->rs_numblocks == 0)
877 : 0 : return InvalidBlockNumber;
878 : : }
879 : :
880 : : /* wrap to the end of the heap when the last page was page 0 */
881 [ # # ]: 0 : if (block == 0)
882 : 0 : block = scan->rs_nblocks;
883 : :
884 : 0 : block--;
885 : :
886 : 0 : return block;
887 : : }
888 : : }
889 : :
890 : : /* ----------------
891 : : * heapgettup - fetch next heap tuple
892 : : *
893 : : * Initialize the scan if not already done; then advance to the next
894 : : * tuple as indicated by "dir"; return the next tuple in scan->rs_ctup,
895 : : * or set scan->rs_ctup.t_data = NULL if no more tuples.
896 : : *
897 : : * Note: the reason nkeys/key are passed separately, even though they are
898 : : * kept in the scan descriptor, is that the caller may not want us to check
899 : : * the scankeys.
900 : : *
901 : : * Note: when we fall off the end of the scan in either direction, we
902 : : * reset rs_inited. This means that a further request with the same
903 : : * scan direction will restart the scan, which is a bit odd, but a
904 : : * request with the opposite scan direction will start a fresh scan
905 : : * in the proper direction. The latter is required behavior for cursors,
906 : : * while the former case is generally undefined behavior in Postgres
907 : : * so we don't care too much.
908 : : * ----------------
909 : : */
910 : : static void
7326 tgl@sss.pgh.pa.us 911 :CBC 7486495 : heapgettup(HeapScanDesc scan,
912 : : ScanDirection dir,
913 : : int nkeys,
914 : : ScanKey key)
915 : : {
916 : 7486495 : HeapTuple tuple = &(scan->rs_ctup);
917 : : Page page;
918 : : OffsetNumber lineoff;
919 : : int linesleft;
920 : :
622 drowley@postgresql.o 921 [ + + ]: 7486495 : if (likely(scan->rs_inited))
922 : : {
923 : : /* continue from previously returned page/tuple */
1048 924 : 7465663 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
925 : 7465663 : page = heapgettup_continue_page(scan, dir, &linesleft, &lineoff);
1044 926 : 7465663 : goto continue_page;
927 : : }
928 : :
929 : : /*
930 : : * advance the scan until we find a qualifying tuple or run out of stuff
931 : : * to scan
932 : : */
933 : : while (true)
934 : : {
622 935 : 112664 : heap_fetch_next_buffer(scan, dir);
936 : :
937 : : /* did we run out of blocks to scan? */
938 [ + + ]: 112664 : if (!BufferIsValid(scan->rs_cbuf))
939 : 20682 : break;
940 : :
941 [ - + ]: 91982 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
942 : :
1044 943 : 91982 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
944 : 91982 : page = heapgettup_start_page(scan, dir, &linesleft, &lineoff);
945 : 7557645 : continue_page:
946 : :
947 : : /*
948 : : * Only continue scanning the page while we have lines left.
949 : : *
950 : : * Note that this protects us from accessing line pointers past
951 : : * PageGetMaxOffsetNumber(); both for forward scans when we resume the
952 : : * table scan, and for when we start scanning a new page.
953 : : */
954 [ + + ]: 7594471 : for (; linesleft > 0; linesleft--, lineoff += dir)
955 : : {
956 : : bool visible;
957 : 7502639 : ItemId lpp = PageGetItemId(page, lineoff);
958 : :
959 [ + + ]: 7502639 : if (!ItemIdIsNormal(lpp))
960 : 31596 : continue;
961 : :
962 : 7471043 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
963 : 7471043 : tuple->t_len = ItemIdGetLength(lpp);
622 964 : 7471043 : ItemPointerSet(&(tuple->t_self), scan->rs_cblock, lineoff);
965 : :
1044 966 : 7471043 : visible = HeapTupleSatisfiesVisibility(tuple,
967 : : scan->rs_base.rs_snapshot,
968 : : scan->rs_cbuf);
969 : :
970 : 7471043 : HeapCheckForSerializableConflictOut(visible, scan->rs_base.rs_rd,
971 : : tuple, scan->rs_cbuf,
972 : : scan->rs_base.rs_snapshot);
973 : :
974 : : /* skip tuples not visible to this snapshot */
975 [ + + ]: 7471043 : if (!visible)
976 : 5230 : continue;
977 : :
978 : : /* skip any tuples that don't match the scan key */
979 [ - + ]: 7465813 : if (key != NULL &&
1044 drowley@postgresql.o 980 [ # # ]:UBC 0 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
981 : : nkeys, key))
982 : 0 : continue;
983 : :
1044 drowley@postgresql.o 984 :CBC 7465813 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
985 : 7465813 : scan->rs_coffset = lineoff;
986 : 7465813 : return;
987 : : }
988 : :
989 : : /*
990 : : * if we get here, it means we've exhausted the items on this page and
991 : : * it's time to move to the next.
992 : : */
7326 tgl@sss.pgh.pa.us 993 : 91832 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
994 : : }
995 : :
996 : : /* end of scan */
1044 drowley@postgresql.o 997 [ - + ]: 20682 : if (BufferIsValid(scan->rs_cbuf))
1044 drowley@postgresql.o 998 :UBC 0 : ReleaseBuffer(scan->rs_cbuf);
999 : :
1044 drowley@postgresql.o 1000 :CBC 20682 : scan->rs_cbuf = InvalidBuffer;
1001 : 20682 : scan->rs_cblock = InvalidBlockNumber;
618 tmunro@postgresql.or 1002 : 20682 : scan->rs_prefetch_block = InvalidBlockNumber;
1044 drowley@postgresql.o 1003 : 20682 : tuple->t_data = NULL;
1004 : 20682 : scan->rs_inited = false;
1005 : : }
1006 : :
1007 : : /* ----------------
1008 : : * heapgettup_pagemode - fetch next heap tuple in page-at-a-time mode
1009 : : *
1010 : : * Same API as heapgettup, but used in page-at-a-time mode
1011 : : *
1012 : : * The internal logic is much the same as heapgettup's too, but there are some
1013 : : * differences: we do not take the buffer content lock (that only needs to
1014 : : * happen inside heap_prepare_pagescan), and we iterate through just the
1015 : : * tuples listed in rs_vistuples[] rather than all tuples on the page. Notice
1016 : : * that lineindex is 0-based, where the corresponding loop variable lineoff in
1017 : : * heapgettup is 1-based.
1018 : : * ----------------
1019 : : */
1020 : : static void
7326 tgl@sss.pgh.pa.us 1021 : 49036328 : heapgettup_pagemode(HeapScanDesc scan,
1022 : : ScanDirection dir,
1023 : : int nkeys,
1024 : : ScanKey key)
1025 : : {
1026 : 49036328 : HeapTuple tuple = &(scan->rs_ctup);
1027 : : Page page;
1028 : : uint32 lineindex;
1029 : : uint32 linesleft;
1030 : :
622 drowley@postgresql.o 1031 [ + + ]: 49036328 : if (likely(scan->rs_inited))
1032 : : {
1033 : : /* continue from previously returned page/tuple */
1048 1034 : 48071538 : page = BufferGetPage(scan->rs_cbuf);
1035 : :
1036 : 48071538 : lineindex = scan->rs_cindex + dir;
1037 [ + + ]: 48071538 : if (ScanDirectionIsForward(dir))
1038 : 48071210 : linesleft = scan->rs_ntuples - lineindex;
1039 : : else
1040 : 328 : linesleft = scan->rs_cindex;
1041 : : /* lineindex now references the next or previous visible tid */
1042 : :
1044 1043 : 48071538 : goto continue_page;
1044 : : }
1045 : :
1046 : : /*
1047 : : * advance the scan until we find a qualifying tuple or run out of stuff
1048 : : * to scan
1049 : : */
1050 : : while (true)
1051 : : {
622 1052 : 3466511 : heap_fetch_next_buffer(scan, dir);
1053 : :
1054 : : /* did we run out of blocks to scan? */
1055 [ + + ]: 3466486 : if (!BufferIsValid(scan->rs_cbuf))
1056 : 806606 : break;
1057 : :
1058 [ - + ]: 2659880 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
1059 : :
1060 : : /* prune the page and determine visible tuple offsets */
1061 : 2659880 : heap_prepare_pagescan((TableScanDesc) scan);
1044 1062 : 2659872 : page = BufferGetPage(scan->rs_cbuf);
1063 : 2659872 : linesleft = scan->rs_ntuples;
1064 [ + + ]: 2659872 : lineindex = ScanDirectionIsForward(dir) ? 0 : linesleft - 1;
1065 : :
1066 : : /* block is the same for all tuples, set it once outside the loop */
260 heikki.linnakangas@i 1067 : 2659872 : ItemPointerSetBlockNumber(&tuple->t_self, scan->rs_cblock);
1068 : :
1069 : : /* lineindex now references the next or previous visible tid */
1044 drowley@postgresql.o 1070 : 50731410 : continue_page:
1071 : :
1072 [ + + ]: 97837315 : for (; linesleft > 0; linesleft--, lineindex += dir)
1073 : : {
1074 : : ItemId lpp;
1075 : : OffsetNumber lineoff;
1076 : :
364 melanieplageman@gmai 1077 [ - + ]: 95335594 : Assert(lineindex <= scan->rs_ntuples);
7326 tgl@sss.pgh.pa.us 1078 : 95335594 : lineoff = scan->rs_vistuples[lineindex];
1127 peter@eisentraut.org 1079 : 95335594 : lpp = PageGetItemId(page, lineoff);
6671 tgl@sss.pgh.pa.us 1080 [ - + ]: 95335594 : Assert(ItemIdIsNormal(lpp));
1081 : :
1127 peter@eisentraut.org 1082 : 95335594 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
7326 tgl@sss.pgh.pa.us 1083 : 95335594 : tuple->t_len = ItemIdGetLength(lpp);
260 heikki.linnakangas@i 1084 : 95335594 : ItemPointerSetOffsetNumber(&tuple->t_self, lineoff);
1085 : :
1086 : : /* skip any tuples that don't match the scan key */
1044 drowley@postgresql.o 1087 [ + + ]: 95335594 : if (key != NULL &&
1088 [ + + ]: 47428744 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1089 : : nkeys, key))
1090 : 47105905 : continue;
1091 : :
1092 : 48229689 : scan->rs_cindex = lineindex;
1093 : 48229689 : return;
1094 : : }
1095 : : }
1096 : :
1097 : : /* end of scan */
1098 [ - + ]: 806606 : if (BufferIsValid(scan->rs_cbuf))
1044 drowley@postgresql.o 1099 :UBC 0 : ReleaseBuffer(scan->rs_cbuf);
1044 drowley@postgresql.o 1100 :CBC 806606 : scan->rs_cbuf = InvalidBuffer;
1101 : 806606 : scan->rs_cblock = InvalidBlockNumber;
618 tmunro@postgresql.or 1102 : 806606 : scan->rs_prefetch_block = InvalidBlockNumber;
1044 drowley@postgresql.o 1103 : 806606 : tuple->t_data = NULL;
1104 : 806606 : scan->rs_inited = false;
1105 : : }
1106 : :
1107 : :
1108 : : /* ----------------------------------------------------------------
1109 : : * heap access method interface
1110 : : * ----------------------------------------------------------------
1111 : : */
1112 : :
1113 : :
1114 : : TableScanDesc
8612 tgl@sss.pgh.pa.us 1115 : 375830 : heap_beginscan(Relation relation, Snapshot snapshot,
1116 : : int nkeys, ScanKey key,
1117 : : ParallelTableScanDesc parallel_scan,
1118 : : uint32 flags)
1119 : : {
1120 : : HeapScanDesc scan;
1121 : :
1122 : : /*
1123 : : * increment relation ref count while scanning relation
1124 : : *
1125 : : * This is just to make really sure the relcache entry won't go away while
1126 : : * the scan has a pointer to it. Caller should be holding the rel open
1127 : : * anyway, so this is redundant in all normal scenarios...
1128 : : */
9170 1129 : 375830 : RelationIncrementReferenceCount(relation);
1130 : :
1131 : : /*
1132 : : * allocate and initialize scan descriptor
1133 : : */
335 melanieplageman@gmai 1134 [ + + ]: 375830 : if (flags & SO_TYPE_BITMAPSCAN)
1135 : : {
7 michael@paquier.xyz 1136 :GNC 10507 : BitmapHeapScanDesc bscan = palloc_object(BitmapHeapScanDescData);
1137 : :
1138 : : /*
1139 : : * Bitmap Heap scans do not have any fields that a normal Heap Scan
1140 : : * does not have, so no special initializations required here.
1141 : : */
335 melanieplageman@gmai 1142 :CBC 10507 : scan = (HeapScanDesc) bscan;
1143 : : }
1144 : : else
7 michael@paquier.xyz 1145 :GNC 365323 : scan = (HeapScanDesc) palloc_object(HeapScanDescData);
1146 : :
2473 andres@anarazel.de 1147 :CBC 375830 : scan->rs_base.rs_rd = relation;
1148 : 375830 : scan->rs_base.rs_snapshot = snapshot;
1149 : 375830 : scan->rs_base.rs_nkeys = nkeys;
2404 1150 : 375830 : scan->rs_base.rs_flags = flags;
2473 1151 : 375830 : scan->rs_base.rs_parallel = parallel_scan;
2404 1152 : 375830 : scan->rs_strategy = NULL; /* set in initscan */
277 melanieplageman@gmai 1153 : 375830 : scan->rs_cbuf = InvalidBuffer;
1154 : :
1155 : : /*
1156 : : * Disable page-at-a-time mode if it's not a MVCC-safe snapshot.
1157 : : */
2404 andres@anarazel.de 1158 [ + + + + : 375830 : if (!(snapshot && IsMVCCSnapshot(snapshot)))
+ + ]
1159 : 29846 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1160 : :
1161 : : /* Check that a historic snapshot is not used for non-catalog tables */
117 heikki.linnakangas@i 1162 [ + + ]:GNC 375830 : if (snapshot &&
1163 [ + + ]: 366816 : IsHistoricMVCCSnapshot(snapshot) &&
1164 [ + - + - : 644 : !RelationIsAccessibleInLogicalDecoding(relation))
- + - - -
- - + - -
- - - - -
- - - ]
1165 : : {
117 heikki.linnakangas@i 1166 [ # # ]:UNC 0 : ereport(ERROR,
1167 : : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
1168 : : errmsg("cannot query non-catalog table \"%s\" during logical decoding",
1169 : : RelationGetRelationName(relation))));
1170 : : }
1171 : :
1172 : : /*
1173 : : * For seqscan and sample scans in a serializable transaction, acquire a
1174 : : * predicate lock on the entire relation. This is required not only to
1175 : : * lock all the matching tuples, but also to conflict with new insertions
1176 : : * into the table. In an indexscan, we take page locks on the index pages
1177 : : * covering the range specified in the scan qual, but in a heap scan there
1178 : : * is nothing more fine-grained to lock. A bitmap scan is a different
1179 : : * story, there we have already scanned the index and locked the index
1180 : : * pages covering the predicate. But in that case we still have to lock
1181 : : * any matching heap tuples. For sample scan we could optimize the locking
1182 : : * to be at least page-level granularity, but we'd need to add per-tuple
1183 : : * locking for that.
1184 : : */
2404 andres@anarazel.de 1185 [ + + ]:CBC 375830 : if (scan->rs_base.rs_flags & (SO_TYPE_SEQSCAN | SO_TYPE_SAMPLESCAN))
1186 : : {
1187 : : /*
1188 : : * Ensure a missing snapshot is noticed reliably, even if the
1189 : : * isolation mode means predicate locking isn't performed (and
1190 : : * therefore the snapshot isn't used here).
1191 : : */
1192 [ - + ]: 354926 : Assert(snapshot);
5285 heikki.linnakangas@i 1193 : 354926 : PredicateLockRelation(relation, snapshot);
1194 : : }
1195 : :
1196 : : /* we only need to set this up once */
7326 tgl@sss.pgh.pa.us 1197 : 375830 : scan->rs_ctup.t_tableOid = RelationGetRelid(relation);
1198 : :
1199 : : /*
1200 : : * Allocate memory to keep track of page allocation for parallel workers
1201 : : * when doing a parallel scan.
1202 : : */
1723 drowley@postgresql.o 1203 [ + + ]: 375830 : if (parallel_scan != NULL)
7 michael@paquier.xyz 1204 :GNC 2058 : scan->rs_parallelworkerdata = palloc_object(ParallelBlockTableScanWorkerData);
1205 : : else
1723 drowley@postgresql.o 1206 :CBC 373772 : scan->rs_parallelworkerdata = NULL;
1207 : :
1208 : : /*
1209 : : * we do this here instead of in initscan() because heap_rescan also calls
1210 : : * initscan() and we don't want to allocate memory again
1211 : : */
8612 tgl@sss.pgh.pa.us 1212 [ + + ]: 375830 : if (nkeys > 0)
7 michael@paquier.xyz 1213 :GNC 214084 : scan->rs_base.rs_key = palloc_array(ScanKeyData, nkeys);
1214 : : else
2473 andres@anarazel.de 1215 :CBC 161746 : scan->rs_base.rs_key = NULL;
1216 : :
6034 tgl@sss.pgh.pa.us 1217 : 375830 : initscan(scan, key, false);
1218 : :
618 tmunro@postgresql.or 1219 : 375828 : scan->rs_read_stream = NULL;
1220 : :
1221 : : /*
1222 : : * Set up a read stream for sequential scans and TID range scans. This
1223 : : * should be done after initscan() because initscan() allocates the
1224 : : * BufferAccessStrategy object passed to the read stream API.
1225 : : */
1226 [ + + ]: 375828 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN ||
1227 [ + + ]: 20977 : scan->rs_base.rs_flags & SO_TYPE_TIDRANGESCAN)
1228 : 355839 : {
1229 : : ReadStreamBlockNumberCB cb;
1230 : :
1231 [ + + ]: 355839 : if (scan->rs_base.rs_parallel)
1232 : 2058 : cb = heap_scan_stream_read_next_parallel;
1233 : : else
1234 : 353781 : cb = heap_scan_stream_read_next_serial;
1235 : :
1236 : : /* ---
1237 : : * It is safe to use batchmode as the only locks taken by `cb`
1238 : : * are never taken while waiting for IO:
1239 : : * - SyncScanLock is used in the non-parallel case
1240 : : * - in the parallel case, only spinlocks and atomics are used
1241 : : * ---
1242 : : */
262 andres@anarazel.de 1243 : 355839 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_SEQUENTIAL |
1244 : : READ_STREAM_USE_BATCHING,
1245 : : scan->rs_strategy,
1246 : : scan->rs_base.rs_rd,
1247 : : MAIN_FORKNUM,
1248 : : cb,
1249 : : scan,
1250 : : 0);
1251 : : }
277 melanieplageman@gmai 1252 [ + + ]: 19989 : else if (scan->rs_base.rs_flags & SO_TYPE_BITMAPSCAN)
1253 : : {
258 1254 : 10507 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_DEFAULT |
1255 : : READ_STREAM_USE_BATCHING,
1256 : : scan->rs_strategy,
1257 : : scan->rs_base.rs_rd,
1258 : : MAIN_FORKNUM,
1259 : : bitmapheap_stream_read_next,
1260 : : scan,
1261 : : sizeof(TBMIterateResult));
1262 : : }
1263 : :
1264 : :
2473 andres@anarazel.de 1265 : 375828 : return (TableScanDesc) scan;
1266 : : }
1267 : :
1268 : : void
1269 : 632555 : heap_rescan(TableScanDesc sscan, ScanKey key, bool set_params,
1270 : : bool allow_strat, bool allow_sync, bool allow_pagemode)
1271 : : {
1272 : 632555 : HeapScanDesc scan = (HeapScanDesc) sscan;
1273 : :
1274 [ + + ]: 632555 : if (set_params)
1275 : : {
2404 1276 [ + - ]: 15 : if (allow_strat)
1277 : 15 : scan->rs_base.rs_flags |= SO_ALLOW_STRAT;
1278 : : else
2404 andres@anarazel.de 1279 :UBC 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_STRAT;
1280 : :
2404 andres@anarazel.de 1281 [ + + ]:CBC 15 : if (allow_sync)
1282 : 6 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
1283 : : else
1284 : 9 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
1285 : :
1286 [ + - + - ]: 15 : if (allow_pagemode && scan->rs_base.rs_snapshot &&
1287 [ - + - - ]: 15 : IsMVCCSnapshot(scan->rs_base.rs_snapshot))
1288 : 15 : scan->rs_base.rs_flags |= SO_ALLOW_PAGEMODE;
1289 : : else
2404 andres@anarazel.de 1290 :UBC 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1291 : : }
1292 : :
1293 : : /*
1294 : : * unpin scan buffers
1295 : : */
8957 tgl@sss.pgh.pa.us 1296 [ + + ]:CBC 632555 : if (BufferIsValid(scan->rs_cbuf))
1297 : : {
1298 : 1754 : ReleaseBuffer(scan->rs_cbuf);
277 melanieplageman@gmai 1299 : 1754 : scan->rs_cbuf = InvalidBuffer;
1300 : : }
1301 : :
1302 : : /*
1303 : : * SO_TYPE_BITMAPSCAN would be cleaned up here, but it does not hold any
1304 : : * additional data vs a normal HeapScan
1305 : : */
1306 : :
1307 : : /*
1308 : : * The read stream is reset on rescan. This must be done before
1309 : : * initscan(), as some state referred to by read_stream_reset() is reset
1310 : : * in initscan().
1311 : : */
618 tmunro@postgresql.or 1312 [ + + ]: 632555 : if (scan->rs_read_stream)
1313 : 632537 : read_stream_reset(scan->rs_read_stream);
1314 : :
1315 : : /*
1316 : : * reinitialize scan descriptor
1317 : : */
6034 tgl@sss.pgh.pa.us 1318 : 632555 : initscan(scan, key, true);
10753 scrappy@hub.org 1319 : 632555 : }
1320 : :
1321 : : void
2473 andres@anarazel.de 1322 : 373459 : heap_endscan(TableScanDesc sscan)
1323 : : {
1324 : 373459 : HeapScanDesc scan = (HeapScanDesc) sscan;
1325 : :
1326 : : /* Note: no locking manipulations needed */
1327 : :
1328 : : /*
1329 : : * unpin scan buffers
1330 : : */
8957 tgl@sss.pgh.pa.us 1331 [ + + ]: 373459 : if (BufferIsValid(scan->rs_cbuf))
1332 : 154922 : ReleaseBuffer(scan->rs_cbuf);
1333 : :
1334 : : /*
1335 : : * Must free the read stream before freeing the BufferAccessStrategy.
1336 : : */
618 tmunro@postgresql.or 1337 [ + + ]: 373459 : if (scan->rs_read_stream)
1338 : 364031 : read_stream_end(scan->rs_read_stream);
1339 : :
1340 : : /*
1341 : : * decrement relation reference count and free scan descriptor storage
1342 : : */
2473 andres@anarazel.de 1343 : 373459 : RelationDecrementReferenceCount(scan->rs_base.rs_rd);
1344 : :
1345 [ + + ]: 373459 : if (scan->rs_base.rs_key)
1346 : 214053 : pfree(scan->rs_base.rs_key);
1347 : :
6776 tgl@sss.pgh.pa.us 1348 [ + + ]: 373459 : if (scan->rs_strategy != NULL)
1349 : 12796 : FreeAccessStrategy(scan->rs_strategy);
1350 : :
1723 drowley@postgresql.o 1351 [ + + ]: 373459 : if (scan->rs_parallelworkerdata != NULL)
1352 : 2058 : pfree(scan->rs_parallelworkerdata);
1353 : :
2404 andres@anarazel.de 1354 [ + + ]: 373459 : if (scan->rs_base.rs_flags & SO_TEMP_SNAPSHOT)
2473 1355 : 32341 : UnregisterSnapshot(scan->rs_base.rs_snapshot);
1356 : :
9587 tgl@sss.pgh.pa.us 1357 : 373459 : pfree(scan);
10753 scrappy@hub.org 1358 : 373459 : }
1359 : :
1360 : : HeapTuple
2473 andres@anarazel.de 1361 : 9132506 : heap_getnext(TableScanDesc sscan, ScanDirection direction)
1362 : : {
1363 : 9132506 : HeapScanDesc scan = (HeapScanDesc) sscan;
1364 : :
1365 : : /*
1366 : : * This is still widely used directly, without going through table AM, so
1367 : : * add a safety check. It's possible we should, at a later point,
1368 : : * downgrade this to an assert. The reason for checking the AM routine,
1369 : : * rather than the AM oid, is that this allows to write regression tests
1370 : : * that create another AM reusing the heap handler.
1371 : : */
1372 [ - + ]: 9132506 : if (unlikely(sscan->rs_rd->rd_tableam != GetHeapamTableAmRoutine()))
2473 andres@anarazel.de 1373 [ # # ]:UBC 0 : ereport(ERROR,
1374 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1375 : : errmsg_internal("only heap AM is supported")));
1376 : :
1377 : : /*
1378 : : * We don't expect direct calls to heap_getnext with valid CheckXidAlive
1379 : : * for catalog or regular tables. See detailed comments in xact.c where
1380 : : * these variables are declared. Normally we have such a check at tableam
1381 : : * level API but this is called from many places so we need to ensure it
1382 : : * here.
1383 : : */
1957 akapila@postgresql.o 1384 [ - + - - :CBC 9132506 : if (unlikely(TransactionIdIsValid(CheckXidAlive) && !bsysscan))
- + ]
1957 akapila@postgresql.o 1385 [ # # ]:UBC 0 : elog(ERROR, "unexpected heap_getnext call during logical decoding");
1386 : :
1387 : : /* Note: no locking manipulations needed */
1388 : :
2404 andres@anarazel.de 1389 [ + + ]:CBC 9132506 : if (scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
7239 neilc@samurai.com 1390 : 2124510 : heapgettup_pagemode(scan, direction,
2473 andres@anarazel.de 1391 : 2124510 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1392 : : else
1393 : 7007996 : heapgettup(scan, direction,
1394 : 7007996 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1395 : :
7326 tgl@sss.pgh.pa.us 1396 [ + + ]: 9132506 : if (scan->rs_ctup.t_data == NULL)
8612 1397 : 59003 : return NULL;
1398 : :
1399 : : /*
1400 : : * if we get here it means we have a new current scan tuple, so point to
1401 : : * the proper return buffer and return the tuple.
1402 : : */
1403 : :
2473 andres@anarazel.de 1404 [ - + - - : 9073503 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
+ - ]
1405 : :
1406 : 9073503 : return &scan->rs_ctup;
1407 : : }
1408 : :
1409 : : bool
1410 : 47385042 : heap_getnextslot(TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
1411 : : {
1412 : 47385042 : HeapScanDesc scan = (HeapScanDesc) sscan;
1413 : :
1414 : : /* Note: no locking manipulations needed */
1415 : :
2404 1416 [ + + ]: 47385042 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1417 : 46906543 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1418 : : else
1419 : 478499 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1420 : :
2473 1421 [ + + ]: 47385015 : if (scan->rs_ctup.t_data == NULL)
1422 : : {
1423 : 768181 : ExecClearTuple(slot);
1424 : 768181 : return false;
1425 : : }
1426 : :
1427 : : /*
1428 : : * if we get here it means we have a new current scan tuple, so point to
1429 : : * the proper return buffer and return the tuple.
1430 : : */
1431 : :
1432 [ + + - + : 46616834 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
+ + ]
1433 : :
1434 : 46616834 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
1435 : : scan->rs_cbuf);
1436 : 46616834 : return true;
1437 : : }
1438 : :
1439 : : void
1754 drowley@postgresql.o 1440 : 1033 : heap_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
1441 : : ItemPointer maxtid)
1442 : : {
1443 : 1033 : HeapScanDesc scan = (HeapScanDesc) sscan;
1444 : : BlockNumber startBlk;
1445 : : BlockNumber numBlks;
1446 : : ItemPointerData highestItem;
1447 : : ItemPointerData lowestItem;
1448 : :
1449 : : /*
1450 : : * For relations without any pages, we can simply leave the TID range
1451 : : * unset. There will be no tuples to scan, therefore no tuples outside
1452 : : * the given TID range.
1453 : : */
1454 [ + + ]: 1033 : if (scan->rs_nblocks == 0)
1455 : 24 : return;
1456 : :
1457 : : /*
1458 : : * Set up some ItemPointers which point to the first and last possible
1459 : : * tuples in the heap.
1460 : : */
1461 : 1027 : ItemPointerSet(&highestItem, scan->rs_nblocks - 1, MaxOffsetNumber);
1462 : 1027 : ItemPointerSet(&lowestItem, 0, FirstOffsetNumber);
1463 : :
1464 : : /*
1465 : : * If the given maximum TID is below the highest possible TID in the
1466 : : * relation, then restrict the range to that, otherwise we scan to the end
1467 : : * of the relation.
1468 : : */
1469 [ + + ]: 1027 : if (ItemPointerCompare(maxtid, &highestItem) < 0)
1470 : 128 : ItemPointerCopy(maxtid, &highestItem);
1471 : :
1472 : : /*
1473 : : * If the given minimum TID is above the lowest possible TID in the
1474 : : * relation, then restrict the range to only scan for TIDs above that.
1475 : : */
1476 [ + + ]: 1027 : if (ItemPointerCompare(mintid, &lowestItem) > 0)
1477 : 911 : ItemPointerCopy(mintid, &lowestItem);
1478 : :
1479 : : /*
1480 : : * Check for an empty range and protect from would be negative results
1481 : : * from the numBlks calculation below.
1482 : : */
1483 [ + + ]: 1027 : if (ItemPointerCompare(&highestItem, &lowestItem) < 0)
1484 : : {
1485 : : /* Set an empty range of blocks to scan */
1486 : 18 : heap_setscanlimits(sscan, 0, 0);
1487 : 18 : return;
1488 : : }
1489 : :
1490 : : /*
1491 : : * Calculate the first block and the number of blocks we must scan. We
1492 : : * could be more aggressive here and perform some more validation to try
1493 : : * and further narrow the scope of blocks to scan by checking if the
1494 : : * lowestItem has an offset above MaxOffsetNumber. In this case, we could
1495 : : * advance startBlk by one. Likewise, if highestItem has an offset of 0
1496 : : * we could scan one fewer blocks. However, such an optimization does not
1497 : : * seem worth troubling over, currently.
1498 : : */
1499 : 1009 : startBlk = ItemPointerGetBlockNumberNoCheck(&lowestItem);
1500 : :
1501 : 1009 : numBlks = ItemPointerGetBlockNumberNoCheck(&highestItem) -
1502 : 1009 : ItemPointerGetBlockNumberNoCheck(&lowestItem) + 1;
1503 : :
1504 : : /* Set the start block and number of blocks to scan */
1505 : 1009 : heap_setscanlimits(sscan, startBlk, numBlks);
1506 : :
1507 : : /* Finally, set the TID range in sscan */
418 melanieplageman@gmai 1508 : 1009 : ItemPointerCopy(&lowestItem, &sscan->st.tidrange.rs_mintid);
1509 : 1009 : ItemPointerCopy(&highestItem, &sscan->st.tidrange.rs_maxtid);
1510 : : }
1511 : :
1512 : : bool
1754 drowley@postgresql.o 1513 : 5182 : heap_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
1514 : : TupleTableSlot *slot)
1515 : : {
1516 : 5182 : HeapScanDesc scan = (HeapScanDesc) sscan;
418 melanieplageman@gmai 1517 : 5182 : ItemPointer mintid = &sscan->st.tidrange.rs_mintid;
1518 : 5182 : ItemPointer maxtid = &sscan->st.tidrange.rs_maxtid;
1519 : :
1520 : : /* Note: no locking manipulations needed */
1521 : : for (;;)
1522 : : {
1754 drowley@postgresql.o 1523 [ + - ]: 5275 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1524 : 5275 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1525 : : else
1754 drowley@postgresql.o 1526 :UBC 0 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1527 : :
1754 drowley@postgresql.o 1528 [ + + ]:CBC 5269 : if (scan->rs_ctup.t_data == NULL)
1529 : : {
1530 : 104 : ExecClearTuple(slot);
1531 : 104 : return false;
1532 : : }
1533 : :
1534 : : /*
1535 : : * heap_set_tidrange will have used heap_setscanlimits to limit the
1536 : : * range of pages we scan to only ones that can contain the TID range
1537 : : * we're scanning for. Here we must filter out any tuples from these
1538 : : * pages that are outside of that range.
1539 : : */
1540 [ + + ]: 5165 : if (ItemPointerCompare(&scan->rs_ctup.t_self, mintid) < 0)
1541 : : {
1542 : 93 : ExecClearTuple(slot);
1543 : :
1544 : : /*
1545 : : * When scanning backwards, the TIDs will be in descending order.
1546 : : * Future tuples in this direction will be lower still, so we can
1547 : : * just return false to indicate there will be no more tuples.
1548 : : */
1549 [ - + ]: 93 : if (ScanDirectionIsBackward(direction))
1754 drowley@postgresql.o 1550 :UBC 0 : return false;
1551 : :
1754 drowley@postgresql.o 1552 :CBC 93 : continue;
1553 : : }
1554 : :
1555 : : /*
1556 : : * Likewise for the final page, we must filter out TIDs greater than
1557 : : * maxtid.
1558 : : */
1559 [ + + ]: 5072 : if (ItemPointerCompare(&scan->rs_ctup.t_self, maxtid) > 0)
1560 : : {
1561 : 56 : ExecClearTuple(slot);
1562 : :
1563 : : /*
1564 : : * When scanning forward, the TIDs will be in ascending order.
1565 : : * Future tuples in this direction will be higher still, so we can
1566 : : * just return false to indicate there will be no more tuples.
1567 : : */
1568 [ + - ]: 56 : if (ScanDirectionIsForward(direction))
1569 : 56 : return false;
1754 drowley@postgresql.o 1570 :UBC 0 : continue;
1571 : : }
1572 : :
1754 drowley@postgresql.o 1573 :CBC 5016 : break;
1574 : : }
1575 : :
1576 : : /*
1577 : : * if we get here it means we have a new current scan tuple, so point to
1578 : : * the proper return buffer and return the tuple.
1579 : : */
1580 [ - + - - : 5016 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
+ - ]
1581 : :
1582 : 5016 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
1583 : 5016 : return true;
1584 : : }
1585 : :
1586 : : /*
1587 : : * heap_fetch - retrieve tuple with given tid
1588 : : *
1589 : : * On entry, tuple->t_self is the TID to fetch. We pin the buffer holding
1590 : : * the tuple, fill in the remaining fields of *tuple, and check the tuple
1591 : : * against the specified snapshot.
1592 : : *
1593 : : * If successful (tuple found and passes snapshot time qual), then *userbuf
1594 : : * is set to the buffer holding the tuple and true is returned. The caller
1595 : : * must unpin the buffer when done with the tuple.
1596 : : *
1597 : : * If the tuple is not found (ie, item number references a deleted slot),
1598 : : * then tuple->t_data is set to NULL, *userbuf is set to InvalidBuffer,
1599 : : * and false is returned.
1600 : : *
1601 : : * If the tuple is found but fails the time qual check, then the behavior
1602 : : * depends on the keep_buf parameter. If keep_buf is false, the results
1603 : : * are the same as for the tuple-not-found case. If keep_buf is true,
1604 : : * then tuple->t_data and *userbuf are returned as for the success case,
1605 : : * and again the caller must unpin the buffer; but false is returned.
1606 : : *
1607 : : * heap_fetch does not follow HOT chains: only the exact TID requested will
1608 : : * be fetched.
1609 : : *
1610 : : * It is somewhat inconsistent that we ereport() on invalid block number but
1611 : : * return false on invalid item number. There are a couple of reasons though.
1612 : : * One is that the caller can relatively easily check the block number for
1613 : : * validity, but cannot check the item number without reading the page
1614 : : * himself. Another is that when we are following a t_ctid link, we can be
1615 : : * reasonably confident that the page number is valid (since VACUUM shouldn't
1616 : : * truncate off the destination page without having killed the referencing
1617 : : * tuple first), but the item number might well not be good.
1618 : : */
1619 : : bool
10753 scrappy@hub.org 1620 : 181478 : heap_fetch(Relation relation,
1621 : : Snapshot snapshot,
1622 : : HeapTuple tuple,
1623 : : Buffer *userbuf,
1624 : : bool keep_buf)
1625 : : {
8608 tgl@sss.pgh.pa.us 1626 : 181478 : ItemPointer tid = &(tuple->t_self);
1627 : : ItemId lp;
1628 : : Buffer buffer;
1629 : : Page page;
1630 : : OffsetNumber offnum;
1631 : : bool valid;
1632 : :
1633 : : /*
1634 : : * Fetch and pin the appropriate page of the relation.
1635 : : */
6467 1636 : 181478 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
1637 : :
1638 : : /*
1639 : : * Need share lock on buffer to examine tuple commit status.
1640 : : */
9864 vadim4o@yahoo.com 1641 : 181469 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
3528 kgrittn@postgresql.o 1642 : 181469 : page = BufferGetPage(buffer);
1643 : :
1644 : : /*
1645 : : * We'd better check for out-of-range offnum in case of VACUUM since the
1646 : : * TID was obtained.
1647 : : */
10328 bruce@momjian.us 1648 : 181469 : offnum = ItemPointerGetOffsetNumber(tid);
6366 tgl@sss.pgh.pa.us 1649 [ + - + + ]: 181469 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1650 : : {
7570 1651 : 3 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2461 andres@anarazel.de 1652 : 3 : ReleaseBuffer(buffer);
1653 : 3 : *userbuf = InvalidBuffer;
7570 tgl@sss.pgh.pa.us 1654 : 3 : tuple->t_data = NULL;
1655 : 3 : return false;
1656 : : }
1657 : :
1658 : : /*
1659 : : * get the item line pointer corresponding to the requested tid
1660 : : */
6366 1661 : 181466 : lp = PageGetItemId(page, offnum);
1662 : :
1663 : : /*
1664 : : * Must check for deleted tuple.
1665 : : */
6671 1666 [ + + ]: 181466 : if (!ItemIdIsNormal(lp))
1667 : : {
9297 vadim4o@yahoo.com 1668 : 354 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2461 andres@anarazel.de 1669 : 354 : ReleaseBuffer(buffer);
1670 : 354 : *userbuf = InvalidBuffer;
8608 tgl@sss.pgh.pa.us 1671 : 354 : tuple->t_data = NULL;
1672 : 354 : return false;
1673 : : }
1674 : :
1675 : : /*
1676 : : * fill in *tuple fields
1677 : : */
6366 1678 : 181112 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
9882 vadim4o@yahoo.com 1679 : 181112 : tuple->t_len = ItemIdGetLength(lp);
7424 tgl@sss.pgh.pa.us 1680 : 181112 : tuple->t_tableOid = RelationGetRelid(relation);
1681 : :
1682 : : /*
1683 : : * check tuple visibility, then release lock
1684 : : */
7326 1685 : 181112 : valid = HeapTupleSatisfiesVisibility(tuple, snapshot, buffer);
1686 : :
5427 heikki.linnakangas@i 1687 [ + + ]: 181112 : if (valid)
2150 tmunro@postgresql.or 1688 : 181057 : PredicateLockTID(relation, &(tuple->t_self), snapshot,
1689 : 181057 : HeapTupleHeaderGetXmin(tuple->t_data));
1690 : :
1691 : 181112 : HeapCheckForSerializableConflictOut(valid, relation, tuple, buffer, snapshot);
1692 : :
5402 heikki.linnakangas@i 1693 : 181112 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1694 : :
8608 tgl@sss.pgh.pa.us 1695 [ + + ]: 181112 : if (valid)
1696 : : {
1697 : : /*
1698 : : * All checks passed, so return the tuple as valid. Caller is now
1699 : : * responsible for releasing the buffer.
1700 : : */
9581 1701 : 181057 : *userbuf = buffer;
1702 : :
8608 1703 : 181057 : return true;
1704 : : }
1705 : :
1706 : : /* Tuple failed time qual, but maybe caller wants to see it anyway. */
1344 1707 [ + + ]: 55 : if (keep_buf)
1708 : 34 : *userbuf = buffer;
1709 : : else
1710 : : {
1711 : 21 : ReleaseBuffer(buffer);
1712 : 21 : *userbuf = InvalidBuffer;
1713 : 21 : tuple->t_data = NULL;
1714 : : }
1715 : :
8608 1716 : 55 : return false;
1717 : : }
1718 : :
1719 : : /*
1720 : : * heap_hot_search_buffer - search HOT chain for tuple satisfying snapshot
1721 : : *
1722 : : * On entry, *tid is the TID of a tuple (either a simple tuple, or the root
1723 : : * of a HOT chain), and buffer is the buffer holding this tuple. We search
1724 : : * for the first chain member satisfying the given snapshot. If one is
1725 : : * found, we update *tid to reference that tuple's offset number, and
1726 : : * return true. If no match, return false without modifying *tid.
1727 : : *
1728 : : * heapTuple is a caller-supplied buffer. When a match is found, we return
1729 : : * the tuple here, in addition to updating *tid. If no match is found, the
1730 : : * contents of this buffer on return are undefined.
1731 : : *
1732 : : * If all_dead is not NULL, we check non-visible tuples to see if they are
1733 : : * globally dead; *all_dead is set true if all members of the HOT chain
1734 : : * are vacuumable, false if not.
1735 : : *
1736 : : * Unlike heap_fetch, the caller must already have pin and (at least) share
1737 : : * lock on the buffer; it is still pinned/locked at exit.
1738 : : */
1739 : : bool
5427 heikki.linnakangas@i 1740 : 21490585 : heap_hot_search_buffer(ItemPointer tid, Relation relation, Buffer buffer,
1741 : : Snapshot snapshot, HeapTuple heapTuple,
1742 : : bool *all_dead, bool first_call)
1743 : : {
1127 peter@eisentraut.org 1744 : 21490585 : Page page = BufferGetPage(buffer);
6663 tgl@sss.pgh.pa.us 1745 : 21490585 : TransactionId prev_xmax = InvalidTransactionId;
1746 : : BlockNumber blkno;
1747 : : OffsetNumber offnum;
1748 : : bool at_chain_start;
1749 : : bool valid;
1750 : : bool skip;
1953 andres@anarazel.de 1751 : 21490585 : GlobalVisState *vistest = NULL;
1752 : :
1753 : : /* If this is not the first call, previous call returned a (live!) tuple */
6663 tgl@sss.pgh.pa.us 1754 [ + + ]: 21490585 : if (all_dead)
5287 rhaas@postgresql.org 1755 : 18197827 : *all_dead = first_call;
1756 : :
2324 heikki.linnakangas@i 1757 : 21490585 : blkno = ItemPointerGetBlockNumber(tid);
6663 tgl@sss.pgh.pa.us 1758 : 21490585 : offnum = ItemPointerGetOffsetNumber(tid);
5287 rhaas@postgresql.org 1759 : 21490585 : at_chain_start = first_call;
1760 : 21490585 : skip = !first_call;
1761 : :
1762 : : /* XXX: we should assert that a snapshot is pushed or registered */
1953 andres@anarazel.de 1763 [ - + ]: 21490585 : Assert(TransactionIdIsValid(RecentXmin));
2324 heikki.linnakangas@i 1764 [ + - ]: 21490585 : Assert(BufferGetBlockNumber(buffer) == blkno);
1765 : :
1766 : : /* Scan through possible multiple members of HOT-chain */
1767 : : for (;;)
6663 tgl@sss.pgh.pa.us 1768 : 1621481 : {
1769 : : ItemId lp;
1770 : :
1771 : : /* check for bogus TID */
1127 peter@eisentraut.org 1772 [ + - + - ]: 23112066 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1773 : : break;
1774 : :
1775 : 23112066 : lp = PageGetItemId(page, offnum);
1776 : :
1777 : : /* check for unused, dead, or redirected items */
6663 tgl@sss.pgh.pa.us 1778 [ + + ]: 23112066 : if (!ItemIdIsNormal(lp))
1779 : : {
1780 : : /* We should only see a redirect at start of chain */
1781 [ + + + - ]: 779363 : if (ItemIdIsRedirected(lp) && at_chain_start)
1782 : : {
1783 : : /* Follow the redirect */
1784 : 422418 : offnum = ItemIdGetRedirect(lp);
1785 : 422418 : at_chain_start = false;
1786 : 422418 : continue;
1787 : : }
1788 : : /* else must be end of chain */
1789 : 356945 : break;
1790 : : }
1791 : :
1792 : : /*
1793 : : * Update heapTuple to point to the element of the HOT chain we're
1794 : : * currently investigating. Having t_self set correctly is important
1795 : : * because the SSI checks and the *Satisfies routine for historical
1796 : : * MVCC snapshots need the correct tid to decide about the visibility.
1797 : : */
1127 peter@eisentraut.org 1798 : 22332703 : heapTuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
5287 rhaas@postgresql.org 1799 : 22332703 : heapTuple->t_len = ItemIdGetLength(lp);
4531 1800 : 22332703 : heapTuple->t_tableOid = RelationGetRelid(relation);
2324 heikki.linnakangas@i 1801 : 22332703 : ItemPointerSet(&heapTuple->t_self, blkno, offnum);
1802 : :
1803 : : /*
1804 : : * Shouldn't see a HEAP_ONLY tuple at chain start.
1805 : : */
5287 rhaas@postgresql.org 1806 [ + + - + ]: 22332703 : if (at_chain_start && HeapTupleIsHeapOnly(heapTuple))
6663 tgl@sss.pgh.pa.us 1807 :UBC 0 : break;
1808 : :
1809 : : /*
1810 : : * The xmin should match the previous xmax value, else chain is
1811 : : * broken.
1812 : : */
6663 tgl@sss.pgh.pa.us 1813 [ + + - + ]:CBC 23531766 : if (TransactionIdIsValid(prev_xmax) &&
2967 alvherre@alvh.no-ip. 1814 : 1199063 : !TransactionIdEquals(prev_xmax,
1815 : : HeapTupleHeaderGetXmin(heapTuple->t_data)))
6663 tgl@sss.pgh.pa.us 1816 :UBC 0 : break;
1817 : :
1818 : : /*
1819 : : * When first_call is true (and thus, skip is initially false) we'll
1820 : : * return the first tuple we find. But on later passes, heapTuple
1821 : : * will initially be pointing to the tuple we returned last time.
1822 : : * Returning it again would be incorrect (and would loop forever), so
1823 : : * we skip it and return the next match we find.
1824 : : */
5287 rhaas@postgresql.org 1825 [ + + ]:CBC 22332703 : if (!skip)
1826 : : {
1827 : : /* If it's visible per the snapshot, we must return it */
1828 : 22246258 : valid = HeapTupleSatisfiesVisibility(heapTuple, snapshot, buffer);
2150 tmunro@postgresql.or 1829 : 22246258 : HeapCheckForSerializableConflictOut(valid, relation, heapTuple,
1830 : : buffer, snapshot);
1831 : :
5287 rhaas@postgresql.org 1832 [ + + ]: 22246253 : if (valid)
1833 : : {
1834 : 14899736 : ItemPointerSetOffsetNumber(tid, offnum);
2150 tmunro@postgresql.or 1835 : 14899736 : PredicateLockTID(relation, &heapTuple->t_self, snapshot,
1836 : 14899736 : HeapTupleHeaderGetXmin(heapTuple->t_data));
5287 rhaas@postgresql.org 1837 [ + + ]: 14899736 : if (all_dead)
1838 : 11887521 : *all_dead = false;
1839 : 14899736 : return true;
1840 : : }
1841 : : }
1842 : 7432962 : skip = false;
1843 : :
1844 : : /*
1845 : : * If we can't see it, maybe no one else can either. At caller
1846 : : * request, check whether all chain members are dead to all
1847 : : * transactions.
1848 : : *
1849 : : * Note: if you change the criterion here for what is "dead", fix the
1850 : : * planner's get_actual_variable_range() function to match.
1851 : : */
1953 andres@anarazel.de 1852 [ + + + + ]: 7432962 : if (all_dead && *all_dead)
1853 : : {
1854 [ + + ]: 6477122 : if (!vistest)
1855 : 6358394 : vistest = GlobalVisTestFor(relation);
1856 : :
1857 [ + + ]: 6477122 : if (!HeapTupleIsSurelyDead(heapTuple, vistest))
1858 : 6121855 : *all_dead = false;
1859 : : }
1860 : :
1861 : : /*
1862 : : * Check to see if HOT chain continues past this tuple; if so fetch
1863 : : * the next offnum and loop around.
1864 : : */
5287 rhaas@postgresql.org 1865 [ + + ]: 7432962 : if (HeapTupleIsHotUpdated(heapTuple))
1866 : : {
1867 [ - + ]: 1199063 : Assert(ItemPointerGetBlockNumber(&heapTuple->t_data->t_ctid) ==
1868 : : blkno);
1869 : 1199063 : offnum = ItemPointerGetOffsetNumber(&heapTuple->t_data->t_ctid);
6663 tgl@sss.pgh.pa.us 1870 : 1199063 : at_chain_start = false;
4711 alvherre@alvh.no-ip. 1871 : 1199063 : prev_xmax = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
1872 : : }
1873 : : else
6607 bruce@momjian.us 1874 : 6233899 : break; /* end of chain */
1875 : : }
1876 : :
5315 heikki.linnakangas@i 1877 : 6590844 : return false;
1878 : : }
1879 : :
1880 : : /*
1881 : : * heap_get_latest_tid - get the latest tid of a specified tuple
1882 : : *
1883 : : * Actually, this gets the latest version that is visible according to the
1884 : : * scan's snapshot. Create a scan using SnapshotDirty to get the very latest,
1885 : : * possibly uncommitted version.
1886 : : *
1887 : : * *tid is both an input and an output parameter: it is updated to
1888 : : * show the latest version of the row. Note that it will not be changed
1889 : : * if no version of the row passes the snapshot test.
1890 : : */
1891 : : void
2406 andres@anarazel.de 1892 : 150 : heap_get_latest_tid(TableScanDesc sscan,
1893 : : ItemPointer tid)
1894 : : {
2401 tgl@sss.pgh.pa.us 1895 : 150 : Relation relation = sscan->rs_rd;
1896 : 150 : Snapshot snapshot = sscan->rs_snapshot;
1897 : : ItemPointerData ctid;
1898 : : TransactionId priorXmax;
1899 : :
1900 : : /*
1901 : : * table_tuple_get_latest_tid() verified that the passed in tid is valid.
1902 : : * Assume that t_ctid links are valid however - there shouldn't be invalid
1903 : : * ones in the table.
1904 : : */
2406 andres@anarazel.de 1905 [ - + ]: 150 : Assert(ItemPointerIsValid(tid));
1906 : :
1907 : : /*
1908 : : * Loop to chase down t_ctid links. At top of loop, ctid is the tuple we
1909 : : * need to examine, and *tid is the TID we will return if ctid turns out
1910 : : * to be bogus.
1911 : : *
1912 : : * Note that we will loop until we reach the end of the t_ctid chain.
1913 : : * Depending on the snapshot passed, there might be at most one visible
1914 : : * version of the row, but we don't try to optimize for that.
1915 : : */
7424 tgl@sss.pgh.pa.us 1916 : 150 : ctid = *tid;
1917 : 150 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
1918 : : for (;;)
1919 : 45 : {
1920 : : Buffer buffer;
1921 : : Page page;
1922 : : OffsetNumber offnum;
1923 : : ItemId lp;
1924 : : HeapTupleData tp;
1925 : : bool valid;
1926 : :
1927 : : /*
1928 : : * Read, pin, and lock the page.
1929 : : */
1930 : 195 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&ctid));
1931 : 195 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
3528 kgrittn@postgresql.o 1932 : 195 : page = BufferGetPage(buffer);
1933 : :
1934 : : /*
1935 : : * Check for bogus item number. This is not treated as an error
1936 : : * condition because it can happen while following a t_ctid link. We
1937 : : * just assume that the prior tid is OK and return it unchanged.
1938 : : */
7424 tgl@sss.pgh.pa.us 1939 : 195 : offnum = ItemPointerGetOffsetNumber(&ctid);
6366 1940 [ + - - + ]: 195 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1941 : : {
7201 tgl@sss.pgh.pa.us 1942 :UBC 0 : UnlockReleaseBuffer(buffer);
7424 1943 : 0 : break;
1944 : : }
6366 tgl@sss.pgh.pa.us 1945 :CBC 195 : lp = PageGetItemId(page, offnum);
6671 1946 [ - + ]: 195 : if (!ItemIdIsNormal(lp))
1947 : : {
7201 tgl@sss.pgh.pa.us 1948 :UBC 0 : UnlockReleaseBuffer(buffer);
7424 1949 : 0 : break;
1950 : : }
1951 : :
1952 : : /* OK to access the tuple */
7424 tgl@sss.pgh.pa.us 1953 :CBC 195 : tp.t_self = ctid;
6366 1954 : 195 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
7424 1955 : 195 : tp.t_len = ItemIdGetLength(lp);
4531 rhaas@postgresql.org 1956 : 195 : tp.t_tableOid = RelationGetRelid(relation);
1957 : :
1958 : : /*
1959 : : * After following a t_ctid link, we might arrive at an unrelated
1960 : : * tuple. Check for XMIN match.
1961 : : */
7424 tgl@sss.pgh.pa.us 1962 [ + + - + ]: 240 : if (TransactionIdIsValid(priorXmax) &&
2967 alvherre@alvh.no-ip. 1963 : 45 : !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(tp.t_data)))
1964 : : {
7201 tgl@sss.pgh.pa.us 1965 :UBC 0 : UnlockReleaseBuffer(buffer);
7424 1966 : 0 : break;
1967 : : }
1968 : :
1969 : : /*
1970 : : * Check tuple visibility; if visible, set it as the new result
1971 : : * candidate.
1972 : : */
7326 tgl@sss.pgh.pa.us 1973 :CBC 195 : valid = HeapTupleSatisfiesVisibility(&tp, snapshot, buffer);
2150 tmunro@postgresql.or 1974 : 195 : HeapCheckForSerializableConflictOut(valid, relation, &tp, buffer, snapshot);
7424 tgl@sss.pgh.pa.us 1975 [ + + ]: 195 : if (valid)
1976 : 138 : *tid = ctid;
1977 : :
1978 : : /*
1979 : : * If there's a valid t_ctid link, follow it, else we're done.
1980 : : */
4711 alvherre@alvh.no-ip. 1981 [ + + + + ]: 276 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
1982 [ + - ]: 138 : HeapTupleHeaderIsOnlyLocked(tp.t_data) ||
2811 andres@anarazel.de 1983 [ + + ]: 114 : HeapTupleHeaderIndicatesMovedPartitions(tp.t_data) ||
7424 tgl@sss.pgh.pa.us 1984 : 57 : ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
1985 : : {
7201 1986 : 150 : UnlockReleaseBuffer(buffer);
7424 1987 : 150 : break;
1988 : : }
1989 : :
1990 : 45 : ctid = tp.t_data->t_ctid;
4711 alvherre@alvh.no-ip. 1991 : 45 : priorXmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
7201 tgl@sss.pgh.pa.us 1992 : 45 : UnlockReleaseBuffer(buffer);
1993 : : } /* end of loop */
9564 inoue@tpf.co.jp 1994 : 150 : }
1995 : :
1996 : :
1997 : : /*
1998 : : * UpdateXmaxHintBits - update tuple hint bits after xmax transaction ends
1999 : : *
2000 : : * This is called after we have waited for the XMAX transaction to terminate.
2001 : : * If the transaction aborted, we guarantee the XMAX_INVALID hint bit will
2002 : : * be set on exit. If the transaction committed, we set the XMAX_COMMITTED
2003 : : * hint bit if possible --- but beware that that may not yet be possible,
2004 : : * if the transaction committed asynchronously.
2005 : : *
2006 : : * Note that if the transaction was a locker only, we set HEAP_XMAX_INVALID
2007 : : * even if it commits.
2008 : : *
2009 : : * Hence callers should look only at XMAX_INVALID.
2010 : : *
2011 : : * Note this is not allowed for tuples whose xmax is a multixact.
2012 : : */
2013 : : static void
6700 tgl@sss.pgh.pa.us 2014 : 221 : UpdateXmaxHintBits(HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
2015 : : {
4711 alvherre@alvh.no-ip. 2016 [ - + ]: 221 : Assert(TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple), xid));
2017 [ - + ]: 221 : Assert(!(tuple->t_infomask & HEAP_XMAX_IS_MULTI));
2018 : :
6700 tgl@sss.pgh.pa.us 2019 [ + - ]: 221 : if (!(tuple->t_infomask & (HEAP_XMAX_COMMITTED | HEAP_XMAX_INVALID)))
2020 : : {
4711 alvherre@alvh.no-ip. 2021 [ + + + + ]: 397 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask) &&
2022 : 176 : TransactionIdDidCommit(xid))
6700 tgl@sss.pgh.pa.us 2023 : 149 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_COMMITTED,
2024 : : xid);
2025 : : else
2026 : 72 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_INVALID,
2027 : : InvalidTransactionId);
2028 : : }
2029 : 221 : }
2030 : :
2031 : :
2032 : : /*
2033 : : * GetBulkInsertState - prepare status object for a bulk insert
2034 : : */
2035 : : BulkInsertState
6250 2036 : 2723 : GetBulkInsertState(void)
2037 : : {
2038 : : BulkInsertState bistate;
2039 : :
7 michael@paquier.xyz 2040 :GNC 2723 : bistate = (BulkInsertState) palloc_object(BulkInsertStateData);
6250 tgl@sss.pgh.pa.us 2041 :CBC 2723 : bistate->strategy = GetAccessStrategy(BAS_BULKWRITE);
2042 : 2723 : bistate->current_buf = InvalidBuffer;
986 andres@anarazel.de 2043 : 2723 : bistate->next_free = InvalidBlockNumber;
2044 : 2723 : bistate->last_free = InvalidBlockNumber;
856 2045 : 2723 : bistate->already_extended_by = 0;
6250 tgl@sss.pgh.pa.us 2046 : 2723 : return bistate;
2047 : : }
2048 : :
2049 : : /*
2050 : : * FreeBulkInsertState - clean up after finishing a bulk insert
2051 : : */
2052 : : void
2053 : 2578 : FreeBulkInsertState(BulkInsertState bistate)
2054 : : {
2055 [ + + ]: 2578 : if (bistate->current_buf != InvalidBuffer)
6033 bruce@momjian.us 2056 : 1999 : ReleaseBuffer(bistate->current_buf);
6250 tgl@sss.pgh.pa.us 2057 : 2578 : FreeAccessStrategy(bistate->strategy);
2058 : 2578 : pfree(bistate);
2059 : 2578 : }
2060 : :
2061 : : /*
2062 : : * ReleaseBulkInsertStatePin - release a buffer currently held in bistate
2063 : : */
2064 : : void
3249 rhaas@postgresql.org 2065 : 80758 : ReleaseBulkInsertStatePin(BulkInsertState bistate)
2066 : : {
2067 [ + + ]: 80758 : if (bistate->current_buf != InvalidBuffer)
2068 : 30021 : ReleaseBuffer(bistate->current_buf);
2069 : 80758 : bistate->current_buf = InvalidBuffer;
2070 : :
2071 : : /*
2072 : : * Despite the name, we also reset bulk relation extension state.
2073 : : * Otherwise we can end up erroring out due to looking for free space in
2074 : : * ->next_free of one partition, even though ->next_free was set when
2075 : : * extending another partition. It could obviously also be bad for
2076 : : * efficiency to look at existing blocks at offsets from another
2077 : : * partition, even if we don't error out.
2078 : : */
796 andres@anarazel.de 2079 : 80758 : bistate->next_free = InvalidBlockNumber;
2080 : 80758 : bistate->last_free = InvalidBlockNumber;
3249 rhaas@postgresql.org 2081 : 80758 : }
2082 : :
2083 : :
2084 : : /*
2085 : : * heap_insert - insert tuple into a heap
2086 : : *
2087 : : * The new tuple is stamped with current transaction ID and the specified
2088 : : * command ID.
2089 : : *
2090 : : * See table_tuple_insert for comments about most of the input flags, except
2091 : : * that this routine directly takes a tuple rather than a slot.
2092 : : *
2093 : : * There's corresponding HEAP_INSERT_ options to all the TABLE_INSERT_
2094 : : * options, and there additionally is HEAP_INSERT_SPECULATIVE which is used to
2095 : : * implement table_tuple_insert_speculative().
2096 : : *
2097 : : * On return the header fields of *tup are updated to match the stored tuple;
2098 : : * in particular tup->t_self receives the actual TID where the tuple was
2099 : : * stored. But note that any toasting of fields within the tuple data is NOT
2100 : : * reflected into *tup.
2101 : : */
2102 : : void
7485 tgl@sss.pgh.pa.us 2103 : 8451135 : heap_insert(Relation relation, HeapTuple tup, CommandId cid,
2104 : : int options, BulkInsertState bistate)
2105 : : {
7762 2106 : 8451135 : TransactionId xid = GetCurrentTransactionId();
2107 : : HeapTuple heaptup;
2108 : : Buffer buffer;
5293 rhaas@postgresql.org 2109 : 8451135 : Buffer vmbuffer = InvalidBuffer;
6223 heikki.linnakangas@i 2110 : 8451135 : bool all_visible_cleared = false;
2111 : :
2112 : : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
1682 tgl@sss.pgh.pa.us 2113 [ - + ]: 8451135 : Assert(HeapTupleHeaderGetNatts(tup->t_data) <=
2114 : : RelationGetNumberOfAttributes(relation));
2115 : :
201 nathan@postgresql.or 2116 : 8451135 : AssertHasSnapshotForToast(relation);
2117 : :
2118 : : /*
2119 : : * Fill in tuple header fields and toast the tuple if necessary.
2120 : : *
2121 : : * Note: below this point, heaptup is the data we actually intend to store
2122 : : * into the relation; tup is the caller's original untoasted data.
2123 : : */
5152 heikki.linnakangas@i 2124 : 8451135 : heaptup = heap_prepare_insert(relation, tup, xid, cid, options);
2125 : :
2126 : : /*
2127 : : * Find buffer to insert this tuple into. If the page is all visible,
2128 : : * this will also pin the requisite visibility map page.
2129 : : */
3700 kgrittn@postgresql.o 2130 : 8451135 : buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
2131 : : InvalidBuffer, options, bistate,
2132 : : &vmbuffer, NULL,
2133 : : 0);
2134 : :
2135 : : /*
2136 : : * We're about to do the actual insert -- but check for conflict first, to
2137 : : * avoid possibly having to roll back work we've just done.
2138 : : *
2139 : : * This is safe without a recheck as long as there is no possibility of
2140 : : * another process scanning the page between this check and the insert
2141 : : * being visible to the scan (i.e., an exclusive buffer content lock is
2142 : : * continuously held from this point until the tuple insert is visible).
2143 : : *
2144 : : * For a heap insert, we only need to check for table-level SSI locks. Our
2145 : : * new tuple can't possibly conflict with existing tuple locks, and heap
2146 : : * page locks are only consolidated versions of tuple locks; they do not
2147 : : * lock "gaps" as index page locks do. So we don't need to specify a
2148 : : * buffer when making the call, which makes for a faster check.
2149 : : */
2150 tmunro@postgresql.or 2150 : 8451135 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2151 : :
2152 : : /* NO EREPORT(ERROR) from here till changes are logged */
9105 tgl@sss.pgh.pa.us 2153 : 8451123 : START_CRIT_SECTION();
2154 : :
3876 andres@anarazel.de 2155 : 8451123 : RelationPutHeapTuple(relation, buffer, heaptup,
2156 : 8451123 : (options & HEAP_INSERT_SPECULATIVE) != 0);
2157 : :
1658 tomas.vondra@postgre 2158 [ + + ]: 8451123 : if (PageIsAllVisible(BufferGetPage(buffer)))
2159 : : {
6223 heikki.linnakangas@i 2160 : 7616 : all_visible_cleared = true;
3528 kgrittn@postgresql.o 2161 : 7616 : PageClearAllVisible(BufferGetPage(buffer));
5293 rhaas@postgresql.org 2162 : 7616 : visibilitymap_clear(relation,
2163 : 7616 : ItemPointerGetBlockNumber(&(heaptup->t_self)),
2164 : : vmbuffer, VISIBILITYMAP_VALID_BITS);
2165 : : }
2166 : :
2167 : : /*
2168 : : * XXX Should we set PageSetPrunable on this page ?
2169 : : *
2170 : : * The inserting transaction may eventually abort thus making this tuple
2171 : : * DEAD and hence available for pruning. Though we don't want to optimize
2172 : : * for aborts, if no other tuple in this page is UPDATEd/DELETEd, the
2173 : : * aborted tuple will never be pruned until next vacuum is triggered.
2174 : : *
2175 : : * If you do add PageSetPrunable here, add it in heap_xlog_insert too.
2176 : : */
2177 : :
7201 tgl@sss.pgh.pa.us 2178 : 8451123 : MarkBufferDirty(buffer);
2179 : :
2180 : : /* XLOG stuff */
2083 noah@leadboat.com 2181 [ + + + + : 8451123 : if (RelationNeedsWAL(relation))
+ + + + ]
2182 : : {
2183 : : xl_heap_insert xlrec;
2184 : : xl_heap_header xlhdr;
2185 : : XLogRecPtr recptr;
3528 kgrittn@postgresql.o 2186 : 7099680 : Page page = BufferGetPage(buffer);
9036 bruce@momjian.us 2187 : 7099680 : uint8 info = XLOG_HEAP_INSERT;
4045 heikki.linnakangas@i 2188 : 7099680 : int bufflags = 0;
2189 : :
2190 : : /*
2191 : : * If this is a catalog, we need to transmit combo CIDs to properly
2192 : : * decode, so log that as well.
2193 : : */
4390 rhaas@postgresql.org 2194 [ + + + - : 7099680 : if (RelationIsAccessibleInLogicalDecoding(relation))
- + - - -
- + + + +
- + - - +
+ ]
2195 : 3305 : log_heap_new_cid(relation, heaptup);
2196 : :
2197 : : /*
2198 : : * If this is the single and first tuple on page, we can reinit the
2199 : : * page instead of restoring the whole thing. Set flag, and hide
2200 : : * buffer references from XLogInsert.
2201 : : */
4045 heikki.linnakangas@i 2202 [ + + + + ]: 7190286 : if (ItemPointerGetOffsetNumber(&(heaptup->t_self)) == FirstOffsetNumber &&
2203 : 90606 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
2204 : : {
2205 : 89484 : info |= XLOG_HEAP_INIT_PAGE;
2206 : 89484 : bufflags |= REGBUF_WILL_INIT;
2207 : : }
2208 : :
2209 : 7099680 : xlrec.offnum = ItemPointerGetOffsetNumber(&heaptup->t_self);
3876 andres@anarazel.de 2210 : 7099680 : xlrec.flags = 0;
2211 [ + + ]: 7099680 : if (all_visible_cleared)
2212 : 7613 : xlrec.flags |= XLH_INSERT_ALL_VISIBLE_CLEARED;
2213 [ + + ]: 7099680 : if (options & HEAP_INSERT_SPECULATIVE)
2214 : 2073 : xlrec.flags |= XLH_INSERT_IS_SPECULATIVE;
4045 heikki.linnakangas@i 2215 [ - + ]: 7099680 : Assert(ItemPointerGetBlockNumber(&heaptup->t_self) == BufferGetBlockNumber(buffer));
2216 : :
2217 : : /*
2218 : : * For logical decoding, we need the tuple even if we're doing a full
2219 : : * page write, so make sure it's included even if we take a full-page
2220 : : * image. (XXX We could alternatively store a pointer into the FPW).
2221 : : */
2625 andres@anarazel.de 2222 [ + + + - : 7099680 : if (RelationIsLogicallyLogged(relation) &&
- + - - -
- + - +
+ ]
2223 [ + + ]: 250423 : !(options & HEAP_INSERT_NO_LOGICAL))
2224 : : {
3876 2225 : 250396 : xlrec.flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
4045 heikki.linnakangas@i 2226 : 250396 : bufflags |= REGBUF_KEEP_DATA;
2227 : :
1957 akapila@postgresql.o 2228 [ + + ]: 250396 : if (IsToastRelation(relation))
2229 : 1786 : xlrec.flags |= XLH_INSERT_ON_TOAST_RELATION;
2230 : : }
2231 : :
4045 heikki.linnakangas@i 2232 : 7099680 : XLogBeginInsert();
309 peter@eisentraut.org 2233 : 7099680 : XLogRegisterData(&xlrec, SizeOfHeapInsert);
2234 : :
4045 heikki.linnakangas@i 2235 : 7099680 : xlhdr.t_infomask2 = heaptup->t_data->t_infomask2;
2236 : 7099680 : xlhdr.t_infomask = heaptup->t_data->t_infomask;
2237 : 7099680 : xlhdr.t_hoff = heaptup->t_data->t_hoff;
2238 : :
2239 : : /*
2240 : : * note we mark xlhdr as belonging to buffer; if XLogInsert decides to
2241 : : * write the whole page to the xlog, we don't need to store
2242 : : * xl_heap_header in the xlog.
2243 : : */
2244 : 7099680 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
309 peter@eisentraut.org 2245 : 7099680 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
2246 : : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
4045 heikki.linnakangas@i 2247 : 7099680 : XLogRegisterBufData(0,
3952 tgl@sss.pgh.pa.us 2248 : 7099680 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2249 : 7099680 : heaptup->t_len - SizeofHeapTupleHeader);
2250 : :
2251 : : /* filtering by origin on a row level is much more efficient */
3282 andres@anarazel.de 2252 : 7099680 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2253 : :
4045 heikki.linnakangas@i 2254 : 7099680 : recptr = XLogInsert(RM_HEAP_ID, info);
2255 : :
9120 vadim4o@yahoo.com 2256 : 7099680 : PageSetLSN(page, recptr);
2257 : : }
2258 : :
9105 tgl@sss.pgh.pa.us 2259 [ - + ]: 8451123 : END_CRIT_SECTION();
2260 : :
7201 2261 : 8451123 : UnlockReleaseBuffer(buffer);
5293 rhaas@postgresql.org 2262 [ + + ]: 8451123 : if (vmbuffer != InvalidBuffer)
2263 : 7898 : ReleaseBuffer(vmbuffer);
2264 : :
2265 : : /*
2266 : : * If tuple is cacheable, mark it for invalidation from the caches in case
2267 : : * we abort. Note it is OK to do this after releasing the buffer, because
2268 : : * the heaptup data structure is all in local memory, not in the shared
2269 : : * buffer.
2270 : : */
5237 tgl@sss.pgh.pa.us 2271 : 8451123 : CacheInvalidateHeapTuple(relation, heaptup, NULL);
2272 : :
2273 : : /* Note: speculative insertions are counted too, even if aborted later */
5152 heikki.linnakangas@i 2274 : 8451123 : pgstat_count_heap_insert(relation, 1);
2275 : :
2276 : : /*
2277 : : * If heaptup is a private copy, release it. Don't forget to copy t_self
2278 : : * back to the caller's image, too.
2279 : : */
7332 tgl@sss.pgh.pa.us 2280 [ + + ]: 8451123 : if (heaptup != tup)
2281 : : {
2282 : 18566 : tup->t_self = heaptup->t_self;
2283 : 18566 : heap_freetuple(heaptup);
2284 : : }
10753 scrappy@hub.org 2285 : 8451123 : }
2286 : :
2287 : : /*
2288 : : * Subroutine for heap_insert(). Prepares a tuple for insertion. This sets the
2289 : : * tuple header fields and toasts the tuple if necessary. Returns a toasted
2290 : : * version of the tuple if it was toasted, or the original tuple if not. Note
2291 : : * that in any case, the header fields are also set in the original tuple.
2292 : : */
2293 : : static HeapTuple
5152 heikki.linnakangas@i 2294 : 10045674 : heap_prepare_insert(Relation relation, HeapTuple tup, TransactionId xid,
2295 : : CommandId cid, int options)
2296 : : {
2297 : : /*
2298 : : * To allow parallel inserts, we need to ensure that they are safe to be
2299 : : * performed in workers. We have the infrastructure to allow parallel
2300 : : * inserts in general except for the cases where inserts generate a new
2301 : : * CommandId (eg. inserts into a table having a foreign key column).
2302 : : */
2995 rhaas@postgresql.org 2303 [ - + ]: 10045674 : if (IsParallelWorker())
3884 rhaas@postgresql.org 2304 [ # # ]:UBC 0 : ereport(ERROR,
2305 : : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2306 : : errmsg("cannot insert tuples in a parallel worker")));
2307 : :
5152 heikki.linnakangas@i 2308 :CBC 10045674 : tup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
2309 : 10045674 : tup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
2310 : 10045674 : tup->t_data->t_infomask |= HEAP_XMAX_INVALID;
4378 rhaas@postgresql.org 2311 : 10045674 : HeapTupleHeaderSetXmin(tup->t_data, xid);
4763 simon@2ndQuadrant.co 2312 [ + + ]: 10045674 : if (options & HEAP_INSERT_FROZEN)
4378 rhaas@postgresql.org 2313 : 102088 : HeapTupleHeaderSetXminFrozen(tup->t_data);
2314 : :
5152 heikki.linnakangas@i 2315 : 10045674 : HeapTupleHeaderSetCmin(tup->t_data, cid);
3101 tgl@sss.pgh.pa.us 2316 : 10045674 : HeapTupleHeaderSetXmax(tup->t_data, 0); /* for cleanliness */
5152 heikki.linnakangas@i 2317 : 10045674 : tup->t_tableOid = RelationGetRelid(relation);
2318 : :
2319 : : /*
2320 : : * If the new tuple is too big for storage or contains already toasted
2321 : : * out-of-line attributes from some other relation, invoke the toaster.
2322 : : */
4672 kgrittn@postgresql.o 2323 [ + + ]: 10045674 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
2324 [ + + ]: 31443 : relation->rd_rel->relkind != RELKIND_MATVIEW)
2325 : : {
2326 : : /* toast table entries should never be recursively toasted */
5152 heikki.linnakangas@i 2327 [ - + ]: 31395 : Assert(!HeapTupleHasExternal(tup));
2328 : 31395 : return tup;
2329 : : }
2330 [ + + + + ]: 10014279 : else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
2266 rhaas@postgresql.org 2331 : 18610 : return heap_toast_insert_or_update(relation, tup, NULL, options);
2332 : : else
5152 heikki.linnakangas@i 2333 : 9995669 : return tup;
2334 : : }
2335 : :
2336 : : /*
2337 : : * Helper for heap_multi_insert() that computes the number of entire pages
2338 : : * that inserting the remaining heaptuples requires. Used to determine how
2339 : : * much the relation needs to be extended by.
2340 : : */
2341 : : static int
986 andres@anarazel.de 2342 : 372537 : heap_multi_insert_pages(HeapTuple *heaptuples, int done, int ntuples, Size saveFreeSpace)
2343 : : {
2344 : 372537 : size_t page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2345 : 372537 : int npages = 1;
2346 : :
2347 [ + + ]: 2646910 : for (int i = done; i < ntuples; i++)
2348 : : {
2349 : 2274373 : size_t tup_sz = sizeof(ItemIdData) + MAXALIGN(heaptuples[i]->t_len);
2350 : :
2351 [ + + ]: 2274373 : if (page_avail < tup_sz)
2352 : : {
2353 : 16322 : npages++;
2354 : 16322 : page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2355 : : }
2356 : 2274373 : page_avail -= tup_sz;
2357 : : }
2358 : :
2359 : 372537 : return npages;
2360 : : }
2361 : :
2362 : : /*
2363 : : * heap_multi_insert - insert multiple tuples into a heap
2364 : : *
2365 : : * This is like heap_insert(), but inserts multiple tuples in one operation.
2366 : : * That's faster than calling heap_insert() in a loop, because when multiple
2367 : : * tuples can be inserted on a single page, we can write just a single WAL
2368 : : * record covering all of them, and only need to lock/unlock the page once.
2369 : : *
2370 : : * Note: this leaks memory into the current memory context. You can create a
2371 : : * temporary context before calling this, if that's a problem.
2372 : : */
2373 : : void
2449 2374 : 365819 : heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
2375 : : CommandId cid, int options, BulkInsertState bistate)
2376 : : {
5152 heikki.linnakangas@i 2377 : 365819 : TransactionId xid = GetCurrentTransactionId();
2378 : : HeapTuple *heaptuples;
2379 : : int i;
2380 : : int ndone;
2381 : : PGAlignedBlock scratch;
2382 : : Page page;
1795 tomas.vondra@postgre 2383 : 365819 : Buffer vmbuffer = InvalidBuffer;
2384 : : bool needwal;
2385 : : Size saveFreeSpace;
4390 rhaas@postgresql.org 2386 [ + + + - : 365819 : bool need_tuple_data = RelationIsLogicallyLogged(relation);
- + - - -
- + - +
+ ]
2387 [ + + + - : 365819 : bool need_cids = RelationIsAccessibleInLogicalDecoding(relation);
- + - - -
- + + - +
- - - - -
- ]
986 andres@anarazel.de 2388 : 365819 : bool starting_with_empty_page = false;
2389 : 365819 : int npages = 0;
2390 : 365819 : int npages_used = 0;
2391 : :
2392 : : /* currently not needed (thus unsupported) for heap_multi_insert() */
1146 peter@eisentraut.org 2393 [ - + ]: 365819 : Assert(!(options & HEAP_INSERT_NO_LOGICAL));
2394 : :
201 nathan@postgresql.or 2395 : 365819 : AssertHasSnapshotForToast(relation);
2396 : :
2083 noah@leadboat.com 2397 [ + + + + : 365819 : needwal = RelationNeedsWAL(relation);
+ + + + ]
615 akorotkov@postgresql 2398 [ + + ]: 365819 : saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
2399 : : HEAP_DEFAULT_FILLFACTOR);
2400 : :
2401 : : /* Toast and set header data in all the slots */
5152 heikki.linnakangas@i 2402 : 365819 : heaptuples = palloc(ntuples * sizeof(HeapTuple));
2403 [ + + ]: 1960358 : for (i = 0; i < ntuples; i++)
2404 : : {
2405 : : HeapTuple tuple;
2406 : :
2449 andres@anarazel.de 2407 : 1594539 : tuple = ExecFetchSlotHeapTuple(slots[i], true, NULL);
2408 : 1594539 : slots[i]->tts_tableOid = RelationGetRelid(relation);
2409 : 1594539 : tuple->t_tableOid = slots[i]->tts_tableOid;
2410 : 1594539 : heaptuples[i] = heap_prepare_insert(relation, tuple, xid, cid,
2411 : : options);
2412 : : }
2413 : :
2414 : : /*
2415 : : * We're about to do the actual inserts -- but check for conflict first,
2416 : : * to minimize the possibility of having to roll back work we've just
2417 : : * done.
2418 : : *
2419 : : * A check here does not definitively prevent a serialization anomaly;
2420 : : * that check MUST be done at least past the point of acquiring an
2421 : : * exclusive buffer content lock on every buffer that will be affected,
2422 : : * and MAY be done after all inserts are reflected in the buffers and
2423 : : * those locks are released; otherwise there is a race condition. Since
2424 : : * multiple buffers can be locked and unlocked in the loop below, and it
2425 : : * would not be feasible to identify and lock all of those buffers before
2426 : : * the loop, we must do a final check at the end.
2427 : : *
2428 : : * The check here could be omitted with no loss of correctness; it is
2429 : : * present strictly as an optimization.
2430 : : *
2431 : : * For heap inserts, we only need to check for table-level SSI locks. Our
2432 : : * new tuples can't possibly conflict with existing tuple locks, and heap
2433 : : * page locks are only consolidated versions of tuple locks; they do not
2434 : : * lock "gaps" as index page locks do. So we don't need to specify a
2435 : : * buffer when making the call, which makes for a faster check.
2436 : : */
2150 tmunro@postgresql.or 2437 : 365819 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2438 : :
5152 heikki.linnakangas@i 2439 : 365819 : ndone = 0;
2440 [ + + ]: 746906 : while (ndone < ntuples)
2441 : : {
2442 : : Buffer buffer;
2443 : 381087 : bool all_visible_cleared = false;
1795 tomas.vondra@postgre 2444 : 381087 : bool all_frozen_set = false;
2445 : : int nthispage;
2446 : :
4195 rhaas@postgresql.org 2447 [ - + ]: 381087 : CHECK_FOR_INTERRUPTS();
2448 : :
2449 : : /*
2450 : : * Compute number of pages needed to fit the to-be-inserted tuples in
2451 : : * the worst case. This will be used to determine how much to extend
2452 : : * the relation by in RelationGetBufferForTuple(), if needed. If we
2453 : : * filled a prior page from scratch, we can just update our last
2454 : : * computation, but if we started with a partially filled page,
2455 : : * recompute from scratch, the number of potentially required pages
2456 : : * can vary due to tuples needing to fit onto the page, page headers
2457 : : * etc.
2458 : : */
986 andres@anarazel.de 2459 [ + + + + ]: 381087 : if (ndone == 0 || !starting_with_empty_page)
2460 : : {
2461 : 372537 : npages = heap_multi_insert_pages(heaptuples, ndone, ntuples,
2462 : : saveFreeSpace);
2463 : 372537 : npages_used = 0;
2464 : : }
2465 : : else
2466 : 8550 : npages_used++;
2467 : :
2468 : : /*
2469 : : * Find buffer where at least the next tuple will fit. If the page is
2470 : : * all-visible, this will also pin the requisite visibility map page.
2471 : : *
2472 : : * Also pin visibility map page if COPY FREEZE inserts tuples into an
2473 : : * empty page. See all_frozen_set below.
2474 : : */
5152 heikki.linnakangas@i 2475 : 381087 : buffer = RelationGetBufferForTuple(relation, heaptuples[ndone]->t_len,
2476 : : InvalidBuffer, options, bistate,
2477 : : &vmbuffer, NULL,
2478 : : npages - npages_used);
3528 kgrittn@postgresql.o 2479 : 381087 : page = BufferGetPage(buffer);
2480 : :
1795 tomas.vondra@postgre 2481 : 381087 : starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
2482 : :
2483 [ + + + + ]: 381087 : if (starting_with_empty_page && (options & HEAP_INSERT_FROZEN))
2484 : : {
2485 : 1661 : all_frozen_set = true;
2486 : : /* Lock the vmbuffer before entering the critical section */
69 melanieplageman@gmai 2487 :GNC 1661 : LockBuffer(vmbuffer, BUFFER_LOCK_EXCLUSIVE);
2488 : : }
2489 : :
2490 : : /* NO EREPORT(ERROR) from here till changes are logged */
5152 heikki.linnakangas@i 2491 :CBC 381087 : START_CRIT_SECTION();
2492 : :
2493 : : /*
2494 : : * RelationGetBufferForTuple has ensured that the first tuple fits.
2495 : : * Put that on the page, and then as many other tuples as fit.
2496 : : */
3876 andres@anarazel.de 2497 : 381087 : RelationPutHeapTuple(relation, buffer, heaptuples[ndone], false);
2498 : :
2499 : : /*
2500 : : * For logical decoding we need combo CIDs to properly decode the
2501 : : * catalog.
2502 : : */
2122 michael@paquier.xyz 2503 [ + + + + ]: 381087 : if (needwal && need_cids)
2504 : 4852 : log_heap_new_cid(relation, heaptuples[ndone]);
2505 : :
4753 heikki.linnakangas@i 2506 [ + + ]: 1594539 : for (nthispage = 1; ndone + nthispage < ntuples; nthispage++)
2507 : : {
5152 2508 : 1228720 : HeapTuple heaptup = heaptuples[ndone + nthispage];
2509 : :
4963 2510 [ + + ]: 1228720 : if (PageGetHeapFreeSpace(page) < MAXALIGN(heaptup->t_len) + saveFreeSpace)
5152 2511 : 15268 : break;
2512 : :
3876 andres@anarazel.de 2513 : 1213452 : RelationPutHeapTuple(relation, buffer, heaptup, false);
2514 : :
2515 : : /*
2516 : : * For logical decoding we need combo CIDs to properly decode the
2517 : : * catalog.
2518 : : */
4045 heikki.linnakangas@i 2519 [ + + + + ]: 1213452 : if (needwal && need_cids)
2520 : 4592 : log_heap_new_cid(relation, heaptup);
2521 : : }
2522 : :
2523 : : /*
2524 : : * If the page is all visible, need to clear that, unless we're only
2525 : : * going to add further frozen rows to it.
2526 : : *
2527 : : * If we're only adding already frozen rows to a previously empty
2528 : : * page, mark it as all-frozen and update the visibility map. We're
2529 : : * already holding a pin on the vmbuffer.
2530 : : */
1795 tomas.vondra@postgre 2531 [ + + + + ]: 381087 : if (PageIsAllVisible(page) && !(options & HEAP_INSERT_FROZEN))
2532 : : {
4941 rhaas@postgresql.org 2533 : 3448 : all_visible_cleared = true;
2534 : 3448 : PageClearAllVisible(page);
2535 : 3448 : visibilitymap_clear(relation,
2536 : : BufferGetBlockNumber(buffer),
2537 : : vmbuffer, VISIBILITYMAP_VALID_BITS);
2538 : : }
1795 tomas.vondra@postgre 2539 [ + + ]: 377639 : else if (all_frozen_set)
2540 : : {
2541 : 1661 : PageSetAllVisible(page);
69 melanieplageman@gmai 2542 :GNC 1661 : visibilitymap_set_vmbits(BufferGetBlockNumber(buffer),
2543 : : vmbuffer,
2544 : : VISIBILITYMAP_ALL_VISIBLE |
2545 : : VISIBILITYMAP_ALL_FROZEN,
2546 : : relation->rd_locator);
2547 : : }
2548 : :
2549 : : /*
2550 : : * XXX Should we set PageSetPrunable on this page ? See heap_insert()
2551 : : */
2552 : :
5152 heikki.linnakangas@i 2553 :CBC 381087 : MarkBufferDirty(buffer);
2554 : :
2555 : : /* XLOG stuff */
2556 [ + + ]: 381087 : if (needwal)
2557 : : {
2558 : : XLogRecPtr recptr;
2559 : : xl_heap_multi_insert *xlrec;
2560 : 376717 : uint8 info = XLOG_HEAP2_MULTI_INSERT;
2561 : : char *tupledata;
2562 : : int totaldatalen;
2664 tgl@sss.pgh.pa.us 2563 : 376717 : char *scratchptr = scratch.data;
2564 : : bool init;
4045 heikki.linnakangas@i 2565 : 376717 : int bufflags = 0;
2566 : :
2567 : : /*
2568 : : * If the page was previously empty, we can reinit the page
2569 : : * instead of restoring the whole thing.
2570 : : */
1795 tomas.vondra@postgre 2571 : 376717 : init = starting_with_empty_page;
2572 : :
2573 : : /* allocate xl_heap_multi_insert struct from the scratch area */
5152 heikki.linnakangas@i 2574 : 376717 : xlrec = (xl_heap_multi_insert *) scratchptr;
2575 : 376717 : scratchptr += SizeOfHeapMultiInsert;
2576 : :
2577 : : /*
2578 : : * Allocate offsets array. Unless we're reinitializing the page,
2579 : : * in that case the tuples are stored in order starting at
2580 : : * FirstOffsetNumber and we don't need to store the offsets
2581 : : * explicitly.
2582 : : */
2583 [ + + ]: 376717 : if (!init)
2584 : 363796 : scratchptr += nthispage * sizeof(OffsetNumber);
2585 : :
2586 : : /* the rest of the scratch space is used for tuple data */
2587 : 376717 : tupledata = scratchptr;
2588 : :
2589 : : /* check that the mutually exclusive flags are not both set */
1680 tgl@sss.pgh.pa.us 2590 [ + + - + ]: 376717 : Assert(!(all_visible_cleared && all_frozen_set));
2591 : :
1795 tomas.vondra@postgre 2592 : 376717 : xlrec->flags = 0;
2593 [ + + ]: 376717 : if (all_visible_cleared)
2594 : 3448 : xlrec->flags = XLH_INSERT_ALL_VISIBLE_CLEARED;
2595 : :
2596 : : /*
2597 : : * We don't have to worry about including a conflict xid in the
2598 : : * WAL record, as HEAP_INSERT_FROZEN intentionally violates
2599 : : * visibility rules.
2600 : : */
2601 [ + + ]: 376717 : if (all_frozen_set)
2602 : 13 : xlrec->flags = XLH_INSERT_ALL_FROZEN_SET;
2603 : :
5152 heikki.linnakangas@i 2604 : 376717 : xlrec->ntuples = nthispage;
2605 : :
2606 : : /*
2607 : : * Write out an xl_multi_insert_tuple and the tuple data itself
2608 : : * for each tuple.
2609 : : */
2610 [ + + ]: 1665841 : for (i = 0; i < nthispage; i++)
2611 : : {
2612 : 1289124 : HeapTuple heaptup = heaptuples[ndone + i];
2613 : : xl_multi_insert_tuple *tuphdr;
2614 : : int datalen;
2615 : :
2616 [ + + ]: 1289124 : if (!init)
2617 : 763816 : xlrec->offsets[i] = ItemPointerGetOffsetNumber(&heaptup->t_self);
2618 : : /* xl_multi_insert_tuple needs two-byte alignment. */
2619 : 1289124 : tuphdr = (xl_multi_insert_tuple *) SHORTALIGN(scratchptr);
2620 : 1289124 : scratchptr = ((char *) tuphdr) + SizeOfMultiInsertTuple;
2621 : :
2622 : 1289124 : tuphdr->t_infomask2 = heaptup->t_data->t_infomask2;
2623 : 1289124 : tuphdr->t_infomask = heaptup->t_data->t_infomask;
2624 : 1289124 : tuphdr->t_hoff = heaptup->t_data->t_hoff;
2625 : :
2626 : : /* write bitmap [+ padding] [+ oid] + data */
3952 tgl@sss.pgh.pa.us 2627 : 1289124 : datalen = heaptup->t_len - SizeofHeapTupleHeader;
5152 heikki.linnakangas@i 2628 : 1289124 : memcpy(scratchptr,
3952 tgl@sss.pgh.pa.us 2629 : 1289124 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2630 : : datalen);
5152 heikki.linnakangas@i 2631 : 1289124 : tuphdr->datalen = datalen;
2632 : 1289124 : scratchptr += datalen;
2633 : : }
2634 : 376717 : totaldatalen = scratchptr - tupledata;
2664 tgl@sss.pgh.pa.us 2635 [ - + ]: 376717 : Assert((scratchptr - scratch.data) < BLCKSZ);
2636 : :
4390 rhaas@postgresql.org 2637 [ + + ]: 376717 : if (need_tuple_data)
3876 andres@anarazel.de 2638 : 72 : xlrec->flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2639 : :
2640 : : /*
2641 : : * Signal that this is the last xl_heap_multi_insert record
2642 : : * emitted by this call to heap_multi_insert(). Needed for logical
2643 : : * decoding so it knows when to cleanup temporary data.
2644 : : */
4045 heikki.linnakangas@i 2645 [ + + ]: 376717 : if (ndone + nthispage == ntuples)
3876 andres@anarazel.de 2646 : 365300 : xlrec->flags |= XLH_INSERT_LAST_IN_MULTI;
2647 : :
5152 heikki.linnakangas@i 2648 [ + + ]: 376717 : if (init)
2649 : : {
2650 : 12921 : info |= XLOG_HEAP_INIT_PAGE;
4045 2651 : 12921 : bufflags |= REGBUF_WILL_INIT;
2652 : : }
2653 : :
2654 : : /*
2655 : : * If we're doing logical decoding, include the new tuple data
2656 : : * even if we take a full-page image of the page.
2657 : : */
2658 [ + + ]: 376717 : if (need_tuple_data)
2659 : 72 : bufflags |= REGBUF_KEEP_DATA;
2660 : :
2661 : 376717 : XLogBeginInsert();
309 peter@eisentraut.org 2662 : 376717 : XLogRegisterData(xlrec, tupledata - scratch.data);
4045 heikki.linnakangas@i 2663 : 376717 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
69 melanieplageman@gmai 2664 [ + + ]:GNC 376717 : if (all_frozen_set)
2665 : 13 : XLogRegisterBuffer(1, vmbuffer, 0);
2666 : :
4045 heikki.linnakangas@i 2667 :CBC 376717 : XLogRegisterBufData(0, tupledata, totaldatalen);
2668 : :
2669 : : /* filtering by origin on a row level is much more efficient */
3282 andres@anarazel.de 2670 : 376717 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2671 : :
4045 heikki.linnakangas@i 2672 : 376717 : recptr = XLogInsert(RM_HEAP2_ID, info);
2673 : :
5152 2674 : 376717 : PageSetLSN(page, recptr);
69 melanieplageman@gmai 2675 [ + + ]:GNC 376717 : if (all_frozen_set)
2676 : : {
2677 [ - + ]: 13 : Assert(BufferIsDirty(vmbuffer));
2678 : 13 : PageSetLSN(BufferGetPage(vmbuffer), recptr);
2679 : : }
2680 : : }
2681 : :
5152 heikki.linnakangas@i 2682 [ - + ]:CBC 381087 : END_CRIT_SECTION();
2683 : :
1795 tomas.vondra@postgre 2684 [ + + ]: 381087 : if (all_frozen_set)
69 melanieplageman@gmai 2685 :GNC 1661 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
2686 : :
1795 tomas.vondra@postgre 2687 :CBC 381087 : UnlockReleaseBuffer(buffer);
5152 heikki.linnakangas@i 2688 : 381087 : ndone += nthispage;
2689 : :
2690 : : /*
2691 : : * NB: Only release vmbuffer after inserting all tuples - it's fairly
2692 : : * likely that we'll insert into subsequent heap pages that are likely
2693 : : * to use the same vm page.
2694 : : */
2695 : : }
2696 : :
2697 : : /* We're done with inserting all tuples, so release the last vmbuffer. */
1795 tomas.vondra@postgre 2698 [ + + ]: 365819 : if (vmbuffer != InvalidBuffer)
2699 : 3533 : ReleaseBuffer(vmbuffer);
2700 : :
2701 : : /*
2702 : : * We're done with the actual inserts. Check for conflicts again, to
2703 : : * ensure that all rw-conflicts in to these inserts are detected. Without
2704 : : * this final check, a sequential scan of the heap may have locked the
2705 : : * table after the "before" check, missing one opportunity to detect the
2706 : : * conflict, and then scanned the table before the new tuples were there,
2707 : : * missing the other chance to detect the conflict.
2708 : : *
2709 : : * For heap inserts, we only need to check for table-level SSI locks. Our
2710 : : * new tuples can't possibly conflict with existing tuple locks, and heap
2711 : : * page locks are only consolidated versions of tuple locks; they do not
2712 : : * lock "gaps" as index page locks do. So we don't need to specify a
2713 : : * buffer when making the call.
2714 : : */
2150 tmunro@postgresql.or 2715 : 365819 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2716 : :
2717 : : /*
2718 : : * If tuples are cacheable, mark them for invalidation from the caches in
2719 : : * case we abort. Note it is OK to do this after releasing the buffer,
2720 : : * because the heaptuples data structure is all in local memory, not in
2721 : : * the shared buffer.
2722 : : */
4402 rhaas@postgresql.org 2723 [ + + ]: 365819 : if (IsCatalogRelation(relation))
2724 : : {
5152 heikki.linnakangas@i 2725 [ + + ]: 1256718 : for (i = 0; i < ntuples; i++)
2726 : 892215 : CacheInvalidateHeapTuple(relation, heaptuples[i], NULL);
2727 : : }
2728 : :
2729 : : /* copy t_self fields back to the caller's slots */
5056 2730 [ + + ]: 1960358 : for (i = 0; i < ntuples; i++)
2449 andres@anarazel.de 2731 : 1594539 : slots[i]->tts_tid = heaptuples[i]->t_self;
2732 : :
5152 heikki.linnakangas@i 2733 : 365819 : pgstat_count_heap_insert(relation, ntuples);
2734 : 365819 : }
2735 : :
2736 : : /*
2737 : : * simple_heap_insert - insert a tuple
2738 : : *
2739 : : * Currently, this routine differs from heap_insert only in supplying
2740 : : * a default command ID and not allowing access to the speedup options.
2741 : : *
2742 : : * This should be used rather than using heap_insert directly in most places
2743 : : * where we are modifying system catalogs.
2744 : : */
2745 : : void
8611 tgl@sss.pgh.pa.us 2746 : 953133 : simple_heap_insert(Relation relation, HeapTuple tup)
2747 : : {
2584 andres@anarazel.de 2748 : 953133 : heap_insert(relation, tup, GetCurrentCommandId(true), 0, NULL);
8611 tgl@sss.pgh.pa.us 2749 : 953133 : }
2750 : :
2751 : : /*
2752 : : * Given infomask/infomask2, compute the bits that must be saved in the
2753 : : * "infobits" field of xl_heap_delete, xl_heap_update, xl_heap_lock,
2754 : : * xl_heap_lock_updated WAL records.
2755 : : *
2756 : : * See fix_infomask_from_infobits.
2757 : : */
2758 : : static uint8
4711 alvherre@alvh.no-ip. 2759 : 1992823 : compute_infobits(uint16 infomask, uint16 infomask2)
2760 : : {
2761 : : return
2762 : 1992823 : ((infomask & HEAP_XMAX_IS_MULTI) != 0 ? XLHL_XMAX_IS_MULTI : 0) |
2763 : 1992823 : ((infomask & HEAP_XMAX_LOCK_ONLY) != 0 ? XLHL_XMAX_LOCK_ONLY : 0) |
2764 : 1992823 : ((infomask & HEAP_XMAX_EXCL_LOCK) != 0 ? XLHL_XMAX_EXCL_LOCK : 0) |
2765 : : /* note we ignore HEAP_XMAX_SHR_LOCK here */
2766 : 3985646 : ((infomask & HEAP_XMAX_KEYSHR_LOCK) != 0 ? XLHL_XMAX_KEYSHR_LOCK : 0) |
2767 : : ((infomask2 & HEAP_KEYS_UPDATED) != 0 ?
2768 : 1992823 : XLHL_KEYS_UPDATED : 0);
2769 : : }
2770 : :
2771 : : /*
2772 : : * Given two versions of the same t_infomask for a tuple, compare them and
2773 : : * return whether the relevant status for a tuple Xmax has changed. This is
2774 : : * used after a buffer lock has been released and reacquired: we want to ensure
2775 : : * that the tuple state continues to be the same it was when we previously
2776 : : * examined it.
2777 : : *
2778 : : * Note the Xmax field itself must be compared separately.
2779 : : */
2780 : : static inline bool
4255 2781 : 5376 : xmax_infomask_changed(uint16 new_infomask, uint16 old_infomask)
2782 : : {
4243 bruce@momjian.us 2783 : 5376 : const uint16 interesting =
2784 : : HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY | HEAP_LOCK_MASK;
2785 : :
4255 alvherre@alvh.no-ip. 2786 [ + + ]: 5376 : if ((new_infomask & interesting) != (old_infomask & interesting))
2787 : 12 : return true;
2788 : :
2789 : 5364 : return false;
2790 : : }
2791 : :
2792 : : /*
2793 : : * heap_delete - delete a tuple
2794 : : *
2795 : : * See table_tuple_delete() for an explanation of the parameters, except that
2796 : : * this routine directly takes a tuple rather than a slot.
2797 : : *
2798 : : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
2799 : : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
2800 : : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
2801 : : * generated by another transaction).
2802 : : */
2803 : : TM_Result
48 peter@eisentraut.org 2804 :GNC 1454502 : heap_delete(Relation relation, const ItemPointerData *tid,
2805 : : CommandId cid, Snapshot crosscheck, bool wait,
2806 : : TM_FailureData *tmfd, bool changingPart)
2807 : : {
2808 : : TM_Result result;
7762 tgl@sss.pgh.pa.us 2809 :CBC 1454502 : TransactionId xid = GetCurrentTransactionId();
2810 : : ItemId lp;
2811 : : HeapTupleData tp;
2812 : : Page page;
2813 : : BlockNumber block;
2814 : : Buffer buffer;
5293 rhaas@postgresql.org 2815 : 1454502 : Buffer vmbuffer = InvalidBuffer;
2816 : : TransactionId new_xmax;
2817 : : uint16 new_infomask,
2818 : : new_infomask2;
7536 tgl@sss.pgh.pa.us 2819 : 1454502 : bool have_tuple_lock = false;
2820 : : bool iscombo;
6223 heikki.linnakangas@i 2821 : 1454502 : bool all_visible_cleared = false;
4243 bruce@momjian.us 2822 : 1454502 : HeapTuple old_key_tuple = NULL; /* replica identity of the tuple */
4390 rhaas@postgresql.org 2823 : 1454502 : bool old_key_copied = false;
2824 : :
10328 bruce@momjian.us 2825 [ - + ]: 1454502 : Assert(ItemPointerIsValid(tid));
2826 : :
201 nathan@postgresql.or 2827 : 1454502 : AssertHasSnapshotForToast(relation);
2828 : :
2829 : : /*
2830 : : * Forbid this during a parallel operation, lest it allocate a combo CID.
2831 : : * Other workers might need that combo CID for visibility checks, and we
2832 : : * have no provision for broadcasting it to them.
2833 : : */
3884 rhaas@postgresql.org 2834 [ - + ]: 1454502 : if (IsInParallelMode())
3884 rhaas@postgresql.org 2835 [ # # ]:UBC 0 : ereport(ERROR,
2836 : : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2837 : : errmsg("cannot delete tuples during a parallel operation")));
2838 : :
5293 rhaas@postgresql.org 2839 :CBC 1454502 : block = ItemPointerGetBlockNumber(tid);
2840 : 1454502 : buffer = ReadBuffer(relation, block);
3528 kgrittn@postgresql.o 2841 : 1454502 : page = BufferGetPage(buffer);
2842 : :
2843 : : /*
2844 : : * Before locking the buffer, pin the visibility map page if it appears to
2845 : : * be necessary. Since we haven't got the lock yet, someone else might be
2846 : : * in the middle of changing this, so we'll need to recheck after we have
2847 : : * the lock.
2848 : : */
5293 rhaas@postgresql.org 2849 [ + + ]: 1454502 : if (PageIsAllVisible(page))
2850 : 355 : visibilitymap_pin(relation, block, &vmbuffer);
2851 : :
9864 vadim4o@yahoo.com 2852 : 1454502 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2853 : :
1182 jdavis@postgresql.or 2854 : 1454502 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
2855 [ - + ]: 1454502 : Assert(ItemIdIsNormal(lp));
2856 : :
2857 : 1454502 : tp.t_tableOid = RelationGetRelid(relation);
2858 : 1454502 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2859 : 1454502 : tp.t_len = ItemIdGetLength(lp);
2860 : 1454502 : tp.t_self = *tid;
2861 : :
2862 : 1 : l1:
2863 : :
2864 : : /*
2865 : : * If we didn't pin the visibility map page and the page has become all
2866 : : * visible while we were busy locking the buffer, we'll have to unlock and
2867 : : * re-lock, to avoid holding the buffer lock across an I/O. That's a bit
2868 : : * unfortunate, but hopefully shouldn't happen often.
2869 : : */
5293 rhaas@postgresql.org 2870 [ + + - + ]: 1454503 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
2871 : : {
5293 rhaas@postgresql.org 2872 :UBC 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2873 : 0 : visibilitymap_pin(relation, block, &vmbuffer);
2874 : 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2875 : : }
2876 : :
4531 rhaas@postgresql.org 2877 :CBC 1454503 : result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
2878 : :
2461 andres@anarazel.de 2879 [ - + ]: 1454503 : if (result == TM_Invisible)
2880 : : {
7201 tgl@sss.pgh.pa.us 2881 :UBC 0 : UnlockReleaseBuffer(buffer);
3790 2882 [ # # ]: 0 : ereport(ERROR,
2883 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2884 : : errmsg("attempted to delete invisible tuple")));
2885 : : }
615 akorotkov@postgresql 2886 [ + + + - ]:CBC 1454503 : else if (result == TM_BeingModified && wait)
2887 : : {
2888 : : TransactionId xwait;
2889 : : uint16 infomask;
2890 : :
2891 : : /* must copy state data before unlocking buffer */
4711 alvherre@alvh.no-ip. 2892 : 40559 : xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
7536 tgl@sss.pgh.pa.us 2893 : 40559 : infomask = tp.t_data->t_infomask;
2894 : :
2895 : : /*
2896 : : * Sleep until concurrent transaction ends -- except when there's a
2897 : : * single locker and it's our own transaction. Note we don't care
2898 : : * which lock mode the locker has, because we need the strongest one.
2899 : : *
2900 : : * Before sleeping, we need to acquire tuple lock to establish our
2901 : : * priority for the tuple (see heap_lock_tuple). LockTuple will
2902 : : * release us when we are next-in-line for the tuple.
2903 : : *
2904 : : * If we are forced to "start over" below, we keep the tuple lock;
2905 : : * this arranges that we stay at the head of the line while rechecking
2906 : : * tuple state.
2907 : : */
7538 2908 [ + + ]: 40559 : if (infomask & HEAP_XMAX_IS_MULTI)
2909 : : {
2374 alvherre@alvh.no-ip. 2910 : 8 : bool current_is_member = false;
2911 : :
3904 2912 [ + - ]: 8 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
2913 : : LockTupleExclusive, ¤t_is_member))
2914 : : {
2915 : 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2916 : :
2917 : : /*
2918 : : * Acquire the lock, if necessary (but skip it when we're
2919 : : * requesting a lock and already have one; avoids deadlock).
2920 : : */
2374 2921 [ + + ]: 8 : if (!current_is_member)
2922 : 6 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2923 : : LockWaitBlock, &have_tuple_lock);
2924 : :
2925 : : /* wait for multixact */
3904 2926 : 8 : MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
2927 : : relation, &(tp.t_self), XLTW_Delete,
2928 : : NULL);
2929 : 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2930 : :
2931 : : /*
2932 : : * If xwait had just locked the tuple then some other xact
2933 : : * could update this tuple before we get to this point. Check
2934 : : * for xmax change, and start over if so.
2935 : : *
2936 : : * We also must start over if we didn't pin the VM page, and
2937 : : * the page has become all visible.
2938 : : */
1182 jdavis@postgresql.or 2939 [ + - + - : 16 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
+ - ]
2940 [ - + ]: 16 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
3904 alvherre@alvh.no-ip. 2941 : 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2942 : : xwait))
3904 alvherre@alvh.no-ip. 2943 :UBC 0 : goto l1;
2944 : : }
2945 : :
2946 : : /*
2947 : : * You might think the multixact is necessarily done here, but not
2948 : : * so: it could have surviving members, namely our own xact or
2949 : : * other subxacts of this backend. It is legal for us to delete
2950 : : * the tuple in either case, however (the latter case is
2951 : : * essentially a situation of upgrading our former shared lock to
2952 : : * exclusive). We don't bother changing the on-disk hint bits
2953 : : * since we are about to overwrite the xmax altogether.
2954 : : */
2955 : : }
3904 alvherre@alvh.no-ip. 2956 [ + + ]:CBC 40551 : else if (!TransactionIdIsCurrentTransactionId(xwait))
2957 : : {
2958 : : /*
2959 : : * Wait for regular transaction to end; but first, acquire tuple
2960 : : * lock.
2961 : : */
2962 : 52 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2963 : 52 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2964 : : LockWaitBlock, &have_tuple_lock);
3969 heikki.linnakangas@i 2965 : 52 : XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
7538 tgl@sss.pgh.pa.us 2966 : 48 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2967 : :
2968 : : /*
2969 : : * xwait is done, but if xwait had just locked the tuple then some
2970 : : * other xact could update this tuple before we get to this point.
2971 : : * Check for xmax change, and start over if so.
2972 : : *
2973 : : * We also must start over if we didn't pin the VM page, and the
2974 : : * page has become all visible.
2975 : : */
1182 jdavis@postgresql.or 2976 [ + - + - : 96 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
+ + ]
2977 [ - + ]: 95 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
4711 alvherre@alvh.no-ip. 2978 : 47 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2979 : : xwait))
7538 tgl@sss.pgh.pa.us 2980 : 1 : goto l1;
2981 : :
2982 : : /* Otherwise check if it committed or aborted */
6700 2983 : 47 : UpdateXmaxHintBits(tp.t_data, buffer, xwait);
2984 : : }
2985 : :
2986 : : /*
2987 : : * We may overwrite if previous xmax aborted, or if it committed but
2988 : : * only locked the tuple without updating it.
2989 : : */
4711 alvherre@alvh.no-ip. 2990 [ + + + + ]: 81088 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
2991 [ + + ]: 40565 : HEAP_XMAX_IS_LOCKED_ONLY(tp.t_data->t_infomask) ||
2992 : 31 : HeapTupleHeaderIsOnlyLocked(tp.t_data))
2461 andres@anarazel.de 2993 : 40527 : result = TM_Ok;
1759 alvherre@alvh.no-ip. 2994 [ + + ]: 27 : else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
2461 andres@anarazel.de 2995 : 23 : result = TM_Updated;
2996 : : else
2997 : 4 : result = TM_Deleted;
2998 : : }
2999 : :
3000 : : /* sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3001 [ + + ]: 1454498 : if (result != TM_Ok)
3002 : : {
3003 [ + + + + : 61 : Assert(result == TM_SelfModified ||
- + - - ]
3004 : : result == TM_Updated ||
3005 : : result == TM_Deleted ||
3006 : : result == TM_BeingModified);
7424 tgl@sss.pgh.pa.us 3007 [ - + ]: 61 : Assert(!(tp.t_data->t_infomask & HEAP_XMAX_INVALID));
2461 andres@anarazel.de 3008 [ + + - + ]: 61 : Assert(result != TM_Updated ||
3009 : : !ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
3010 : : }
3011 : :
750 heikki.linnakangas@i 3012 [ + + + - ]: 1454498 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3013 : : {
3014 : : /* Perform additional check for transaction-snapshot mode RI updates */
3015 [ + - ]: 1 : if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
3016 : 1 : result = TM_Updated;
3017 : : }
3018 : :
3019 [ + + ]: 1454498 : if (result != TM_Ok)
3020 : : {
2461 andres@anarazel.de 3021 : 62 : tmfd->ctid = tp.t_data->t_ctid;
3022 : 62 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
3023 [ + + ]: 62 : if (result == TM_SelfModified)
3024 : 21 : tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
3025 : : else
3026 : 41 : tmfd->cmax = InvalidCommandId;
615 akorotkov@postgresql 3027 : 62 : UnlockReleaseBuffer(buffer);
7536 tgl@sss.pgh.pa.us 3028 [ + + ]: 62 : if (have_tuple_lock)
4711 alvherre@alvh.no-ip. 3029 : 27 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
5293 rhaas@postgresql.org 3030 [ - + ]: 62 : if (vmbuffer != InvalidBuffer)
5293 rhaas@postgresql.org 3031 :UBC 0 : ReleaseBuffer(vmbuffer);
9864 vadim4o@yahoo.com 3032 :CBC 62 : return result;
3033 : : }
3034 : :
3035 : : /*
3036 : : * We're about to do the actual delete -- check for conflict first, to
3037 : : * avoid possibly having to roll back work we've just done.
3038 : : *
3039 : : * This is safe without a recheck as long as there is no possibility of
3040 : : * another process scanning the page between this check and the delete
3041 : : * being visible to the scan (i.e., an exclusive buffer content lock is
3042 : : * continuously held from this point until the tuple delete is visible).
3043 : : */
2150 tmunro@postgresql.or 3044 : 1454436 : CheckForSerializableConflictIn(relation, tid, BufferGetBlockNumber(buffer));
3045 : :
3046 : : /* replace cid with a combo CID if necessary */
6886 tgl@sss.pgh.pa.us 3047 : 1454422 : HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
3048 : :
3049 : : /*
3050 : : * Compute replica identity tuple before entering the critical section so
3051 : : * we don't PANIC upon a memory allocation failure.
3052 : : */
4390 rhaas@postgresql.org 3053 : 1454422 : old_key_tuple = ExtractReplicaIdentity(relation, &tp, true, &old_key_copied);
3054 : :
3055 : : /*
3056 : : * If this is the first possibly-multixact-able operation in the current
3057 : : * transaction, set my per-backend OldestMemberMXactId setting. We can be
3058 : : * certain that the transaction will never become a member of any older
3059 : : * MultiXactIds than that. (We have to do this even if we end up just
3060 : : * using our own TransactionId below, since some other backend could
3061 : : * incorporate our XID into a MultiXact immediately afterwards.)
3062 : : */
4275 heikki.linnakangas@i 3063 : 1454422 : MultiXactIdSetOldestMember();
3064 : :
3065 : 1454422 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(tp.t_data),
3066 : 1454422 : tp.t_data->t_infomask, tp.t_data->t_infomask2,
3067 : : xid, LockTupleExclusive, true,
3068 : : &new_xmax, &new_infomask, &new_infomask2);
3069 : :
9105 tgl@sss.pgh.pa.us 3070 : 1454422 : START_CRIT_SECTION();
3071 : :
3072 : : /*
3073 : : * If this transaction commits, the tuple will become DEAD sooner or
3074 : : * later. Set flag that this page is a candidate for pruning once our xid
3075 : : * falls below the OldestXmin horizon. If the transaction finally aborts,
3076 : : * the subsequent page pruning will be a no-op and the hint will be
3077 : : * cleared.
3078 : : */
6366 3079 [ - + + + : 1454422 : PageSetPrunable(page, xid);
+ + ]
3080 : :
6223 heikki.linnakangas@i 3081 [ + + ]: 1454422 : if (PageIsAllVisible(page))
3082 : : {
3083 : 355 : all_visible_cleared = true;
3084 : 355 : PageClearAllVisible(page);
5293 rhaas@postgresql.org 3085 : 355 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
3086 : : vmbuffer, VISIBILITYMAP_VALID_BITS);
3087 : : }
3088 : :
3089 : : /* store transaction information of xact deleting the tuple */
4711 alvherre@alvh.no-ip. 3090 : 1454422 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3091 : 1454422 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3092 : 1454422 : tp.t_data->t_infomask |= new_infomask;
3093 : 1454422 : tp.t_data->t_infomask2 |= new_infomask2;
6663 tgl@sss.pgh.pa.us 3094 : 1454422 : HeapTupleHeaderClearHotUpdated(tp.t_data);
4711 alvherre@alvh.no-ip. 3095 : 1454422 : HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
6886 tgl@sss.pgh.pa.us 3096 : 1454422 : HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
3097 : : /* Make sure there is no forward chain link in t_ctid */
8527 3098 : 1454422 : tp.t_data->t_ctid = tp.t_self;
3099 : :
3100 : : /* Signal that this is actually a move into another partition */
2811 andres@anarazel.de 3101 [ + + ]: 1454422 : if (changingPart)
3102 : 492 : HeapTupleHeaderSetMovedPartitions(tp.t_data);
3103 : :
7201 tgl@sss.pgh.pa.us 3104 : 1454422 : MarkBufferDirty(buffer);
3105 : :
3106 : : /*
3107 : : * XLOG stuff
3108 : : *
3109 : : * NB: heap_abort_speculative() uses the same xlog record and replay
3110 : : * routines.
3111 : : */
5483 rhaas@postgresql.org 3112 [ + + + + : 1454422 : if (RelationNeedsWAL(relation))
+ + + + ]
3113 : : {
3114 : : xl_heap_delete xlrec;
3115 : : xl_heap_header xlhdr;
3116 : : XLogRecPtr recptr;
3117 : :
3118 : : /*
3119 : : * For logical decode we need combo CIDs to properly decode the
3120 : : * catalog
3121 : : */
4390 3122 [ + + + - : 1391803 : if (RelationIsAccessibleInLogicalDecoding(relation))
- + - - -
- + + - +
- - - - -
- ]
3123 : 6161 : log_heap_new_cid(relation, &tp);
3124 : :
2811 andres@anarazel.de 3125 : 1391803 : xlrec.flags = 0;
3126 [ + + ]: 1391803 : if (all_visible_cleared)
3127 : 355 : xlrec.flags |= XLH_DELETE_ALL_VISIBLE_CLEARED;
3128 [ + + ]: 1391803 : if (changingPart)
3129 : 492 : xlrec.flags |= XLH_DELETE_IS_PARTITION_MOVE;
4711 alvherre@alvh.no-ip. 3130 : 2783606 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
3131 : 1391803 : tp.t_data->t_infomask2);
4045 heikki.linnakangas@i 3132 : 1391803 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
4711 alvherre@alvh.no-ip. 3133 : 1391803 : xlrec.xmax = new_xmax;
3134 : :
4045 heikki.linnakangas@i 3135 [ + + ]: 1391803 : if (old_key_tuple != NULL)
3136 : : {
3137 [ + + ]: 47017 : if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
3876 andres@anarazel.de 3138 : 130 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_TUPLE;
3139 : : else
3140 : 46887 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_KEY;
3141 : : }
3142 : :
4045 heikki.linnakangas@i 3143 : 1391803 : XLogBeginInsert();
309 peter@eisentraut.org 3144 : 1391803 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
3145 : :
4045 heikki.linnakangas@i 3146 : 1391803 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3147 : :
3148 : : /*
3149 : : * Log replica identity of the deleted tuple if there is one
3150 : : */
4390 rhaas@postgresql.org 3151 [ + + ]: 1391803 : if (old_key_tuple != NULL)
3152 : : {
3153 : 47017 : xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
3154 : 47017 : xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
3155 : 47017 : xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
3156 : :
309 peter@eisentraut.org 3157 : 47017 : XLogRegisterData(&xlhdr, SizeOfHeapHeader);
4045 heikki.linnakangas@i 3158 : 47017 : XLogRegisterData((char *) old_key_tuple->t_data
3159 : : + SizeofHeapTupleHeader,
3160 : 47017 : old_key_tuple->t_len
3161 : : - SizeofHeapTupleHeader);
3162 : : }
3163 : :
3164 : : /* filtering by origin on a row level is much more efficient */
3282 andres@anarazel.de 3165 : 1391803 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
3166 : :
4045 heikki.linnakangas@i 3167 : 1391803 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
3168 : :
6366 tgl@sss.pgh.pa.us 3169 : 1391803 : PageSetLSN(page, recptr);
3170 : : }
3171 : :
9105 3172 [ - + ]: 1454422 : END_CRIT_SECTION();
3173 : :
9102 3174 : 1454422 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3175 : :
5293 rhaas@postgresql.org 3176 [ + + ]: 1454422 : if (vmbuffer != InvalidBuffer)
3177 : 355 : ReleaseBuffer(vmbuffer);
3178 : :
3179 : : /*
3180 : : * If the tuple has toasted out-of-line attributes, we need to delete
3181 : : * those items too. We have to do this before releasing the buffer
3182 : : * because we need to look at the contents of the tuple, but it's OK to
3183 : : * release the content lock on the buffer first.
3184 : : */
4672 kgrittn@postgresql.o 3185 [ + + ]: 1454422 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3186 [ + + ]: 2602 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3187 : : {
3188 : : /* toast table entries should never be recursively toasted */
6833 tgl@sss.pgh.pa.us 3189 [ - + ]: 2592 : Assert(!HeapTupleHasExternal(&tp));
3190 : : }
3191 [ + + ]: 1451830 : else if (HeapTupleHasExternal(&tp))
2266 rhaas@postgresql.org 3192 : 299 : heap_toast_delete(relation, &tp, false);
3193 : :
3194 : : /*
3195 : : * Mark tuple for invalidation from system caches at next command
3196 : : * boundary. We have to do this before releasing the buffer because we
3197 : : * need to look at the contents of the tuple.
3198 : : */
5237 tgl@sss.pgh.pa.us 3199 : 1454422 : CacheInvalidateHeapTuple(relation, &tp, NULL);
3200 : :
3201 : : /* Now we can release the buffer */
615 akorotkov@postgresql 3202 : 1454422 : ReleaseBuffer(buffer);
3203 : :
3204 : : /*
3205 : : * Release the lmgr tuple lock, if we had it.
3206 : : */
7536 tgl@sss.pgh.pa.us 3207 [ + + ]: 1454422 : if (have_tuple_lock)
4711 alvherre@alvh.no-ip. 3208 : 26 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3209 : :
6779 tgl@sss.pgh.pa.us 3210 : 1454422 : pgstat_count_heap_delete(relation);
3211 : :
4390 rhaas@postgresql.org 3212 [ + + + + ]: 1454422 : if (old_key_tuple != NULL && old_key_copied)
3213 : 46888 : heap_freetuple(old_key_tuple);
3214 : :
2461 andres@anarazel.de 3215 : 1454422 : return TM_Ok;
3216 : : }
3217 : :
3218 : : /*
3219 : : * simple_heap_delete - delete a tuple
3220 : : *
3221 : : * This routine may be used to delete a tuple when concurrent updates of
3222 : : * the target tuple are not expected (for example, because we have a lock
3223 : : * on the relation associated with the tuple). Any failure is reported
3224 : : * via ereport().
3225 : : */
3226 : : void
48 peter@eisentraut.org 3227 :GNC 645029 : simple_heap_delete(Relation relation, const ItemPointerData *tid)
3228 : : {
3229 : : TM_Result result;
3230 : : TM_FailureData tmfd;
3231 : :
8129 tgl@sss.pgh.pa.us 3232 :CBC 645029 : result = heap_delete(relation, tid,
3233 : : GetCurrentCommandId(true), InvalidSnapshot,
3234 : : true /* wait for commit */ ,
3235 : : &tmfd, false /* changingPart */ );
9094 3236 [ - + - - : 645029 : switch (result)
- ]
3237 : : {
2461 andres@anarazel.de 3238 :UBC 0 : case TM_SelfModified:
3239 : : /* Tuple was already updated in current command? */
8185 tgl@sss.pgh.pa.us 3240 [ # # ]: 0 : elog(ERROR, "tuple already updated by self");
3241 : : break;
3242 : :
2461 andres@anarazel.de 3243 :CBC 645029 : case TM_Ok:
3244 : : /* done successfully */
9094 tgl@sss.pgh.pa.us 3245 : 645029 : break;
3246 : :
2461 andres@anarazel.de 3247 :UBC 0 : case TM_Updated:
8185 tgl@sss.pgh.pa.us 3248 [ # # ]: 0 : elog(ERROR, "tuple concurrently updated");
3249 : : break;
3250 : :
2461 andres@anarazel.de 3251 : 0 : case TM_Deleted:
3252 [ # # ]: 0 : elog(ERROR, "tuple concurrently deleted");
3253 : : break;
3254 : :
9094 tgl@sss.pgh.pa.us 3255 : 0 : default:
8185 3256 [ # # ]: 0 : elog(ERROR, "unrecognized heap_delete status: %u", result);
3257 : : break;
3258 : : }
9094 tgl@sss.pgh.pa.us 3259 :CBC 645029 : }
3260 : :
3261 : : /*
3262 : : * heap_update - replace a tuple
3263 : : *
3264 : : * See table_tuple_update() for an explanation of the parameters, except that
3265 : : * this routine directly takes a tuple rather than a slot.
3266 : : *
3267 : : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
3268 : : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
3269 : : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
3270 : : * generated by another transaction).
3271 : : */
3272 : : TM_Result
48 peter@eisentraut.org 3273 :GNC 312128 : heap_update(Relation relation, const ItemPointerData *otid, HeapTuple newtup,
3274 : : CommandId cid, Snapshot crosscheck, bool wait,
3275 : : TM_FailureData *tmfd, LockTupleMode *lockmode,
3276 : : TU_UpdateIndexes *update_indexes)
3277 : : {
3278 : : TM_Result result;
7762 tgl@sss.pgh.pa.us 3279 :CBC 312128 : TransactionId xid = GetCurrentTransactionId();
3280 : : Bitmapset *hot_attrs;
3281 : : Bitmapset *sum_attrs;
3282 : : Bitmapset *key_attrs;
3283 : : Bitmapset *id_attrs;
3284 : : Bitmapset *interesting_attrs;
3285 : : Bitmapset *modified_attrs;
3286 : : ItemId lp;
3287 : : HeapTupleData oldtup;
3288 : : HeapTuple heaptup;
4390 rhaas@postgresql.org 3289 : 312128 : HeapTuple old_key_tuple = NULL;
3290 : 312128 : bool old_key_copied = false;
3291 : : Page page;
3292 : : BlockNumber block;
3293 : : MultiXactStatus mxact_status;
3294 : : Buffer buffer,
3295 : : newbuf,
5293 3296 : 312128 : vmbuffer = InvalidBuffer,
3297 : 312128 : vmbuffer_new = InvalidBuffer;
3298 : : bool need_toast;
3299 : : Size newtupsize,
3300 : : pagefree;
7536 tgl@sss.pgh.pa.us 3301 : 312128 : bool have_tuple_lock = false;
3302 : : bool iscombo;
6663 3303 : 312128 : bool use_hot_update = false;
1003 tomas.vondra@postgre 3304 : 312128 : bool summarized_update = false;
3305 : : bool key_intact;
6223 heikki.linnakangas@i 3306 : 312128 : bool all_visible_cleared = false;
3307 : 312128 : bool all_visible_cleared_new = false;
3308 : : bool checked_lockers;
3309 : : bool locker_remains;
1402 akapila@postgresql.o 3310 : 312128 : 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 : :
10328 bruce@momjian.us 3318 [ - + ]: 312128 : Assert(ItemPointerIsValid(otid));
3319 : :
3320 : : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
1682 tgl@sss.pgh.pa.us 3321 [ - + ]: 312128 : Assert(HeapTupleHeaderGetNatts(newtup->t_data) <=
3322 : : RelationGetNumberOfAttributes(relation));
3323 : :
201 nathan@postgresql.or 3324 : 312128 : 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 : : */
3884 rhaas@postgresql.org 3331 [ - + ]: 312128 : if (IsInParallelMode())
3884 rhaas@postgresql.org 3332 [ # # ]:UBC 0 : ereport(ERROR,
3333 : : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
3334 : : errmsg("cannot update tuples during a parallel operation")));
3335 : :
3336 : : #ifdef USE_ASSERT_CHECKING
449 noah@leadboat.com 3337 :CBC 312128 : 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 : : */
1003 tomas.vondra@postgre 3356 : 312128 : hot_attrs = RelationGetIndexAttrBitmap(relation,
3357 : : INDEX_ATTR_BITMAP_HOT_BLOCKING);
3358 : 312128 : sum_attrs = RelationGetIndexAttrBitmap(relation,
3359 : : INDEX_ATTR_BITMAP_SUMMARIZED);
4390 rhaas@postgresql.org 3360 : 312128 : key_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_KEY);
3361 : 312128 : id_attrs = RelationGetIndexAttrBitmap(relation,
3362 : : INDEX_ATTR_BITMAP_IDENTITY_KEY);
1482 pg@bowt.ie 3363 : 312128 : interesting_attrs = NULL;
3364 : 312128 : interesting_attrs = bms_add_members(interesting_attrs, hot_attrs);
1003 tomas.vondra@postgre 3365 : 312128 : interesting_attrs = bms_add_members(interesting_attrs, sum_attrs);
1482 pg@bowt.ie 3366 : 312128 : interesting_attrs = bms_add_members(interesting_attrs, key_attrs);
3367 : 312128 : interesting_attrs = bms_add_members(interesting_attrs, id_attrs);
3368 : :
5293 rhaas@postgresql.org 3369 : 312128 : block = ItemPointerGetBlockNumber(otid);
3370 : : INJECTION_POINT("heap_update-before-pin", NULL);
3371 : 312128 : buffer = ReadBuffer(relation, block);
3528 kgrittn@postgresql.o 3372 : 312128 : 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 : : */
5293 rhaas@postgresql.org 3380 [ + + ]: 312128 : if (PageIsAllVisible(page))
3381 : 1636 : visibilitymap_pin(relation, block, &vmbuffer);
3382 : :
9864 vadim4o@yahoo.com 3383 : 312128 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3384 : :
6366 tgl@sss.pgh.pa.us 3385 : 312128 : 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 : : */
326 noah@leadboat.com 3410 [ - + ]: 312128 : if (!ItemIdIsNormal(lp))
3411 : : {
326 noah@leadboat.com 3412 [ # # ]:UBC 0 : Assert(RelationSupportsSysCache(RelationGetRelid(relation)));
3413 : :
3414 : 0 : UnlockReleaseBuffer(buffer);
3415 [ # # ]: 0 : Assert(!have_tuple_lock);
3416 [ # # ]: 0 : if (vmbuffer != InvalidBuffer)
3417 : 0 : ReleaseBuffer(vmbuffer);
3418 : 0 : tmfd->ctid = *otid;
3419 : 0 : tmfd->xmax = InvalidTransactionId;
3420 : 0 : tmfd->cmax = InvalidCommandId;
3421 : 0 : *update_indexes = TU_None;
3422 : :
3423 : 0 : bms_free(hot_attrs);
3424 : 0 : bms_free(sum_attrs);
3425 : 0 : bms_free(key_attrs);
3426 : 0 : bms_free(id_attrs);
3427 : : /* modified_attrs not yet initialized */
3428 : 0 : bms_free(interesting_attrs);
3429 : 0 : return TM_Deleted;
3430 : : }
3431 : :
3432 : : /*
3433 : : * Fill in enough data in oldtup for HeapDetermineColumnsInfo to work
3434 : : * properly.
3435 : : */
4702 alvherre@alvh.no-ip. 3436 :CBC 312128 : oldtup.t_tableOid = RelationGetRelid(relation);
6366 tgl@sss.pgh.pa.us 3437 : 312128 : oldtup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
9882 vadim4o@yahoo.com 3438 : 312128 : oldtup.t_len = ItemIdGetLength(lp);
3439 : 312128 : oldtup.t_self = *otid;
3440 : :
3441 : : /* the new tuple is ready, except for this: */
4702 alvherre@alvh.no-ip. 3442 : 312128 : 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 : : */
1402 akapila@postgresql.o 3452 : 312128 : 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 : : */
3185 alvherre@alvh.no-ip. 3467 [ + + ]: 312128 : if (!bms_overlap(modified_attrs, key_attrs))
3468 : : {
2806 simon@2ndQuadrant.co 3469 : 307862 : *lockmode = LockTupleNoKeyExclusive;
4711 alvherre@alvh.no-ip. 3470 : 307862 : mxact_status = MultiXactStatusNoKeyUpdate;
3471 : 307862 : 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 : 307862 : MultiXactIdSetOldestMember();
3483 : : }
3484 : : else
3485 : : {
2806 simon@2ndQuadrant.co 3486 : 4266 : *lockmode = LockTupleExclusive;
4711 alvherre@alvh.no-ip. 3487 : 4266 : mxact_status = MultiXactStatusUpdate;
3488 : 4266 : 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 : :
9864 vadim4o@yahoo.com 3498 : 1 : l2:
4711 alvherre@alvh.no-ip. 3499 : 312129 : checked_lockers = false;
3500 : 312129 : locker_remains = false;
4531 rhaas@postgresql.org 3501 : 312129 : result = HeapTupleSatisfiesUpdate(&oldtup, cid, buffer);
3502 : :
3503 : : /* see below about the "no wait" case */
615 akorotkov@postgresql 3504 [ + + - + ]: 312129 : Assert(result != TM_BeingModified || wait);
3505 : :
2461 andres@anarazel.de 3506 [ - + ]: 312129 : if (result == TM_Invisible)
3507 : : {
7201 tgl@sss.pgh.pa.us 3508 :UBC 0 : UnlockReleaseBuffer(buffer);
3790 3509 [ # # ]: 0 : ereport(ERROR,
3510 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3511 : : errmsg("attempted to update invisible tuple")));
3512 : : }
615 akorotkov@postgresql 3513 [ + + + - ]:CBC 312129 : else if (result == TM_BeingModified && wait)
3514 : : {
3515 : : TransactionId xwait;
3516 : : uint16 infomask;
4711 alvherre@alvh.no-ip. 3517 : 36098 : 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 : 36098 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
7536 tgl@sss.pgh.pa.us 3534 : 36098 : 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 : : */
7538 3558 [ + + ]: 36098 : if (infomask & HEAP_XMAX_IS_MULTI)
3559 : : {
3560 : : TransactionId update_xact;
3561 : : int remain;
2374 alvherre@alvh.no-ip. 3562 : 179 : bool current_is_member = false;
3563 : :
3904 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 : : */
2374 3573 [ - + ]: 8 : if (!current_is_member)
2374 alvherre@alvh.no-ip. 3574 :UBC 0 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3575 : : LockWaitBlock, &have_tuple_lock);
3576 : :
3577 : : /* wait for multixact */
3904 alvherre@alvh.no-ip. 3578 :CBC 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) ||
3101 tgl@sss.pgh.pa.us 3592 : 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3593 : : xwait))
3904 alvherre@alvh.no-ip. 3594 :UBC 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 : : */
4711 alvherre@alvh.no-ip. 3617 [ + + ]:CBC 179 : if (!HEAP_XMAX_IS_LOCKED_ONLY(oldtup.t_data->t_infomask))
3618 : 8 : update_xact = HeapTupleGetUpdateXid(oldtup.t_data);
3619 : : else
3904 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 : : */
4711 3627 [ + + + + ]: 187 : if (!TransactionIdIsValid(update_xact) ||
3628 : 8 : TransactionIdDidAbort(update_xact))
3629 : 172 : can_continue = true;
3630 : : }
3904 3631 [ + + ]: 35919 : 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 : 35810 : checked_lockers = true;
3638 : 35810 : locker_remains = true;
3639 : 35810 : can_continue = true;
3640 : : }
3641 [ + + + + ]: 109 : 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 : 80 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2806 simon@2ndQuadrant.co 3659 : 80 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3660 : : LockWaitBlock, &have_tuple_lock);
3904 alvherre@alvh.no-ip. 3661 : 80 : XactLockTableWait(xwait, relation, &oldtup.t_self,
3662 : : XLTW_Update);
3663 : 80 : checked_lockers = true;
3664 : 80 : 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 [ + + - + ]: 159 : if (xmax_infomask_changed(oldtup.t_data->t_infomask, infomask) ||
3672 : 79 : !TransactionIdEquals(xwait,
3673 : : HeapTupleHeaderGetRawXmax(oldtup.t_data)))
3674 : 1 : goto l2;
3675 : :
3676 : : /* Otherwise check if it committed or aborted */
3677 : 79 : UpdateXmaxHintBits(oldtup.t_data, buffer, xwait);
3678 [ + + ]: 79 : if (oldtup.t_data->t_infomask & HEAP_XMAX_INVALID)
4711 3679 : 22 : can_continue = true;
3680 : : }
3681 : :
2461 andres@anarazel.de 3682 [ + + ]: 36097 : if (can_continue)
3683 : 36033 : result = TM_Ok;
1759 alvherre@alvh.no-ip. 3684 [ + + ]: 64 : else if (!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid))
2461 andres@anarazel.de 3685 : 59 : result = TM_Updated;
3686 : : else
3687 : 5 : result = TM_Deleted;
3688 : : }
3689 : :
3690 : : /* Sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3691 [ + + ]: 312128 : if (result != TM_Ok)
3692 : : {
3693 [ + + + + : 161 : Assert(result == TM_SelfModified ||
- + - - ]
3694 : : result == TM_Updated ||
3695 : : result == TM_Deleted ||
3696 : : result == TM_BeingModified);
7424 tgl@sss.pgh.pa.us 3697 [ - + ]: 161 : Assert(!(oldtup.t_data->t_infomask & HEAP_XMAX_INVALID));
2461 andres@anarazel.de 3698 [ + + - + ]: 161 : Assert(result != TM_Updated ||
3699 : : !ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
3700 : : }
3701 : :
750 heikki.linnakangas@i 3702 [ + + + - ]: 312128 : 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 [ + + ]: 312128 : if (result != TM_Ok)
3710 : : {
2461 andres@anarazel.de 3711 : 162 : tmfd->ctid = oldtup.t_data->t_ctid;
3712 : 162 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(oldtup.t_data);
3713 [ + + ]: 162 : if (result == TM_SelfModified)
3714 : 52 : tmfd->cmax = HeapTupleHeaderGetCmax(oldtup.t_data);
3715 : : else
3716 : 110 : tmfd->cmax = InvalidCommandId;
615 akorotkov@postgresql 3717 : 162 : UnlockReleaseBuffer(buffer);
7536 tgl@sss.pgh.pa.us 3718 [ + + ]: 162 : if (have_tuple_lock)
2806 simon@2ndQuadrant.co 3719 : 57 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
5293 rhaas@postgresql.org 3720 [ - + ]: 162 : if (vmbuffer != InvalidBuffer)
5293 rhaas@postgresql.org 3721 :UBC 0 : ReleaseBuffer(vmbuffer);
1003 tomas.vondra@postgre 3722 :CBC 162 : *update_indexes = TU_None;
3723 : :
6663 tgl@sss.pgh.pa.us 3724 : 162 : bms_free(hot_attrs);
1003 tomas.vondra@postgre 3725 : 162 : bms_free(sum_attrs);
4711 alvherre@alvh.no-ip. 3726 : 162 : bms_free(key_attrs);
3402 tgl@sss.pgh.pa.us 3727 : 162 : bms_free(id_attrs);
3185 alvherre@alvh.no-ip. 3728 : 162 : bms_free(modified_attrs);
3729 : 162 : bms_free(interesting_attrs);
9864 vadim4o@yahoo.com 3730 : 162 : 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 : : */
5287 rhaas@postgresql.org 3742 [ + + - + ]: 311966 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3743 : : {
5287 rhaas@postgresql.org 3744 :UBC 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3745 : 0 : visibilitymap_pin(relation, block, &vmbuffer);
3746 : 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
5195 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 : : */
4711 alvherre@alvh.no-ip. 3756 :CBC 311966 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3757 : 311966 : oldtup.t_data->t_infomask,
3758 : 311966 : 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 [ + + + - ]: 347977 : if ((oldtup.t_data->t_infomask & HEAP_XMAX_INVALID) ||
3463 3771 [ + + ]: 72022 : HEAP_LOCKED_UPGRADED(oldtup.t_data->t_infomask) ||
4711 3772 [ - + ]: 35840 : (checked_lockers && !locker_remains))
3773 : 275955 : xmax_new_tuple = InvalidTransactionId;
3774 : : else
3775 : 36011 : xmax_new_tuple = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3776 : :
3777 [ + + ]: 311966 : if (!TransactionIdIsValid(xmax_new_tuple))
3778 : : {
3779 : 275955 : infomask_new_tuple = HEAP_XMAX_INVALID;
3780 : 275955 : 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 [ + + ]: 36011 : 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 : 35839 : infomask_new_tuple = HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_LOCK_ONLY;
3798 : 35839 : 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 : : */
9882 vadim4o@yahoo.com 3806 : 311966 : newtup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
6663 tgl@sss.pgh.pa.us 3807 : 311966 : newtup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
7762 3808 : 311966 : HeapTupleHeaderSetXmin(newtup->t_data, xid);
8586 bruce@momjian.us 3809 : 311966 : HeapTupleHeaderSetCmin(newtup->t_data, cid);
4711 alvherre@alvh.no-ip. 3810 : 311966 : newtup->t_data->t_infomask |= HEAP_UPDATED | infomask_new_tuple;
3811 : 311966 : newtup->t_data->t_infomask2 |= infomask2_new_tuple;
3812 : 311966 : 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 : : */
6886 tgl@sss.pgh.pa.us 3818 : 311966 : 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 : : */
4672 kgrittn@postgresql.o 3831 [ - + ]: 311966 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
4672 kgrittn@postgresql.o 3832 [ # # ]:UBC 0 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3833 : : {
3834 : : /* toast table entries should never be recursively toasted */
6833 tgl@sss.pgh.pa.us 3835 [ # # ]: 0 : Assert(!HeapTupleHasExternal(&oldtup));
3836 [ # # ]: 0 : Assert(!HeapTupleHasExternal(newtup));
3837 : 0 : need_toast = false;
3838 : : }
3839 : : else
6833 tgl@sss.pgh.pa.us 3840 [ + + ]:CBC 935512 : need_toast = (HeapTupleHasExternal(&oldtup) ||
3841 [ + + ]: 623546 : HeapTupleHasExternal(newtup) ||
3842 [ + + ]: 311556 : newtup->t_len > TOAST_TUPLE_THRESHOLD);
3843 : :
6366 3844 : 311966 : pagefree = PageGetHeapFreeSpace(page);
3845 : :
6891 3846 : 311966 : newtupsize = MAXALIGN(newtup->t_len);
3847 : :
8981 3848 [ + + + + ]: 311966 : if (need_toast || newtupsize > pagefree)
9232 vadim4o@yahoo.com 3849 : 152128 : {
3850 : : TransactionId xmax_lock_old_tuple;
3851 : : uint16 infomask_lock_old_tuple,
3852 : : infomask2_lock_old_tuple;
3439 andres@anarazel.de 3853 : 152128 : 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 : : */
3442 3873 : 152128 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3874 : 152128 : oldtup.t_data->t_infomask,
3875 : 152128 : 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 [ - + ]: 152128 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(infomask_lock_old_tuple));
3881 : :
3882 : 152128 : START_CRIT_SECTION();
3883 : :
3884 : : /* Clear obsolete visibility flags ... */
4711 alvherre@alvh.no-ip. 3885 : 152128 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3886 : 152128 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6663 tgl@sss.pgh.pa.us 3887 : 152128 : HeapTupleClearHotUpdated(&oldtup);
3888 : : /* ... and store info about transaction updating this tuple */
3442 andres@anarazel.de 3889 [ - + ]: 152128 : Assert(TransactionIdIsValid(xmax_lock_old_tuple));
3890 : 152128 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_lock_old_tuple);
3891 : 152128 : oldtup.t_data->t_infomask |= infomask_lock_old_tuple;
3892 : 152128 : oldtup.t_data->t_infomask2 |= infomask2_lock_old_tuple;
6886 tgl@sss.pgh.pa.us 3893 : 152128 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
3894 : :
3895 : : /* temporarily make it look not-updated, but locked */
7332 3896 : 152128 : oldtup.t_data->t_ctid = oldtup.t_self;
3897 : :
3898 : : /*
3899 : : * Clear all-frozen bit on visibility map if needed. We could
3900 : : * immediately reset ALL_VISIBLE, but given that the WAL logging
3901 : : * overhead would be unchanged, that doesn't seem necessarily
3902 : : * worthwhile.
3903 : : */
1709 3904 [ + + + + ]: 153054 : if (PageIsAllVisible(page) &&
3439 andres@anarazel.de 3905 : 926 : visibilitymap_clear(relation, block, vmbuffer,
3906 : : VISIBILITYMAP_ALL_FROZEN))
3907 : 719 : cleared_all_frozen = true;
3908 : :
3442 3909 : 152128 : MarkBufferDirty(buffer);
3910 : :
3911 [ + + + + : 152128 : if (RelationNeedsWAL(relation))
+ - + + ]
3912 : : {
3913 : : xl_heap_lock xlrec;
3914 : : XLogRecPtr recptr;
3915 : :
3916 : 141998 : XLogBeginInsert();
3917 : 141998 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3918 : :
3919 : 141998 : xlrec.offnum = ItemPointerGetOffsetNumber(&oldtup.t_self);
981 pg@bowt.ie 3920 : 141998 : xlrec.xmax = xmax_lock_old_tuple;
3442 andres@anarazel.de 3921 : 283996 : xlrec.infobits_set = compute_infobits(oldtup.t_data->t_infomask,
3922 : 141998 : oldtup.t_data->t_infomask2);
3439 3923 : 141998 : xlrec.flags =
3924 : 141998 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
309 peter@eisentraut.org 3925 : 141998 : XLogRegisterData(&xlrec, SizeOfHeapLock);
3442 andres@anarazel.de 3926 : 141998 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
3927 : 141998 : PageSetLSN(page, recptr);
3928 : : }
3929 : :
3930 [ - + ]: 152128 : END_CRIT_SECTION();
3931 : :
9232 vadim4o@yahoo.com 3932 : 152128 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3933 : :
3934 : : /*
3935 : : * Let the toaster do its thing, if needed.
3936 : : *
3937 : : * Note: below this point, heaptup is the data we actually intend to
3938 : : * store into the relation; newtup is the caller's original untoasted
3939 : : * data.
3940 : : */
9102 tgl@sss.pgh.pa.us 3941 [ + + ]: 152128 : if (need_toast)
3942 : : {
3943 : : /* Note we always use WAL and FSM during updates */
2266 rhaas@postgresql.org 3944 : 1773 : heaptup = heap_toast_insert_or_update(relation, newtup, &oldtup, 0);
7332 tgl@sss.pgh.pa.us 3945 : 1773 : newtupsize = MAXALIGN(heaptup->t_len);
3946 : : }
3947 : : else
3948 : 150355 : heaptup = newtup;
3949 : :
3950 : : /*
3951 : : * Now, do we need a new page for the tuple, or not? This is a bit
3952 : : * tricky since someone else could have added tuples to the page while
3953 : : * we weren't looking. We have to recheck the available space after
3954 : : * reacquiring the buffer lock. But don't bother to do that if the
3955 : : * former amount of free space is still not enough; it's unlikely
3956 : : * there's more free now than before.
3957 : : *
3958 : : * What's more, if we need to get a new page, we will need to acquire
3959 : : * buffer locks on both old and new pages. To avoid deadlock against
3960 : : * some other backend trying to get the same two locks in the other
3961 : : * order, we must be consistent about the order we get the locks in.
3962 : : * We use the rule "lock the lower-numbered page of the relation
3963 : : * first". To implement this, we must do RelationGetBufferForTuple
3964 : : * while not holding the lock on the old page, and we must rely on it
3965 : : * to get the locks on both pages in the correct order.
3966 : : *
3967 : : * Another consideration is that we need visibility map page pin(s) if
3968 : : * we will have to clear the all-visible flag on either page. If we
3969 : : * call RelationGetBufferForTuple, we rely on it to acquire any such
3970 : : * pins; but if we don't, we have to handle that here. Hence we need
3971 : : * a loop.
3972 : : */
3973 : : for (;;)
3974 : : {
1709 3975 [ + + ]: 152128 : if (newtupsize > pagefree)
3976 : : {
3977 : : /* It doesn't fit, must use RelationGetBufferForTuple. */
3978 : 151586 : newbuf = RelationGetBufferForTuple(relation, heaptup->t_len,
3979 : : buffer, 0, NULL,
3980 : : &vmbuffer_new, &vmbuffer,
3981 : : 0);
3982 : : /* We're all done. */
3983 : 151586 : break;
3984 : : }
3985 : : /* Acquire VM page pin if needed and we don't have it. */
3986 [ + + - + ]: 542 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
1709 tgl@sss.pgh.pa.us 3987 :UBC 0 : visibilitymap_pin(relation, block, &vmbuffer);
3988 : : /* Re-acquire the lock on the old tuple's page. */
8981 tgl@sss.pgh.pa.us 3989 :CBC 542 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3990 : : /* Re-check using the up-to-date free space */
6366 3991 : 542 : pagefree = PageGetHeapFreeSpace(page);
1709 3992 [ + - ]: 542 : if (newtupsize > pagefree ||
3993 [ + + - + ]: 542 : (vmbuffer == InvalidBuffer && PageIsAllVisible(page)))
3994 : : {
3995 : : /*
3996 : : * Rats, it doesn't fit anymore, or somebody just now set the
3997 : : * all-visible flag. We must now unlock and loop to avoid
3998 : : * deadlock. Fortunately, this path should seldom be taken.
3999 : : */
8981 tgl@sss.pgh.pa.us 4000 :UBC 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4001 : : }
4002 : : else
4003 : : {
4004 : : /* We're all done. */
8981 tgl@sss.pgh.pa.us 4005 :CBC 542 : newbuf = buffer;
1709 4006 : 542 : break;
4007 : : }
4008 : : }
4009 : : }
4010 : : else
4011 : : {
4012 : : /* No TOAST work needed, and it'll fit on same page */
9102 4013 : 159838 : newbuf = buffer;
7332 4014 : 159838 : heaptup = newtup;
4015 : : }
4016 : :
4017 : : /*
4018 : : * We're about to do the actual update -- check for conflict first, to
4019 : : * avoid possibly having to roll back work we've just done.
4020 : : *
4021 : : * This is safe without a recheck as long as there is no possibility of
4022 : : * another process scanning the pages between this check and the update
4023 : : * being visible to the scan (i.e., exclusive buffer content lock(s) are
4024 : : * continuously held from this point until the tuple update is visible).
4025 : : *
4026 : : * For the new tuple the only check needed is at the relation level, but
4027 : : * since both tuples are in the same relation and the check for oldtup
4028 : : * will include checking the relation level, there is no benefit to a
4029 : : * separate check for the new tuple.
4030 : : */
1709 tmunro@postgresql.or 4031 : 311966 : CheckForSerializableConflictIn(relation, &oldtup.t_self,
4032 : : BufferGetBlockNumber(buffer));
4033 : :
4034 : : /*
4035 : : * At this point newbuf and buffer are both pinned and locked, and newbuf
4036 : : * has enough space for the new tuple. If they are the same buffer, only
4037 : : * one pin is held.
4038 : : */
4039 : :
6663 tgl@sss.pgh.pa.us 4040 [ + + ]: 311954 : if (newbuf == buffer)
4041 : : {
4042 : : /*
4043 : : * Since the new tuple is going into the same page, we might be able
4044 : : * to do a HOT update. Check if any of the index columns have been
4045 : : * changed.
4046 : : */
1482 pg@bowt.ie 4047 [ + + ]: 160368 : if (!bms_overlap(modified_attrs, hot_attrs))
4048 : : {
6663 tgl@sss.pgh.pa.us 4049 : 148133 : use_hot_update = true;
4050 : :
4051 : : /*
4052 : : * If none of the columns that are used in hot-blocking indexes
4053 : : * were updated, we can apply HOT, but we do still need to check
4054 : : * if we need to update the summarizing indexes, and update those
4055 : : * indexes if the columns were updated, or we may fail to detect
4056 : : * e.g. value bound changes in BRIN minmax indexes.
4057 : : */
1003 tomas.vondra@postgre 4058 [ + + ]: 148133 : if (bms_overlap(modified_attrs, sum_attrs))
4059 : 1641 : summarized_update = true;
4060 : : }
4061 : : }
4062 : : else
4063 : : {
4064 : : /* Set a hint that the old page could use prune/defrag */
6366 tgl@sss.pgh.pa.us 4065 : 151586 : PageSetFull(page);
4066 : : }
4067 : :
4068 : : /*
4069 : : * Compute replica identity tuple before entering the critical section so
4070 : : * we don't PANIC upon a memory allocation failure.
4071 : : * ExtractReplicaIdentity() will return NULL if nothing needs to be
4072 : : * logged. Pass old key required as true only if the replica identity key
4073 : : * columns are modified or it has external data.
4074 : : */
3185 alvherre@alvh.no-ip. 4075 : 311954 : old_key_tuple = ExtractReplicaIdentity(relation, &oldtup,
1402 akapila@postgresql.o 4076 [ + + + + ]: 311954 : bms_overlap(modified_attrs, id_attrs) ||
4077 : : id_has_external,
4078 : : &old_key_copied);
4079 : :
4080 : : /* NO EREPORT(ERROR) from here till changes are logged */
9105 tgl@sss.pgh.pa.us 4081 : 311954 : START_CRIT_SECTION();
4082 : :
4083 : : /*
4084 : : * If this transaction commits, the old tuple will become DEAD sooner or
4085 : : * later. Set flag that this page is a candidate for pruning once our xid
4086 : : * falls below the OldestXmin horizon. If the transaction finally aborts,
4087 : : * the subsequent page pruning will be a no-op and the hint will be
4088 : : * cleared.
4089 : : *
4090 : : * XXX Should we set hint on newbuf as well? If the transaction aborts,
4091 : : * there would be a prunable tuple in the newbuf; but for now we choose
4092 : : * not to optimize for aborts. Note that heap_xlog_update must be kept in
4093 : : * sync if this decision changes.
4094 : : */
6366 4095 [ - + + + : 311954 : PageSetPrunable(page, xid);
+ + ]
4096 : :
6663 4097 [ + + ]: 311954 : if (use_hot_update)
4098 : : {
4099 : : /* Mark the old tuple as HOT-updated */
4100 : 148133 : HeapTupleSetHotUpdated(&oldtup);
4101 : : /* And mark the new tuple as heap-only */
4102 : 148133 : HeapTupleSetHeapOnly(heaptup);
4103 : : /* Mark the caller's copy too, in case different from heaptup */
4104 : 148133 : HeapTupleSetHeapOnly(newtup);
4105 : : }
4106 : : else
4107 : : {
4108 : : /* Make sure tuples are correctly marked as not-HOT */
4109 : 163821 : HeapTupleClearHotUpdated(&oldtup);
4110 : 163821 : HeapTupleClearHeapOnly(heaptup);
4111 : 163821 : HeapTupleClearHeapOnly(newtup);
4112 : : }
4113 : :
3101 4114 : 311954 : RelationPutHeapTuple(relation, newbuf, heaptup, false); /* insert new tuple */
4115 : :
4116 : :
4117 : : /* Clear obsolete visibility flags, possibly set by ourselves above... */
3442 andres@anarazel.de 4118 : 311954 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
4119 : 311954 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
4120 : : /* ... and store info about transaction updating this tuple */
4121 [ - + ]: 311954 : Assert(TransactionIdIsValid(xmax_old_tuple));
4122 : 311954 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_old_tuple);
4123 : 311954 : oldtup.t_data->t_infomask |= infomask_old_tuple;
4124 : 311954 : oldtup.t_data->t_infomask2 |= infomask2_old_tuple;
4125 : 311954 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
4126 : :
4127 : : /* record address of new tuple in t_ctid of old one */
7332 tgl@sss.pgh.pa.us 4128 : 311954 : oldtup.t_data->t_ctid = heaptup->t_self;
4129 : :
4130 : : /* clear PD_ALL_VISIBLE flags, reset all visibilitymap bits */
3528 kgrittn@postgresql.o 4131 [ + + ]: 311954 : if (PageIsAllVisible(BufferGetPage(buffer)))
4132 : : {
5959 tgl@sss.pgh.pa.us 4133 : 1635 : all_visible_cleared = true;
3528 kgrittn@postgresql.o 4134 : 1635 : PageClearAllVisible(BufferGetPage(buffer));
5287 rhaas@postgresql.org 4135 : 1635 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
4136 : : vmbuffer, VISIBILITYMAP_VALID_BITS);
4137 : : }
3528 kgrittn@postgresql.o 4138 [ + + + + ]: 311954 : if (newbuf != buffer && PageIsAllVisible(BufferGetPage(newbuf)))
4139 : : {
5959 tgl@sss.pgh.pa.us 4140 : 1016 : all_visible_cleared_new = true;
3528 kgrittn@postgresql.o 4141 : 1016 : PageClearAllVisible(BufferGetPage(newbuf));
5287 rhaas@postgresql.org 4142 : 1016 : visibilitymap_clear(relation, BufferGetBlockNumber(newbuf),
4143 : : vmbuffer_new, VISIBILITYMAP_VALID_BITS);
4144 : : }
4145 : :
7201 tgl@sss.pgh.pa.us 4146 [ + + ]: 311954 : if (newbuf != buffer)
4147 : 151586 : MarkBufferDirty(newbuf);
4148 : 311954 : MarkBufferDirty(buffer);
4149 : :
4150 : : /* XLOG stuff */
5483 rhaas@postgresql.org 4151 [ + + + + : 311954 : if (RelationNeedsWAL(relation))
+ + + + ]
4152 : : {
4153 : : XLogRecPtr recptr;
4154 : :
4155 : : /*
4156 : : * For logical decoding we need combo CIDs to properly decode the
4157 : : * catalog.
4158 : : */
4390 4159 [ + + + - : 300589 : if (RelationIsAccessibleInLogicalDecoding(relation))
- + - - -
- + + - +
- - - - -
- ]
4160 : : {
4161 : 2540 : log_heap_new_cid(relation, &oldtup);
4162 : 2540 : log_heap_new_cid(relation, heaptup);
4163 : : }
4164 : :
4165 : 300589 : recptr = log_heap_update(relation, buffer,
4166 : : newbuf, &oldtup, heaptup,
4167 : : old_key_tuple,
4168 : : all_visible_cleared,
4169 : : all_visible_cleared_new);
9298 vadim4o@yahoo.com 4170 [ + + ]: 300589 : if (newbuf != buffer)
4171 : : {
3528 kgrittn@postgresql.o 4172 : 141462 : PageSetLSN(BufferGetPage(newbuf), recptr);
4173 : : }
4174 : 300589 : PageSetLSN(BufferGetPage(buffer), recptr);
4175 : : }
4176 : :
9105 tgl@sss.pgh.pa.us 4177 [ - + ]: 311954 : END_CRIT_SECTION();
4178 : :
9298 vadim4o@yahoo.com 4179 [ + + ]: 311954 : if (newbuf != buffer)
4180 : 151586 : LockBuffer(newbuf, BUFFER_LOCK_UNLOCK);
9864 4181 : 311954 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4182 : :
4183 : : /*
4184 : : * Mark old tuple for invalidation from system caches at next command
4185 : : * boundary, and mark the new tuple for invalidation in case we abort. We
4186 : : * have to do this before releasing the buffer because oldtup is in the
4187 : : * buffer. (heaptup is all in local memory, but it's necessary to process
4188 : : * both tuple versions in one call to inval.c so we can avoid redundant
4189 : : * sinval messages.)
4190 : : */
5237 tgl@sss.pgh.pa.us 4191 : 311954 : CacheInvalidateHeapTuple(relation, &oldtup, heaptup);
4192 : :
4193 : : /* Now we can release the buffer(s) */
9110 4194 [ + + ]: 311954 : if (newbuf != buffer)
7201 4195 : 151586 : ReleaseBuffer(newbuf);
615 akorotkov@postgresql 4196 : 311954 : ReleaseBuffer(buffer);
5293 rhaas@postgresql.org 4197 [ + + ]: 311954 : if (BufferIsValid(vmbuffer_new))
4198 : 1019 : ReleaseBuffer(vmbuffer_new);
4199 [ + + ]: 311954 : if (BufferIsValid(vmbuffer))
4200 : 1636 : ReleaseBuffer(vmbuffer);
4201 : :
4202 : : /*
4203 : : * Release the lmgr tuple lock, if we had it.
4204 : : */
7536 tgl@sss.pgh.pa.us 4205 [ + + ]: 311954 : if (have_tuple_lock)
2806 simon@2ndQuadrant.co 4206 : 22 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
4207 : :
1000 pg@bowt.ie 4208 : 311954 : pgstat_count_heap_update(relation, use_hot_update, newbuf != buffer);
4209 : :
4210 : : /*
4211 : : * If heaptup is a private copy, release it. Don't forget to copy t_self
4212 : : * back to the caller's image, too.
4213 : : */
7332 tgl@sss.pgh.pa.us 4214 [ + + ]: 311954 : if (heaptup != newtup)
4215 : : {
4216 : 1725 : newtup->t_self = heaptup->t_self;
4217 : 1725 : heap_freetuple(heaptup);
4218 : : }
4219 : :
4220 : : /*
4221 : : * If it is a HOT update, the update may still need to update summarized
4222 : : * indexes, lest we fail to update those summaries and get incorrect
4223 : : * results (for example, minmax bounds of the block may change with this
4224 : : * update).
4225 : : */
1003 tomas.vondra@postgre 4226 [ + + ]: 311954 : if (use_hot_update)
4227 : : {
4228 [ + + ]: 148133 : if (summarized_update)
4229 : 1641 : *update_indexes = TU_Summarizing;
4230 : : else
4231 : 146492 : *update_indexes = TU_None;
4232 : : }
4233 : : else
4234 : 163821 : *update_indexes = TU_All;
4235 : :
4390 rhaas@postgresql.org 4236 [ + + + + ]: 311954 : if (old_key_tuple != NULL && old_key_copied)
4237 : 84 : heap_freetuple(old_key_tuple);
4238 : :
6663 tgl@sss.pgh.pa.us 4239 : 311954 : bms_free(hot_attrs);
1003 tomas.vondra@postgre 4240 : 311954 : bms_free(sum_attrs);
4711 alvherre@alvh.no-ip. 4241 : 311954 : bms_free(key_attrs);
3402 tgl@sss.pgh.pa.us 4242 : 311954 : bms_free(id_attrs);
3185 alvherre@alvh.no-ip. 4243 : 311954 : bms_free(modified_attrs);
4244 : 311954 : bms_free(interesting_attrs);
4245 : :
2461 andres@anarazel.de 4246 : 311954 : return TM_Ok;
4247 : : }
4248 : :
4249 : : #ifdef USE_ASSERT_CHECKING
4250 : : /*
4251 : : * Confirm adequate lock held during heap_update(), per rules from
4252 : : * README.tuplock section "Locking to write inplace-updated tables".
4253 : : */
4254 : : static void
449 noah@leadboat.com 4255 : 312128 : check_lock_if_inplace_updateable_rel(Relation relation,
4256 : : const ItemPointerData *otid,
4257 : : HeapTuple newtup)
4258 : : {
4259 : : /* LOCKTAG_TUPLE acceptable for any catalog */
4260 [ + + ]: 312128 : switch (RelationGetRelid(relation))
4261 : : {
4262 : 70105 : case RelationRelationId:
4263 : : case DatabaseRelationId:
4264 : : {
4265 : : LOCKTAG tuptag;
4266 : :
4267 : 70105 : SET_LOCKTAG_TUPLE(tuptag,
4268 : : relation->rd_lockInfo.lockRelId.dbId,
4269 : : relation->rd_lockInfo.lockRelId.relId,
4270 : : ItemPointerGetBlockNumber(otid),
4271 : : ItemPointerGetOffsetNumber(otid));
4272 [ + + ]: 70105 : if (LockHeldByMe(&tuptag, InplaceUpdateTupleLock, false))
4273 : 31135 : return;
4274 : : }
4275 : 38970 : break;
4276 : 242023 : default:
4277 [ - + ]: 242023 : Assert(!IsInplaceUpdateRelation(relation));
4278 : 242023 : return;
4279 : : }
4280 : :
4281 [ + - - ]: 38970 : switch (RelationGetRelid(relation))
4282 : : {
4283 : 38970 : case RelationRelationId:
4284 : : {
4285 : : /* LOCKTAG_TUPLE or LOCKTAG_RELATION ok */
4286 : 38970 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(newtup);
4287 : 38970 : Oid relid = classForm->oid;
4288 : : Oid dbid;
4289 : : LOCKTAG tag;
4290 : :
4291 [ + + ]: 38970 : if (IsSharedRelation(relid))
4292 : 44 : dbid = InvalidOid;
4293 : : else
4294 : 38926 : dbid = MyDatabaseId;
4295 : :
4296 [ + + ]: 38970 : if (classForm->relkind == RELKIND_INDEX)
4297 : : {
4298 : 996 : Relation irel = index_open(relid, AccessShareLock);
4299 : :
4300 : 996 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4301 : 996 : index_close(irel, AccessShareLock);
4302 : : }
4303 : : else
4304 : 37974 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4305 : :
4306 [ + + ]: 38970 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, false) &&
4307 [ - + ]: 35433 : !LockHeldByMe(&tag, ShareRowExclusiveLock, true))
449 noah@leadboat.com 4308 [ # # ]:UBC 0 : elog(WARNING,
4309 : : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4310 : : NameStr(classForm->relname),
4311 : : relid,
4312 : : classForm->relkind,
4313 : : ItemPointerGetBlockNumber(otid),
4314 : : ItemPointerGetOffsetNumber(otid));
4315 : : }
449 noah@leadboat.com 4316 :CBC 38970 : break;
449 noah@leadboat.com 4317 :UBC 0 : case DatabaseRelationId:
4318 : : {
4319 : : /* LOCKTAG_TUPLE required */
4320 : 0 : Form_pg_database dbForm = (Form_pg_database) GETSTRUCT(newtup);
4321 : :
4322 [ # # ]: 0 : elog(WARNING,
4323 : : "missing lock on database \"%s\" (OID %u) @ TID (%u,%u)",
4324 : : NameStr(dbForm->datname),
4325 : : dbForm->oid,
4326 : : ItemPointerGetBlockNumber(otid),
4327 : : ItemPointerGetOffsetNumber(otid));
4328 : : }
4329 : 0 : break;
4330 : : }
4331 : : }
4332 : :
4333 : : /*
4334 : : * Confirm adequate relation lock held, per rules from README.tuplock section
4335 : : * "Locking to write inplace-updated tables".
4336 : : */
4337 : : static void
449 noah@leadboat.com 4338 :CBC 93207 : check_inplace_rel_lock(HeapTuple oldtup)
4339 : : {
4340 : 93207 : Form_pg_class classForm = (Form_pg_class) GETSTRUCT(oldtup);
4341 : 93207 : Oid relid = classForm->oid;
4342 : : Oid dbid;
4343 : : LOCKTAG tag;
4344 : :
4345 [ + + ]: 93207 : if (IsSharedRelation(relid))
4346 : 8859 : dbid = InvalidOid;
4347 : : else
4348 : 84348 : dbid = MyDatabaseId;
4349 : :
4350 [ + + ]: 93207 : if (classForm->relkind == RELKIND_INDEX)
4351 : : {
4352 : 40499 : Relation irel = index_open(relid, AccessShareLock);
4353 : :
4354 : 40499 : SET_LOCKTAG_RELATION(tag, dbid, irel->rd_index->indrelid);
4355 : 40499 : index_close(irel, AccessShareLock);
4356 : : }
4357 : : else
4358 : 52708 : SET_LOCKTAG_RELATION(tag, dbid, relid);
4359 : :
4360 [ - + ]: 93207 : if (!LockHeldByMe(&tag, ShareUpdateExclusiveLock, true))
449 noah@leadboat.com 4361 [ # # ]:UBC 0 : elog(WARNING,
4362 : : "missing lock for relation \"%s\" (OID %u, relkind %c) @ TID (%u,%u)",
4363 : : NameStr(classForm->relname),
4364 : : relid,
4365 : : classForm->relkind,
4366 : : ItemPointerGetBlockNumber(&oldtup->t_self),
4367 : : ItemPointerGetOffsetNumber(&oldtup->t_self));
449 noah@leadboat.com 4368 :CBC 93207 : }
4369 : : #endif
4370 : :
4371 : : /*
4372 : : * Check if the specified attribute's values are the same. Subroutine for
4373 : : * HeapDetermineColumnsInfo.
4374 : : */
4375 : : static bool
1402 akapila@postgresql.o 4376 : 765911 : heap_attr_equals(TupleDesc tupdesc, int attrnum, Datum value1, Datum value2,
4377 : : bool isnull1, bool isnull2)
4378 : : {
4379 : : /*
4380 : : * If one value is NULL and other is not, then they are certainly not
4381 : : * equal
4382 : : */
6663 tgl@sss.pgh.pa.us 4383 [ + + ]: 765911 : if (isnull1 != isnull2)
4384 : 45 : return false;
4385 : :
4386 : : /*
4387 : : * If both are NULL, they can be considered equal.
4388 : : */
4389 [ + + ]: 765866 : if (isnull1)
4390 : 4991 : return true;
4391 : :
4392 : : /*
4393 : : * We do simple binary comparison of the two datums. This may be overly
4394 : : * strict because there can be multiple binary representations for the
4395 : : * same logical value. But we should be OK as long as there are no false
4396 : : * positives. Using a type-specific equality operator is messy because
4397 : : * there could be multiple notions of equality in different operator
4398 : : * classes; furthermore, we cannot safely invoke user-defined functions
4399 : : * while holding exclusive buffer lock.
4400 : : */
4401 [ - + ]: 760875 : if (attrnum <= 0)
4402 : : {
4403 : : /* The only allowed system columns are OIDs, so do this */
6663 tgl@sss.pgh.pa.us 4404 :UBC 0 : return (DatumGetObjectId(value1) == DatumGetObjectId(value2));
4405 : : }
4406 : : else
4407 : : {
4408 : : CompactAttribute *att;
4409 : :
6663 tgl@sss.pgh.pa.us 4410 [ - + ]:CBC 760875 : Assert(attrnum <= tupdesc->natts);
362 drowley@postgresql.o 4411 : 760875 : att = TupleDescCompactAttr(tupdesc, attrnum - 1);
6663 tgl@sss.pgh.pa.us 4412 : 760875 : return datumIsEqual(value1, value2, att->attbyval, att->attlen);
4413 : : }
4414 : : }
4415 : :
4416 : : /*
4417 : : * Check which columns are being updated.
4418 : : *
4419 : : * Given an updated tuple, determine (and return into the output bitmapset),
4420 : : * from those listed as interesting, the set of columns that changed.
4421 : : *
4422 : : * has_external indicates if any of the unmodified attributes (from those
4423 : : * listed as interesting) of the old tuple is a member of external_cols and is
4424 : : * stored externally.
4425 : : */
4426 : : static Bitmapset *
1402 akapila@postgresql.o 4427 : 312128 : HeapDetermineColumnsInfo(Relation relation,
4428 : : Bitmapset *interesting_cols,
4429 : : Bitmapset *external_cols,
4430 : : HeapTuple oldtup, HeapTuple newtup,
4431 : : bool *has_external)
4432 : : {
4433 : : int attidx;
3136 bruce@momjian.us 4434 : 312128 : Bitmapset *modified = NULL;
1402 akapila@postgresql.o 4435 : 312128 : TupleDesc tupdesc = RelationGetDescr(relation);
4436 : :
1021 tgl@sss.pgh.pa.us 4437 : 312128 : attidx = -1;
4438 [ + + ]: 1078039 : while ((attidx = bms_next_member(interesting_cols, attidx)) >= 0)
4439 : : {
4440 : : /* attidx is zero-based, attrnum is the normal attribute number */
4441 : 765911 : AttrNumber attrnum = attidx + FirstLowInvalidHeapAttributeNumber;
4442 : : Datum value1,
4443 : : value2;
4444 : : bool isnull1,
4445 : : isnull2;
4446 : :
4447 : : /*
4448 : : * If it's a whole-tuple reference, say "not equal". It's not really
4449 : : * worth supporting this case, since it could only succeed after a
4450 : : * no-op update, which is hardly a case worth optimizing for.
4451 : : */
1402 akapila@postgresql.o 4452 [ - + ]: 765911 : if (attrnum == 0)
4453 : : {
1021 tgl@sss.pgh.pa.us 4454 :UBC 0 : modified = bms_add_member(modified, attidx);
1402 akapila@postgresql.o 4455 :CBC 733573 : continue;
4456 : : }
4457 : :
4458 : : /*
4459 : : * Likewise, automatically say "not equal" for any system attribute
4460 : : * other than tableOID; we cannot expect these to be consistent in a
4461 : : * HOT chain, or even to be set correctly yet in the new tuple.
4462 : : */
4463 [ - + ]: 765911 : if (attrnum < 0)
4464 : : {
1402 akapila@postgresql.o 4465 [ # # ]:UBC 0 : if (attrnum != TableOidAttributeNumber)
4466 : : {
1021 tgl@sss.pgh.pa.us 4467 : 0 : modified = bms_add_member(modified, attidx);
1402 akapila@postgresql.o 4468 : 0 : continue;
4469 : : }
4470 : : }
4471 : :
4472 : : /*
4473 : : * Extract the corresponding values. XXX this is pretty inefficient
4474 : : * if there are many indexed columns. Should we do a single
4475 : : * heap_deform_tuple call on each tuple, instead? But that doesn't
4476 : : * work for system columns ...
4477 : : */
1402 akapila@postgresql.o 4478 :CBC 765911 : value1 = heap_getattr(oldtup, attrnum, tupdesc, &isnull1);
4479 : 765911 : value2 = heap_getattr(newtup, attrnum, tupdesc, &isnull2);
4480 : :
4481 [ + + ]: 765911 : if (!heap_attr_equals(tupdesc, attrnum, value1,
4482 : : value2, isnull1, isnull2))
4483 : : {
1021 tgl@sss.pgh.pa.us 4484 : 27106 : modified = bms_add_member(modified, attidx);
1402 akapila@postgresql.o 4485 : 27106 : continue;
4486 : : }
4487 : :
4488 : : /*
4489 : : * No need to check attributes that can't be stored externally. Note
4490 : : * that system attributes can't be stored externally.
4491 : : */
4492 [ + - + + ]: 738805 : if (attrnum < 0 || isnull1 ||
362 drowley@postgresql.o 4493 [ + + ]: 733814 : TupleDescCompactAttr(tupdesc, attrnum - 1)->attlen != -1)
1402 akapila@postgresql.o 4494 : 706467 : continue;
4495 : :
4496 : : /*
4497 : : * Check if the old tuple's attribute is stored externally and is a
4498 : : * member of external_cols.
4499 : : */
4500 [ + + + + ]: 32343 : if (VARATT_IS_EXTERNAL((struct varlena *) DatumGetPointer(value1)) &&
1021 tgl@sss.pgh.pa.us 4501 : 5 : bms_is_member(attidx, external_cols))
1402 akapila@postgresql.o 4502 : 2 : *has_external = true;
4503 : : }
4504 : :
3185 alvherre@alvh.no-ip. 4505 : 312128 : return modified;
4506 : : }
4507 : :
4508 : : /*
4509 : : * simple_heap_update - replace a tuple
4510 : : *
4511 : : * This routine may be used to update a tuple when concurrent updates of
4512 : : * the target tuple are not expected (for example, because we have a lock
4513 : : * on the relation associated with the tuple). Any failure is reported
4514 : : * via ereport().
4515 : : */
4516 : : void
48 peter@eisentraut.org 4517 :GNC 117419 : simple_heap_update(Relation relation, const ItemPointerData *otid, HeapTuple tup,
4518 : : TU_UpdateIndexes *update_indexes)
4519 : : {
4520 : : TM_Result result;
4521 : : TM_FailureData tmfd;
4522 : : LockTupleMode lockmode;
4523 : :
8129 tgl@sss.pgh.pa.us 4524 :CBC 117419 : result = heap_update(relation, otid, tup,
4525 : : GetCurrentCommandId(true), InvalidSnapshot,
4526 : : true /* wait for commit */ ,
4527 : : &tmfd, &lockmode, update_indexes);
9094 4528 [ - + - - : 117419 : switch (result)
- ]
4529 : : {
2461 andres@anarazel.de 4530 :UBC 0 : case TM_SelfModified:
4531 : : /* Tuple was already updated in current command? */
8185 tgl@sss.pgh.pa.us 4532 [ # # ]: 0 : elog(ERROR, "tuple already updated by self");
4533 : : break;
4534 : :
2461 andres@anarazel.de 4535 :CBC 117419 : case TM_Ok:
4536 : : /* done successfully */
9094 tgl@sss.pgh.pa.us 4537 : 117419 : break;
4538 : :
2461 andres@anarazel.de 4539 :UBC 0 : case TM_Updated:
8185 tgl@sss.pgh.pa.us 4540 [ # # ]: 0 : elog(ERROR, "tuple concurrently updated");
4541 : : break;
4542 : :
2461 andres@anarazel.de 4543 : 0 : case TM_Deleted:
4544 [ # # ]: 0 : elog(ERROR, "tuple concurrently deleted");
4545 : : break;
4546 : :
9094 tgl@sss.pgh.pa.us 4547 : 0 : default:
8185 4548 [ # # ]: 0 : elog(ERROR, "unrecognized heap_update status: %u", result);
4549 : : break;
4550 : : }
9094 tgl@sss.pgh.pa.us 4551 :CBC 117419 : }
4552 : :
4553 : :
4554 : : /*
4555 : : * Return the MultiXactStatus corresponding to the given tuple lock mode.
4556 : : */
4557 : : static MultiXactStatus
4711 alvherre@alvh.no-ip. 4558 : 115444 : get_mxact_status_for_lock(LockTupleMode mode, bool is_update)
4559 : : {
4560 : : int retval;
4561 : :
4562 [ + + ]: 115444 : if (is_update)
4563 : 215 : retval = tupleLockExtraInfo[mode].updstatus;
4564 : : else
4565 : 115229 : retval = tupleLockExtraInfo[mode].lockstatus;
4566 : :
4567 [ - + ]: 115444 : if (retval == -1)
4711 alvherre@alvh.no-ip. 4568 [ # # # # ]:UBC 0 : elog(ERROR, "invalid lock tuple mode %d/%s", mode,
4569 : : is_update ? "true" : "false");
4570 : :
4558 alvherre@alvh.no-ip. 4571 :CBC 115444 : return (MultiXactStatus) retval;
4572 : : }
4573 : :
4574 : : /*
4575 : : * heap_lock_tuple - lock a tuple in shared or exclusive mode
4576 : : *
4577 : : * Note that this acquires a buffer pin, which the caller must release.
4578 : : *
4579 : : * Input parameters:
4580 : : * relation: relation containing tuple (caller must hold suitable lock)
4581 : : * cid: current command ID (used for visibility test, and stored into
4582 : : * tuple's cmax if lock is successful)
4583 : : * mode: indicates if shared or exclusive tuple lock is desired
4584 : : * wait_policy: what to do if tuple lock is not available
4585 : : * follow_updates: if true, follow the update chain to also lock descendant
4586 : : * tuples.
4587 : : *
4588 : : * Output parameters:
4589 : : * *tuple: all fields filled in
4590 : : * *buffer: set to buffer holding tuple (pinned but not locked at exit)
4591 : : * *tmfd: filled in failure cases (see below)
4592 : : *
4593 : : * Function results are the same as the ones for table_tuple_lock().
4594 : : *
4595 : : * In the failure cases other than TM_Invisible, the routine fills
4596 : : * *tmfd with the tuple's t_ctid, t_xmax (resolving a possible MultiXact,
4597 : : * if necessary), and t_cmax (the last only for TM_SelfModified,
4598 : : * since we cannot obtain cmax from a combo CID generated by another
4599 : : * transaction).
4600 : : * See comments for struct TM_FailureData for additional info.
4601 : : *
4602 : : * See README.tuplock for a thorough explanation of this mechanism.
4603 : : */
4604 : : TM_Result
615 akorotkov@postgresql 4605 : 158395 : heap_lock_tuple(Relation relation, HeapTuple tuple,
4606 : : CommandId cid, LockTupleMode mode, LockWaitPolicy wait_policy,
4607 : : bool follow_updates,
4608 : : Buffer *buffer, TM_FailureData *tmfd)
4609 : : {
4610 : : TM_Result result;
4611 : 158395 : ItemPointer tid = &(tuple->t_self);
4612 : : ItemId lp;
4613 : : Page page;
3439 andres@anarazel.de 4614 : 158395 : Buffer vmbuffer = InvalidBuffer;
4615 : : BlockNumber block;
4616 : : TransactionId xid,
4617 : : xmax;
4618 : : uint16 old_infomask,
4619 : : new_infomask,
4620 : : new_infomask2;
3904 alvherre@alvh.no-ip. 4621 : 158395 : bool first_time = true;
2374 4622 : 158395 : bool skip_tuple_lock = false;
7536 tgl@sss.pgh.pa.us 4623 : 158395 : bool have_tuple_lock = false;
3439 andres@anarazel.de 4624 : 158395 : bool cleared_all_frozen = false;
4625 : :
615 akorotkov@postgresql 4626 : 158395 : *buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
3439 andres@anarazel.de 4627 : 158395 : block = ItemPointerGetBlockNumber(tid);
4628 : :
4629 : : /*
4630 : : * Before locking the buffer, pin the visibility map page if it appears to
4631 : : * be necessary. Since we haven't got the lock yet, someone else might be
4632 : : * in the middle of changing this, so we'll need to recheck after we have
4633 : : * the lock.
4634 : : */
615 akorotkov@postgresql 4635 [ + + ]: 158395 : if (PageIsAllVisible(BufferGetPage(*buffer)))
3439 andres@anarazel.de 4636 : 1666 : visibilitymap_pin(relation, block, &vmbuffer);
4637 : :
615 akorotkov@postgresql 4638 : 158395 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4639 : :
4640 : 158395 : page = BufferGetPage(*buffer);
6366 tgl@sss.pgh.pa.us 4641 : 158395 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
6671 4642 [ - + ]: 158395 : Assert(ItemIdIsNormal(lp));
4643 : :
6366 4644 : 158395 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
9864 vadim4o@yahoo.com 4645 : 158395 : tuple->t_len = ItemIdGetLength(lp);
7424 tgl@sss.pgh.pa.us 4646 : 158395 : tuple->t_tableOid = RelationGetRelid(relation);
4647 : :
9864 vadim4o@yahoo.com 4648 : 12 : l3:
615 akorotkov@postgresql 4649 : 158407 : result = HeapTupleSatisfiesUpdate(tuple, cid, *buffer);
4650 : :
2461 andres@anarazel.de 4651 [ + + ]: 158407 : if (result == TM_Invisible)
4652 : : {
4653 : : /*
4654 : : * This is possible, but only when locking a tuple for ON CONFLICT
4655 : : * UPDATE. We return this value here rather than throwing an error in
4656 : : * order to give that case the opportunity to throw a more specific
4657 : : * error.
4658 : : */
4659 : 12 : result = TM_Invisible;
3439 4660 : 12 : goto out_locked;
4661 : : }
2461 4662 [ + + + + ]: 158395 : else if (result == TM_BeingModified ||
4663 [ - + ]: 77195 : result == TM_Updated ||
4664 : : result == TM_Deleted)
4665 : : {
4666 : : TransactionId xwait;
4667 : : uint16 infomask;
4668 : : uint16 infomask2;
4669 : : bool require_sleep;
4670 : : ItemPointerData t_ctid;
4671 : :
4672 : : /* must copy state data before unlocking buffer */
4711 alvherre@alvh.no-ip. 4673 : 81200 : xwait = HeapTupleHeaderGetRawXmax(tuple->t_data);
7536 tgl@sss.pgh.pa.us 4674 : 81200 : infomask = tuple->t_data->t_infomask;
4711 alvherre@alvh.no-ip. 4675 : 81200 : infomask2 = tuple->t_data->t_infomask2;
4676 : 81200 : ItemPointerCopy(&tuple->t_data->t_ctid, &t_ctid);
4677 : :
615 akorotkov@postgresql 4678 : 81200 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4679 : :
4680 : : /*
4681 : : * If any subtransaction of the current top transaction already holds
4682 : : * a lock as strong as or stronger than what we're requesting, we
4683 : : * effectively hold the desired lock already. We *must* succeed
4684 : : * without trying to take the tuple lock, else we will deadlock
4685 : : * against anyone wanting to acquire a stronger lock.
4686 : : *
4687 : : * Note we only do this the first time we loop on the HTSU result;
4688 : : * there is no point in testing in subsequent passes, because
4689 : : * evidently our own transaction cannot have acquired a new lock after
4690 : : * the first time we checked.
4691 : : */
3904 alvherre@alvh.no-ip. 4692 [ + + ]: 81200 : if (first_time)
4693 : : {
4694 : 81192 : first_time = false;
4695 : :
4696 [ + + ]: 81192 : if (infomask & HEAP_XMAX_IS_MULTI)
4697 : : {
4698 : : int i;
4699 : : int nmembers;
4700 : : MultiXactMember *members;
4701 : :
4702 : : /*
4703 : : * We don't need to allow old multixacts here; if that had
4704 : : * been the case, HeapTupleSatisfiesUpdate would have returned
4705 : : * MayBeUpdated and we wouldn't be here.
4706 : : */
4707 : : nmembers =
4708 : 73306 : GetMultiXactIdMembers(xwait, &members, false,
4709 : 73306 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
4710 : :
4711 [ + + ]: 1422692 : for (i = 0; i < nmembers; i++)
4712 : : {
4713 : : /* only consider members of our own transaction */
4714 [ + + ]: 1349400 : if (!TransactionIdIsCurrentTransactionId(members[i].xid))
4715 : 1349351 : continue;
4716 : :
4717 [ + + ]: 49 : if (TUPLOCK_from_mxstatus(members[i].status) >= mode)
4718 : : {
4711 4719 : 14 : pfree(members);
2461 andres@anarazel.de 4720 : 14 : result = TM_Ok;
3439 4721 : 14 : goto out_unlocked;
4722 : : }
4723 : : else
4724 : : {
4725 : : /*
4726 : : * Disable acquisition of the heavyweight tuple lock.
4727 : : * Otherwise, when promoting a weaker lock, we might
4728 : : * deadlock with another locker that has acquired the
4729 : : * heavyweight tuple lock and is waiting for our
4730 : : * transaction to finish.
4731 : : *
4732 : : * Note that in this case we still need to wait for
4733 : : * the multixact if required, to avoid acquiring
4734 : : * conflicting locks.
4735 : : */
2374 alvherre@alvh.no-ip. 4736 : 35 : skip_tuple_lock = true;
4737 : : }
4738 : : }
4739 : :
3904 4740 [ + - ]: 73292 : if (members)
4741 : 73292 : pfree(members);
4742 : : }
4743 [ + + ]: 7886 : else if (TransactionIdIsCurrentTransactionId(xwait))
4744 : : {
4745 [ + + + + : 6576 : switch (mode)
- ]
4746 : : {
4747 : 177 : case LockTupleKeyShare:
4748 [ - + - - : 177 : Assert(HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) ||
- - ]
4749 : : HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4750 : : HEAP_XMAX_IS_EXCL_LOCKED(infomask));
2461 andres@anarazel.de 4751 : 177 : result = TM_Ok;
3439 4752 : 177 : goto out_unlocked;
3904 alvherre@alvh.no-ip. 4753 : 22 : case LockTupleShare:
4754 [ + + - + ]: 28 : if (HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4755 : 6 : HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4756 : : {
2461 andres@anarazel.de 4757 : 16 : result = TM_Ok;
3439 4758 : 16 : goto out_unlocked;
4759 : : }
3904 alvherre@alvh.no-ip. 4760 : 6 : break;
4761 : 72 : case LockTupleNoKeyExclusive:
4762 [ + + ]: 72 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4763 : : {
2461 andres@anarazel.de 4764 : 60 : result = TM_Ok;
3439 4765 : 60 : goto out_unlocked;
4766 : : }
3904 alvherre@alvh.no-ip. 4767 : 12 : break;
4768 : 6305 : case LockTupleExclusive:
4769 [ + + ]: 6305 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask) &&
4770 [ + + ]: 1265 : infomask2 & HEAP_KEYS_UPDATED)
4771 : : {
2461 andres@anarazel.de 4772 : 1244 : result = TM_Ok;
3439 4773 : 1244 : goto out_unlocked;
4774 : : }
3904 alvherre@alvh.no-ip. 4775 : 5061 : break;
4776 : : }
4777 : : }
4778 : : }
4779 : :
4780 : : /*
4781 : : * Initially assume that we will have to wait for the locking
4782 : : * transaction(s) to finish. We check various cases below in which
4783 : : * this can be turned off.
4784 : : */
4711 4785 : 79689 : require_sleep = true;
4786 [ + + ]: 79689 : if (mode == LockTupleKeyShare)
4787 : : {
4788 : : /*
4789 : : * If we're requesting KeyShare, and there's no update present, we
4790 : : * don't need to wait. Even if there is an update, we can still
4791 : : * continue if the key hasn't been modified.
4792 : : *
4793 : : * However, if there are updates, we need to walk the update chain
4794 : : * to mark future versions of the row as locked, too. That way,
4795 : : * if somebody deletes that future version, we're protected
4796 : : * against the key going away. This locking of future versions
4797 : : * could block momentarily, if a concurrent transaction is
4798 : : * deleting a key; or it could return a value to the effect that
4799 : : * the transaction deleting the key has already committed. So we
4800 : : * do this before re-locking the buffer; otherwise this would be
4801 : : * prone to deadlocks.
4802 : : *
4803 : : * Note that the TID we're locking was grabbed before we unlocked
4804 : : * the buffer. For it to change while we're not looking, the
4805 : : * other properties we're testing for below after re-locking the
4806 : : * buffer would also change, in which case we would restart this
4807 : : * loop above.
4808 : : */
4809 [ + + ]: 73865 : if (!(infomask2 & HEAP_KEYS_UPDATED))
4810 : : {
4811 : : bool updated;
4812 : :
4813 : 73822 : updated = !HEAP_XMAX_IS_LOCKED_ONLY(infomask);
4814 : :
4815 : : /*
4816 : : * If there are updates, follow the update chain; bail out if
4817 : : * that cannot be done.
4818 : : */
4819 [ + - + + ]: 73822 : if (follow_updates && updated)
4820 : : {
4821 : : TM_Result res;
4822 : :
4823 : 2169 : res = heap_lock_updated_tuple(relation, tuple, &t_ctid,
4824 : : GetCurrentTransactionId(),
4825 : : mode);
2461 andres@anarazel.de 4826 [ + + ]: 2169 : if (res != TM_Ok)
4827 : : {
4711 alvherre@alvh.no-ip. 4828 : 6 : result = res;
4829 : : /* recovery code expects to have buffer lock held */
615 akorotkov@postgresql 4830 : 6 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4711 alvherre@alvh.no-ip. 4831 : 195 : goto failed;
4832 : : }
4833 : : }
4834 : :
615 akorotkov@postgresql 4835 : 73816 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4836 : :
4837 : : /*
4838 : : * Make sure it's still an appropriate lock, else start over.
4839 : : * Also, if it wasn't updated before we released the lock, but
4840 : : * is updated now, we start over too; the reason is that we
4841 : : * now need to follow the update chain to lock the new
4842 : : * versions.
4843 : : */
4711 alvherre@alvh.no-ip. 4844 [ + + ]: 73816 : if (!HeapTupleHeaderIsOnlyLocked(tuple->t_data) &&
4845 [ + - ]: 2151 : ((tuple->t_data->t_infomask2 & HEAP_KEYS_UPDATED) ||
4846 [ - + ]: 2151 : !updated))
4847 : 12 : goto l3;
4848 : :
4849 : : /* Things look okay, so we can skip sleeping */
4850 : 73816 : require_sleep = false;
4851 : :
4852 : : /*
4853 : : * Note we allow Xmax to change here; other updaters/lockers
4854 : : * could have modified it before we grabbed the buffer lock.
4855 : : * However, this is not a problem, because with the recheck we
4856 : : * just did we ensure that they still don't conflict with the
4857 : : * lock we want.
4858 : : */
4859 : : }
4860 : : }
4861 [ + + ]: 5824 : else if (mode == LockTupleShare)
4862 : : {
4863 : : /*
4864 : : * If we're requesting Share, we can similarly avoid sleeping if
4865 : : * there's no update and no exclusive lock present.
4866 : : */
4867 [ + - ]: 443 : if (HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
4868 [ + + ]: 443 : !HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4869 : : {
615 akorotkov@postgresql 4870 : 437 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4871 : :
4872 : : /*
4873 : : * Make sure it's still an appropriate lock, else start over.
4874 : : * See above about allowing xmax to change.
4875 : : */
4711 alvherre@alvh.no-ip. 4876 [ + - - + ]: 874 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
4877 : 437 : HEAP_XMAX_IS_EXCL_LOCKED(tuple->t_data->t_infomask))
4711 alvherre@alvh.no-ip. 4878 :UBC 0 : goto l3;
4711 alvherre@alvh.no-ip. 4879 :CBC 437 : require_sleep = false;
4880 : : }
4881 : : }
4882 [ + + ]: 5381 : else if (mode == LockTupleNoKeyExclusive)
4883 : : {
4884 : : /*
4885 : : * If we're requesting NoKeyExclusive, we might also be able to
4886 : : * avoid sleeping; just ensure that there no conflicting lock
4887 : : * already acquired.
4888 : : */
4889 [ + + ]: 166 : if (infomask & HEAP_XMAX_IS_MULTI)
4890 : : {
4009 4891 [ + + ]: 26 : if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
4892 : : mode, NULL))
4893 : : {
4894 : : /*
4895 : : * No conflict, but if the xmax changed under us in the
4896 : : * meantime, start over.
4897 : : */
615 akorotkov@postgresql 4898 : 13 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4009 alvherre@alvh.no-ip. 4899 [ + - - + ]: 26 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4900 : 13 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4901 : : xwait))
4009 alvherre@alvh.no-ip. 4902 :UBC 0 : goto l3;
4903 : :
4904 : : /* otherwise, we're good */
4009 alvherre@alvh.no-ip. 4905 :CBC 13 : require_sleep = false;
4906 : : }
4907 : : }
4711 4908 [ + + ]: 140 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
4909 : : {
615 akorotkov@postgresql 4910 : 18 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4911 : :
4912 : : /* if the xmax changed in the meantime, start over */
4255 alvherre@alvh.no-ip. 4913 [ + - - + ]: 36 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
2148 4914 : 18 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4915 : : xwait))
4711 alvherre@alvh.no-ip. 4916 :UBC 0 : goto l3;
4917 : : /* otherwise, we're good */
4711 alvherre@alvh.no-ip. 4918 :CBC 18 : require_sleep = false;
4919 : : }
4920 : : }
4921 : :
4922 : : /*
4923 : : * As a check independent from those above, we can also avoid sleeping
4924 : : * if the current transaction is the sole locker of the tuple. Note
4925 : : * that the strength of the lock already held is irrelevant; this is
4926 : : * not about recording the lock in Xmax (which will be done regardless
4927 : : * of this optimization, below). Also, note that the cases where we
4928 : : * hold a lock stronger than we are requesting are already handled
4929 : : * above by not doing anything.
4930 : : *
4931 : : * Note we only deal with the non-multixact case here; MultiXactIdWait
4932 : : * is well equipped to deal with this situation on its own.
4933 : : */
3904 4934 [ + + + + : 85042 : if (require_sleep && !(infomask & HEAP_XMAX_IS_MULTI) &&
+ + ]
4935 : 5359 : TransactionIdIsCurrentTransactionId(xwait))
4936 : : {
4937 : : /* ... but if the xmax changed in the meantime, start over */
615 akorotkov@postgresql 4938 : 5061 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
3904 alvherre@alvh.no-ip. 4939 [ + - - + ]: 10122 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4940 : 5061 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4941 : : xwait))
3904 alvherre@alvh.no-ip. 4942 :UBC 0 : goto l3;
3904 alvherre@alvh.no-ip. 4943 [ - + ]:CBC 5061 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask));
4944 : 5061 : require_sleep = false;
4945 : : }
4946 : :
4947 : : /*
4948 : : * Time to sleep on the other transaction/multixact, if necessary.
4949 : : *
4950 : : * If the other transaction is an update/delete that's already
4951 : : * committed, then sleeping cannot possibly do any good: if we're
4952 : : * required to sleep, get out to raise an error instead.
4953 : : *
4954 : : * By here, we either have already acquired the buffer exclusive lock,
4955 : : * or we must wait for the locking transaction or multixact; so below
4956 : : * we ensure that we grab buffer lock after the sleep.
4957 : : */
2461 andres@anarazel.de 4958 [ + + + + : 79683 : if (require_sleep && (result == TM_Updated || result == TM_Deleted))
- + ]
4959 : : {
615 akorotkov@postgresql 4960 : 151 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
3442 alvherre@alvh.no-ip. 4961 : 151 : goto failed;
4962 : : }
4963 [ + + ]: 79532 : else if (require_sleep)
4964 : : {
4965 : : /*
4966 : : * Acquire tuple lock to establish our priority for the tuple, or
4967 : : * die trying. LockTuple will release us when we are next-in-line
4968 : : * for the tuple. We must do this even if we are share-locking,
4969 : : * but not if we already have a weaker lock on the tuple.
4970 : : *
4971 : : * If we are forced to "start over" below, we keep the tuple lock;
4972 : : * this arranges that we stay at the head of the line while
4973 : : * rechecking tuple state.
4974 : : */
2374 4975 [ + + ]: 187 : if (!skip_tuple_lock &&
4976 [ + + ]: 171 : !heap_acquire_tuplock(relation, tid, mode, wait_policy,
4977 : : &have_tuple_lock))
4978 : : {
4979 : : /*
4980 : : * This can only happen if wait_policy is Skip and the lock
4981 : : * couldn't be obtained.
4982 : : */
2461 andres@anarazel.de 4983 : 1 : result = TM_WouldBlock;
4984 : : /* recovery code expects to have buffer lock held */
615 akorotkov@postgresql 4985 : 1 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4009 alvherre@alvh.no-ip. 4986 : 1 : goto failed;
4987 : : }
4988 : :
4711 4989 [ + + ]: 185 : if (infomask & HEAP_XMAX_IS_MULTI)
4990 : : {
4991 : 40 : MultiXactStatus status = get_mxact_status_for_lock(mode, false);
4992 : :
4993 : : /* We only ever lock tuples, never update them */
4994 [ - + ]: 40 : if (status >= MultiXactStatusNoKeyUpdate)
4711 alvherre@alvh.no-ip. 4995 [ # # ]:UBC 0 : elog(ERROR, "invalid lock mode in heap_lock_tuple");
4996 : :
4997 : : /* wait for multixact to end, or die trying */
4089 alvherre@alvh.no-ip. 4998 [ + + + - ]:CBC 40 : switch (wait_policy)
4999 : : {
5000 : 36 : case LockWaitBlock:
5001 : 36 : MultiXactIdWait((MultiXactId) xwait, status, infomask,
3101 tgl@sss.pgh.pa.us 5002 :GIC 36 : relation, &tuple->t_self, XLTW_Lock, NULL);
4089 alvherre@alvh.no-ip. 5003 :CBC 36 : break;
5004 : 2 : case LockWaitSkip:
5005 [ + - ]: 2 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5006 : : status, infomask, relation,
5007 : : NULL, false))
5008 : : {
2461 andres@anarazel.de 5009 : 2 : result = TM_WouldBlock;
5010 : : /* recovery code expects to have buffer lock held */
615 akorotkov@postgresql 5011 : 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4089 alvherre@alvh.no-ip. 5012 : 2 : goto failed;
5013 : : }
4089 alvherre@alvh.no-ip. 5014 :UBC 0 : break;
4089 alvherre@alvh.no-ip. 5015 :CBC 2 : case LockWaitError:
5016 [ + - ]: 2 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
5017 : : status, infomask, relation,
5018 : : NULL, log_lock_failures))
5019 [ + - ]: 2 : ereport(ERROR,
5020 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5021 : : errmsg("could not obtain lock on row in relation \"%s\"",
5022 : : RelationGetRelationName(relation))));
5023 : :
4089 alvherre@alvh.no-ip. 5024 :UBC 0 : break;
5025 : : }
5026 : :
5027 : : /*
5028 : : * Of course, the multixact might not be done here: if we're
5029 : : * requesting a light lock mode, other transactions with light
5030 : : * locks could still be alive, as well as locks owned by our
5031 : : * own xact or other subxacts of this backend. We need to
5032 : : * preserve the surviving MultiXact members. Note that it
5033 : : * isn't absolutely necessary in the latter case, but doing so
5034 : : * is simpler.
5035 : : */
5036 : : }
5037 : : else
5038 : : {
5039 : : /* wait for regular transaction to end, or die trying */
4089 alvherre@alvh.no-ip. 5040 [ + + + - ]:CBC 145 : switch (wait_policy)
5041 : : {
5042 : 106 : case LockWaitBlock:
3969 heikki.linnakangas@i 5043 : 106 : XactLockTableWait(xwait, relation, &tuple->t_self,
5044 : : XLTW_Lock);
4089 alvherre@alvh.no-ip. 5045 : 106 : break;
5046 : 33 : case LockWaitSkip:
278 fujii@postgresql.org 5047 [ + - ]: 33 : if (!ConditionalXactLockTableWait(xwait, false))
5048 : : {
2461 andres@anarazel.de 5049 : 33 : result = TM_WouldBlock;
5050 : : /* recovery code expects to have buffer lock held */
615 akorotkov@postgresql 5051 : 33 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4089 alvherre@alvh.no-ip. 5052 : 33 : goto failed;
5053 : : }
4089 alvherre@alvh.no-ip. 5054 :UBC 0 : break;
4089 alvherre@alvh.no-ip. 5055 :CBC 6 : case LockWaitError:
197 fujii@postgresql.org 5056 [ + - ]: 6 : if (!ConditionalXactLockTableWait(xwait, log_lock_failures))
4089 alvherre@alvh.no-ip. 5057 [ + - ]: 6 : ereport(ERROR,
5058 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5059 : : errmsg("could not obtain lock on row in relation \"%s\"",
5060 : : RelationGetRelationName(relation))));
4089 alvherre@alvh.no-ip. 5061 :UBC 0 : break;
5062 : : }
5063 : : }
5064 : :
5065 : : /* if there are updates, follow the update chain */
3904 alvherre@alvh.no-ip. 5066 [ + + + + ]:CBC 142 : if (follow_updates && !HEAP_XMAX_IS_LOCKED_ONLY(infomask))
5067 : : {
5068 : : TM_Result res;
5069 : :
5070 : 57 : res = heap_lock_updated_tuple(relation, tuple, &t_ctid,
5071 : : GetCurrentTransactionId(),
5072 : : mode);
2461 andres@anarazel.de 5073 [ + + ]: 57 : if (res != TM_Ok)
5074 : : {
3904 alvherre@alvh.no-ip. 5075 : 2 : result = res;
5076 : : /* recovery code expects to have buffer lock held */
615 akorotkov@postgresql 5077 : 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
3904 alvherre@alvh.no-ip. 5078 : 2 : goto failed;
5079 : : }
5080 : : }
5081 : :
615 akorotkov@postgresql 5082 : 140 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
5083 : :
5084 : : /*
5085 : : * xwait is done, but if xwait had just locked the tuple then some
5086 : : * other xact could update this tuple before we get to this point.
5087 : : * Check for xmax change, and start over if so.
5088 : : */
3904 alvherre@alvh.no-ip. 5089 [ + + + + ]: 270 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
5090 : 130 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
5091 : : xwait))
5092 : 12 : goto l3;
5093 : :
5094 [ + + ]: 128 : if (!(infomask & HEAP_XMAX_IS_MULTI))
5095 : : {
5096 : : /*
5097 : : * Otherwise check if it committed or aborted. Note we cannot
5098 : : * be here if the tuple was only locked by somebody who didn't
5099 : : * conflict with us; that would have been handled above. So
5100 : : * that transaction must necessarily be gone by now. But
5101 : : * don't check for this in the multixact case, because some
5102 : : * locker transactions might still be running.
5103 : : */
615 akorotkov@postgresql 5104 : 95 : UpdateXmaxHintBits(tuple->t_data, *buffer, xwait);
5105 : : }
5106 : : }
5107 : :
5108 : : /* By here, we're certain that we hold buffer exclusive lock again */
5109 : :
5110 : : /*
5111 : : * We may lock if previous xmax aborted, or if it committed but only
5112 : : * locked the tuple without updating it; or if we didn't have to wait
5113 : : * at all for whatever reason.
5114 : : */
4711 alvherre@alvh.no-ip. 5115 [ + + ]: 79473 : if (!require_sleep ||
5116 [ + + + + ]: 226 : (tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
5117 [ + + ]: 180 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
5118 : 82 : HeapTupleHeaderIsOnlyLocked(tuple->t_data))
2461 andres@anarazel.de 5119 : 79397 : result = TM_Ok;
1759 alvherre@alvh.no-ip. 5120 [ + + ]: 76 : else if (!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
2461 andres@anarazel.de 5121 : 57 : result = TM_Updated;
5122 : : else
5123 : 19 : result = TM_Deleted;
5124 : : }
5125 : :
4711 alvherre@alvh.no-ip. 5126 : 77195 : failed:
2461 andres@anarazel.de 5127 [ + + ]: 156863 : if (result != TM_Ok)
5128 : : {
5129 [ + + + + : 277 : Assert(result == TM_SelfModified || result == TM_Updated ||
+ + - + ]
5130 : : result == TM_Deleted || result == TM_WouldBlock);
5131 : :
5132 : : /*
5133 : : * When locking a tuple under LockWaitSkip semantics and we fail with
5134 : : * TM_WouldBlock above, it's possible for concurrent transactions to
5135 : : * release the lock and set HEAP_XMAX_INVALID in the meantime. So
5136 : : * this assert is slightly different from the equivalent one in
5137 : : * heap_delete and heap_update.
5138 : : */
1443 alvherre@alvh.no-ip. 5139 [ + + - + ]: 277 : Assert((result == TM_WouldBlock) ||
5140 : : !(tuple->t_data->t_infomask & HEAP_XMAX_INVALID));
2461 andres@anarazel.de 5141 [ + + - + ]: 277 : Assert(result != TM_Updated ||
5142 : : !ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid));
5143 : 277 : tmfd->ctid = tuple->t_data->t_ctid;
5144 : 277 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
5145 [ + + ]: 277 : if (result == TM_SelfModified)
5146 : 6 : tmfd->cmax = HeapTupleHeaderGetCmax(tuple->t_data);
5147 : : else
5148 : 271 : tmfd->cmax = InvalidCommandId;
3439 5149 : 277 : goto out_locked;
5150 : : }
5151 : :
5152 : : /*
5153 : : * If we didn't pin the visibility map page and the page has become all
5154 : : * visible while we were busy locking the buffer, or during some
5155 : : * subsequent window during which we had it unlocked, we'll have to unlock
5156 : : * and re-lock, to avoid holding the buffer lock across I/O. That's a bit
5157 : : * unfortunate, especially since we'll now have to recheck whether the
5158 : : * tuple has been locked or updated under us, but hopefully it won't
5159 : : * happen very often.
5160 : : */
3422 5161 [ + + - + ]: 156586 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
5162 : : {
615 akorotkov@postgresql 5163 :UBC 0 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
3422 andres@anarazel.de 5164 : 0 : visibilitymap_pin(relation, block, &vmbuffer);
615 akorotkov@postgresql 5165 : 0 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
3422 andres@anarazel.de 5166 : 0 : goto l3;
5167 : : }
5168 : :
4711 alvherre@alvh.no-ip. 5169 :CBC 156586 : xmax = HeapTupleHeaderGetRawXmax(tuple->t_data);
5170 : 156586 : old_infomask = tuple->t_data->t_infomask;
5171 : :
5172 : : /*
5173 : : * If this is the first possibly-multixact-able operation in the current
5174 : : * transaction, set my per-backend OldestMemberMXactId setting. We can be
5175 : : * certain that the transaction will never become a member of any older
5176 : : * MultiXactIds than that. (We have to do this even if we end up just
5177 : : * using our own TransactionId below, since some other backend could
5178 : : * incorporate our XID into a MultiXact immediately afterwards.)
5179 : : */
5180 : 156586 : MultiXactIdSetOldestMember();
5181 : :
5182 : : /*
5183 : : * Compute the new xmax and infomask to store into the tuple. Note we do
5184 : : * not modify the tuple just yet, because that would leave it in the wrong
5185 : : * state if multixact.c elogs.
5186 : : */
5187 : 156586 : compute_new_xmax_infomask(xmax, old_infomask, tuple->t_data->t_infomask2,
5188 : : GetCurrentTransactionId(), mode, false,
5189 : : &xid, &new_infomask, &new_infomask2);
5190 : :
7538 tgl@sss.pgh.pa.us 5191 : 156586 : START_CRIT_SECTION();
5192 : :
5193 : : /*
5194 : : * Store transaction information of xact locking the tuple.
5195 : : *
5196 : : * Note: Cmax is meaningless in this context, so don't set it; this avoids
5197 : : * possibly generating a useless combo CID. Moreover, if we're locking a
5198 : : * previously updated tuple, it's important to preserve the Cmax.
5199 : : *
5200 : : * Also reset the HOT UPDATE bit, but only if there's no update; otherwise
5201 : : * we would break the HOT chain.
5202 : : */
4711 alvherre@alvh.no-ip. 5203 : 156586 : tuple->t_data->t_infomask &= ~HEAP_XMAX_BITS;
5204 : 156586 : tuple->t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5205 : 156586 : tuple->t_data->t_infomask |= new_infomask;
5206 : 156586 : tuple->t_data->t_infomask2 |= new_infomask2;
5207 [ + + ]: 156586 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5208 : 154439 : HeapTupleHeaderClearHotUpdated(tuple->t_data);
7762 tgl@sss.pgh.pa.us 5209 : 156586 : HeapTupleHeaderSetXmax(tuple->t_data, xid);
5210 : :
5211 : : /*
5212 : : * Make sure there is no forward chain link in t_ctid. Note that in the
5213 : : * cases where the tuple has been updated, we must not overwrite t_ctid,
5214 : : * because it was set by the updater. Moreover, if the tuple has been
5215 : : * updated, we need to follow the update chain to lock the new versions of
5216 : : * the tuple as well.
5217 : : */
4711 alvherre@alvh.no-ip. 5218 [ + + ]: 156586 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
5219 : 154439 : tuple->t_data->t_ctid = *tid;
5220 : :
5221 : : /* Clear only the all-frozen bit on visibility map if needed */
3439 andres@anarazel.de 5222 [ + + + + ]: 158252 : if (PageIsAllVisible(page) &&
5223 : 1666 : visibilitymap_clear(relation, block, vmbuffer,
5224 : : VISIBILITYMAP_ALL_FROZEN))
5225 : 14 : cleared_all_frozen = true;
5226 : :
5227 : :
615 akorotkov@postgresql 5228 : 156586 : MarkBufferDirty(*buffer);
5229 : :
5230 : : /*
5231 : : * XLOG stuff. You might think that we don't need an XLOG record because
5232 : : * there is no state change worth restoring after a crash. You would be
5233 : : * wrong however: we have just written either a TransactionId or a
5234 : : * MultiXactId that may never have been seen on disk before, and we need
5235 : : * to make sure that there are XLOG entries covering those ID numbers.
5236 : : * Else the same IDs might be re-used after a crash, which would be
5237 : : * disastrous if this page made it to disk before the crash. Essentially
5238 : : * we have to enforce the WAL log-before-data rule even in this case.
5239 : : * (Also, in a PITR log-shipping or 2PC environment, we have to have XLOG
5240 : : * entries for everything anyway.)
5241 : : */
5483 rhaas@postgresql.org 5242 [ + + + + : 156586 : if (RelationNeedsWAL(relation))
+ - + - ]
5243 : : {
5244 : : xl_heap_lock xlrec;
5245 : : XLogRecPtr recptr;
5246 : :
4045 heikki.linnakangas@i 5247 : 156242 : XLogBeginInsert();
615 akorotkov@postgresql 5248 : 156242 : XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD);
5249 : :
4045 heikki.linnakangas@i 5250 : 156242 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
981 pg@bowt.ie 5251 : 156242 : xlrec.xmax = xid;
4711 alvherre@alvh.no-ip. 5252 : 312484 : xlrec.infobits_set = compute_infobits(new_infomask,
5253 : 156242 : tuple->t_data->t_infomask2);
3439 andres@anarazel.de 5254 : 156242 : xlrec.flags = cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
309 peter@eisentraut.org 5255 : 156242 : XLogRegisterData(&xlrec, SizeOfHeapLock);
5256 : :
5257 : : /* we don't decode row locks atm, so no need to log the origin */
5258 : :
4045 heikki.linnakangas@i 5259 : 156242 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
5260 : :
6366 tgl@sss.pgh.pa.us 5261 : 156242 : PageSetLSN(page, recptr);
5262 : : }
5263 : :
7538 5264 [ - + ]: 156586 : END_CRIT_SECTION();
5265 : :
2461 andres@anarazel.de 5266 : 156586 : result = TM_Ok;
5267 : :
3439 5268 : 156875 : out_locked:
615 akorotkov@postgresql 5269 : 156875 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
5270 : :
3439 andres@anarazel.de 5271 : 158386 : out_unlocked:
5272 [ + + ]: 158386 : if (BufferIsValid(vmbuffer))
5273 : 1666 : ReleaseBuffer(vmbuffer);
5274 : :
5275 : : /*
5276 : : * Don't update the visibility map here. Locking a tuple doesn't change
5277 : : * visibility info.
5278 : : */
5279 : :
5280 : : /*
5281 : : * Now that we have successfully marked the tuple as locked, we can
5282 : : * release the lmgr tuple lock, if we had it.
5283 : : */
7536 tgl@sss.pgh.pa.us 5284 [ + + ]: 158386 : if (have_tuple_lock)
4711 alvherre@alvh.no-ip. 5285 : 156 : UnlockTupleTuplock(relation, tid, mode);
5286 : :
3439 andres@anarazel.de 5287 : 158386 : return result;
5288 : : }
5289 : :
5290 : : /*
5291 : : * Acquire heavyweight lock on the given tuple, in preparation for acquiring
5292 : : * its normal, Xmax-based tuple lock.
5293 : : *
5294 : : * have_tuple_lock is an input and output parameter: on input, it indicates
5295 : : * whether the lock has previously been acquired (and this function does
5296 : : * nothing in that case). If this function returns success, have_tuple_lock
5297 : : * has been flipped to true.
5298 : : *
5299 : : * Returns false if it was unable to obtain the lock; this can only happen if
5300 : : * wait_policy is Skip.
5301 : : */
5302 : : static bool
48 peter@eisentraut.org 5303 :GNC 309 : heap_acquire_tuplock(Relation relation, const ItemPointerData *tid, LockTupleMode mode,
5304 : : LockWaitPolicy wait_policy, bool *have_tuple_lock)
5305 : : {
4009 alvherre@alvh.no-ip. 5306 [ + + ]:CBC 309 : if (*have_tuple_lock)
5307 : 9 : return true;
5308 : :
5309 [ + + + - ]: 300 : switch (wait_policy)
5310 : : {
5311 : 259 : case LockWaitBlock:
5312 : 259 : LockTupleTuplock(relation, tid, mode);
5313 : 259 : break;
5314 : :
5315 : 34 : case LockWaitSkip:
278 fujii@postgresql.org 5316 [ + + ]: 34 : if (!ConditionalLockTupleTuplock(relation, tid, mode, false))
4009 alvherre@alvh.no-ip. 5317 : 1 : return false;
5318 : 33 : break;
5319 : :
5320 : 7 : case LockWaitError:
197 fujii@postgresql.org 5321 [ + + ]: 7 : if (!ConditionalLockTupleTuplock(relation, tid, mode, log_lock_failures))
4009 alvherre@alvh.no-ip. 5322 [ + - ]: 1 : ereport(ERROR,
5323 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5324 : : errmsg("could not obtain lock on row in relation \"%s\"",
5325 : : RelationGetRelationName(relation))));
5326 : 6 : break;
5327 : : }
5328 : 298 : *have_tuple_lock = true;
5329 : :
5330 : 298 : return true;
5331 : : }
5332 : :
5333 : : /*
5334 : : * Given an original set of Xmax and infomask, and a transaction (identified by
5335 : : * add_to_xmax) acquiring a new lock of some mode, compute the new Xmax and
5336 : : * corresponding infomasks to use on the tuple.
5337 : : *
5338 : : * Note that this might have side effects such as creating a new MultiXactId.
5339 : : *
5340 : : * Most callers will have called HeapTupleSatisfiesUpdate before this function;
5341 : : * that will have set the HEAP_XMAX_INVALID bit if the xmax was a MultiXactId
5342 : : * but it was not running anymore. There is a race condition, which is that the
5343 : : * MultiXactId may have finished since then, but that uncommon case is handled
5344 : : * either here, or within MultiXactIdExpand.
5345 : : *
5346 : : * There is a similar race condition possible when the old xmax was a regular
5347 : : * TransactionId. We test TransactionIdIsInProgress again just to narrow the
5348 : : * window, but it's still possible to end up creating an unnecessary
5349 : : * MultiXactId. Fortunately this is harmless.
5350 : : */
5351 : : static void
4711 5352 : 2077283 : compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
5353 : : uint16 old_infomask2, TransactionId add_to_xmax,
5354 : : LockTupleMode mode, bool is_update,
5355 : : TransactionId *result_xmax, uint16 *result_infomask,
5356 : : uint16 *result_infomask2)
5357 : : {
5358 : : TransactionId new_xmax;
5359 : : uint16 new_infomask,
5360 : : new_infomask2;
5361 : :
4381 5362 [ + - ]: 2077283 : Assert(TransactionIdIsCurrentTransactionId(add_to_xmax));
5363 : :
4711 5364 : 2181252 : l5:
5365 : 2181252 : new_infomask = 0;
5366 : 2181252 : new_infomask2 = 0;
5367 [ + + ]: 2181252 : if (old_infomask & HEAP_XMAX_INVALID)
5368 : : {
5369 : : /*
5370 : : * No previous locker; we just insert our own TransactionId.
5371 : : *
5372 : : * Note that it's critical that this case be the first one checked,
5373 : : * because there are several blocks below that come back to this one
5374 : : * to implement certain optimizations; old_infomask might contain
5375 : : * other dirty bits in those cases, but we don't really care.
5376 : : */
5377 [ + + ]: 2000653 : if (is_update)
5378 : : {
5379 : 1766173 : new_xmax = add_to_xmax;
5380 [ + + ]: 1766173 : if (mode == LockTupleExclusive)
5381 : 1490725 : new_infomask2 |= HEAP_KEYS_UPDATED;
5382 : : }
5383 : : else
5384 : : {
5385 : 234480 : new_infomask |= HEAP_XMAX_LOCK_ONLY;
5386 [ + + + + : 234480 : switch (mode)
- ]
5387 : : {
5388 : 2633 : case LockTupleKeyShare:
5389 : 2633 : new_xmax = add_to_xmax;
5390 : 2633 : new_infomask |= HEAP_XMAX_KEYSHR_LOCK;
5391 : 2633 : break;
5392 : 745 : case LockTupleShare:
5393 : 745 : new_xmax = add_to_xmax;
5394 : 745 : new_infomask |= HEAP_XMAX_SHR_LOCK;
5395 : 745 : break;
5396 : 135433 : case LockTupleNoKeyExclusive:
5397 : 135433 : new_xmax = add_to_xmax;
5398 : 135433 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5399 : 135433 : break;
5400 : 95669 : case LockTupleExclusive:
5401 : 95669 : new_xmax = add_to_xmax;
5402 : 95669 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5403 : 95669 : new_infomask2 |= HEAP_KEYS_UPDATED;
5404 : 95669 : break;
4711 alvherre@alvh.no-ip. 5405 :UBC 0 : default:
5406 : 0 : new_xmax = InvalidTransactionId; /* silence compiler */
5407 [ # # ]: 0 : elog(ERROR, "invalid lock mode");
5408 : : }
5409 : : }
5410 : : }
4711 alvherre@alvh.no-ip. 5411 [ + + ]:CBC 180599 : else if (old_infomask & HEAP_XMAX_IS_MULTI)
5412 : : {
5413 : : MultiXactStatus new_status;
5414 : :
5415 : : /*
5416 : : * Currently we don't allow XMAX_COMMITTED to be set for multis, so
5417 : : * cross-check.
5418 : : */
5419 [ - + ]: 75570 : Assert(!(old_infomask & HEAP_XMAX_COMMITTED));
5420 : :
5421 : : /*
5422 : : * A multixact together with LOCK_ONLY set but neither lock bit set
5423 : : * (i.e. a pg_upgraded share locked tuple) cannot possibly be running
5424 : : * anymore. This check is critical for databases upgraded by
5425 : : * pg_upgrade; both MultiXactIdIsRunning and MultiXactIdExpand assume
5426 : : * that such multis are never passed.
5427 : : */
3463 5428 [ - + ]: 75570 : if (HEAP_LOCKED_UPGRADED(old_infomask))
5429 : : {
4711 alvherre@alvh.no-ip. 5430 :UBC 0 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5431 : 0 : old_infomask |= HEAP_XMAX_INVALID;
5432 : 0 : goto l5;
5433 : : }
5434 : :
5435 : : /*
5436 : : * If the XMAX is already a MultiXactId, then we need to expand it to
5437 : : * include add_to_xmax; but if all the members were lockers and are
5438 : : * all gone, we can do away with the IS_MULTI bit and just set
5439 : : * add_to_xmax as the only locker/updater. If all lockers are gone
5440 : : * and we have an updater that aborted, we can also do without a
5441 : : * multi.
5442 : : *
5443 : : * The cost of doing GetMultiXactIdMembers would be paid by
5444 : : * MultiXactIdExpand if we weren't to do this, so this check is not
5445 : : * incurring extra work anyhow.
5446 : : */
4159 alvherre@alvh.no-ip. 5447 [ + + ]:CBC 75570 : if (!MultiXactIdIsRunning(xmax, HEAP_XMAX_IS_LOCKED_ONLY(old_infomask)))
5448 : : {
4711 5449 [ + + ]: 24 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) ||
3904 5450 [ + - ]: 8 : !TransactionIdDidCommit(MultiXactIdGetUpdateXid(xmax,
5451 : : old_infomask)))
5452 : : {
5453 : : /*
5454 : : * Reset these bits and restart; otherwise fall through to
5455 : : * create a new multi below.
5456 : : */
4711 5457 : 24 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5458 : 24 : old_infomask |= HEAP_XMAX_INVALID;
5459 : 24 : goto l5;
5460 : : }
5461 : : }
5462 : :
5463 : 75546 : new_status = get_mxact_status_for_lock(mode, is_update);
5464 : :
5465 : 75546 : new_xmax = MultiXactIdExpand((MultiXactId) xmax, add_to_xmax,
5466 : : new_status);
5467 : 75546 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5468 : : }
5469 [ + + ]: 105029 : else if (old_infomask & HEAP_XMAX_COMMITTED)
5470 : : {
5471 : : /*
5472 : : * It's a committed update, so we need to preserve him as updater of
5473 : : * the tuple.
5474 : : */
5475 : : MultiXactStatus status;
5476 : : MultiXactStatus new_status;
5477 : :
5478 [ - + ]: 13 : if (old_infomask2 & HEAP_KEYS_UPDATED)
4711 alvherre@alvh.no-ip. 5479 :UBC 0 : status = MultiXactStatusUpdate;
5480 : : else
4711 alvherre@alvh.no-ip. 5481 :CBC 13 : status = MultiXactStatusNoKeyUpdate;
5482 : :
5483 : 13 : new_status = get_mxact_status_for_lock(mode, is_update);
5484 : :
5485 : : /*
5486 : : * since it's not running, it's obviously impossible for the old
5487 : : * updater to be identical to the current one, so we need not check
5488 : : * for that case as we do in the block above.
5489 : : */
5490 : 13 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5491 : 13 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5492 : : }
5493 [ + + ]: 105016 : else if (TransactionIdIsInProgress(xmax))
5494 : : {
5495 : : /*
5496 : : * If the XMAX is a valid, in-progress TransactionId, then we need to
5497 : : * create a new MultiXactId that includes both the old locker or
5498 : : * updater and our own TransactionId.
5499 : : */
5500 : : MultiXactStatus new_status;
5501 : : MultiXactStatus old_status;
5502 : : LockTupleMode old_mode;
5503 : :
5504 [ + + ]: 105007 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5505 : : {
5506 [ + + ]: 104981 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
4381 5507 : 5671 : old_status = MultiXactStatusForKeyShare;
4711 5508 [ + + ]: 99310 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
4381 5509 : 433 : old_status = MultiXactStatusForShare;
4711 5510 [ + - ]: 98877 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5511 : : {
5512 [ + + ]: 98877 : if (old_infomask2 & HEAP_KEYS_UPDATED)
4381 5513 : 92739 : old_status = MultiXactStatusForUpdate;
5514 : : else
5515 : 6138 : old_status = MultiXactStatusForNoKeyUpdate;
5516 : : }
5517 : : else
5518 : : {
5519 : : /*
5520 : : * LOCK_ONLY can be present alone only when a page has been
5521 : : * upgraded by pg_upgrade. But in that case,
5522 : : * TransactionIdIsInProgress() should have returned false. We
5523 : : * assume it's no longer locked in this case.
5524 : : */
4711 alvherre@alvh.no-ip. 5525 [ # # ]:UBC 0 : elog(WARNING, "LOCK_ONLY found for Xid in progress %u", xmax);
5526 : 0 : old_infomask |= HEAP_XMAX_INVALID;
5527 : 0 : old_infomask &= ~HEAP_XMAX_LOCK_ONLY;
5528 : 0 : goto l5;
5529 : : }
5530 : : }
5531 : : else
5532 : : {
5533 : : /* it's an update, but which kind? */
4711 alvherre@alvh.no-ip. 5534 [ - + ]:CBC 26 : if (old_infomask2 & HEAP_KEYS_UPDATED)
4381 alvherre@alvh.no-ip. 5535 :UBC 0 : old_status = MultiXactStatusUpdate;
5536 : : else
4381 alvherre@alvh.no-ip. 5537 :CBC 26 : old_status = MultiXactStatusNoKeyUpdate;
5538 : : }
5539 : :
5540 : 105007 : old_mode = TUPLOCK_from_mxstatus(old_status);
5541 : :
5542 : : /*
5543 : : * If the lock to be acquired is for the same TransactionId as the
5544 : : * existing lock, there's an optimization possible: consider only the
5545 : : * strongest of both locks as the only one present, and restart.
5546 : : */
4711 5547 [ + + ]: 105007 : if (xmax == add_to_xmax)
5548 : : {
5549 : : /*
5550 : : * Note that it's not possible for the original tuple to be
5551 : : * updated: we wouldn't be here because the tuple would have been
5552 : : * invisible and we wouldn't try to update it. As a subtlety,
5553 : : * this code can also run when traversing an update chain to lock
5554 : : * future versions of a tuple. But we wouldn't be here either,
5555 : : * because the add_to_xmax would be different from the original
5556 : : * updater.
5557 : : */
4381 5558 [ - + ]: 103937 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5559 : :
5560 : : /* acquire the strongest of both */
5561 [ + + ]: 103937 : if (mode < old_mode)
5562 : 52176 : mode = old_mode;
5563 : : /* mustn't touch is_update */
5564 : :
5565 : 103937 : old_infomask |= HEAP_XMAX_INVALID;
5566 : 103937 : goto l5;
5567 : : }
5568 : :
5569 : : /* otherwise, just fall back to creating a new multixact */
5570 : 1070 : new_status = get_mxact_status_for_lock(mode, is_update);
5571 : 1070 : new_xmax = MultiXactIdCreate(xmax, old_status,
5572 : : add_to_xmax, new_status);
4711 5573 : 1070 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5574 : : }
5575 [ + + + + ]: 14 : else if (!HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) &&
5576 : 5 : TransactionIdDidCommit(xmax))
5577 : 1 : {
5578 : : /*
5579 : : * It's a committed update, so we gotta preserve him as updater of the
5580 : : * tuple.
5581 : : */
5582 : : MultiXactStatus status;
5583 : : MultiXactStatus new_status;
5584 : :
5585 [ - + ]: 1 : if (old_infomask2 & HEAP_KEYS_UPDATED)
4711 alvherre@alvh.no-ip. 5586 :UBC 0 : status = MultiXactStatusUpdate;
5587 : : else
4711 alvherre@alvh.no-ip. 5588 :CBC 1 : status = MultiXactStatusNoKeyUpdate;
5589 : :
5590 : 1 : new_status = get_mxact_status_for_lock(mode, is_update);
5591 : :
5592 : : /*
5593 : : * since it's not running, it's obviously impossible for the old
5594 : : * updater to be identical to the current one, so we need not check
5595 : : * for that case as we do in the block above.
5596 : : */
5597 : 1 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5598 : 1 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5599 : : }
5600 : : else
5601 : : {
5602 : : /*
5603 : : * Can get here iff the locking/updating transaction was running when
5604 : : * the infomask was extracted from the tuple, but finished before
5605 : : * TransactionIdIsInProgress got to run. Deal with it as if there was
5606 : : * no locker at all in the first place.
5607 : : */
5608 : 8 : old_infomask |= HEAP_XMAX_INVALID;
5609 : 8 : goto l5;
5610 : : }
5611 : :
5612 : 2077283 : *result_infomask = new_infomask;
5613 : 2077283 : *result_infomask2 = new_infomask2;
5614 : 2077283 : *result_xmax = new_xmax;
5615 : 2077283 : }
5616 : :
5617 : : /*
5618 : : * Subroutine for heap_lock_updated_tuple_rec.
5619 : : *
5620 : : * Given a hypothetical multixact status held by the transaction identified
5621 : : * with the given xid, does the current transaction need to wait, fail, or can
5622 : : * it continue if it wanted to acquire a lock of the given mode? "needwait"
5623 : : * is set to true if waiting is necessary; if it can continue, then TM_Ok is
5624 : : * returned. If the lock is already held by the current transaction, return
5625 : : * TM_SelfModified. In case of a conflict with another transaction, a
5626 : : * different HeapTupleSatisfiesUpdate return code is returned.
5627 : : *
5628 : : * The held status is said to be hypothetical because it might correspond to a
5629 : : * lock held by a single Xid, i.e. not a real MultiXactId; we express it this
5630 : : * way for simplicity of API.
5631 : : */
5632 : : static TM_Result
4403 5633 : 38774 : test_lockmode_for_conflict(MultiXactStatus status, TransactionId xid,
5634 : : LockTupleMode mode, HeapTuple tup,
5635 : : bool *needwait)
5636 : : {
5637 : : MultiXactStatus wantedstatus;
5638 : :
5639 : 38774 : *needwait = false;
5640 : 38774 : wantedstatus = get_mxact_status_for_lock(mode, false);
5641 : :
5642 : : /*
5643 : : * Note: we *must* check TransactionIdIsInProgress before
5644 : : * TransactionIdDidAbort/Commit; see comment at top of heapam_visibility.c
5645 : : * for an explanation.
5646 : : */
5647 [ - + ]: 38774 : if (TransactionIdIsCurrentTransactionId(xid))
5648 : : {
5649 : : /*
5650 : : * The tuple has already been locked by our own transaction. This is
5651 : : * very rare but can happen if multiple transactions are trying to
5652 : : * lock an ancient version of the same tuple.
5653 : : */
2461 andres@anarazel.de 5654 :UBC 0 : return TM_SelfModified;
5655 : : }
4403 alvherre@alvh.no-ip. 5656 [ + + ]:CBC 38774 : else if (TransactionIdIsInProgress(xid))
5657 : : {
5658 : : /*
5659 : : * If the locking transaction is running, what we do depends on
5660 : : * whether the lock modes conflict: if they do, then we must wait for
5661 : : * it to finish; otherwise we can fall through to lock this tuple
5662 : : * version without waiting.
5663 : : */
5664 [ + + ]: 36539 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5665 : 36539 : LOCKMODE_from_mxstatus(wantedstatus)))
5666 : : {
5667 : 8 : *needwait = true;
5668 : : }
5669 : :
5670 : : /*
5671 : : * If we set needwait above, then this value doesn't matter;
5672 : : * otherwise, this value signals to caller that it's okay to proceed.
5673 : : */
2461 andres@anarazel.de 5674 : 36539 : return TM_Ok;
5675 : : }
4403 alvherre@alvh.no-ip. 5676 [ + + ]: 2235 : else if (TransactionIdDidAbort(xid))
2461 andres@anarazel.de 5677 : 206 : return TM_Ok;
4403 alvherre@alvh.no-ip. 5678 [ + - ]: 2029 : else if (TransactionIdDidCommit(xid))
5679 : : {
5680 : : /*
5681 : : * The other transaction committed. If it was only a locker, then the
5682 : : * lock is completely gone now and we can return success; but if it
5683 : : * was an update, then what we do depends on whether the two lock
5684 : : * modes conflict. If they conflict, then we must report error to
5685 : : * caller. But if they don't, we can fall through to allow the current
5686 : : * transaction to lock the tuple.
5687 : : *
5688 : : * Note: the reason we worry about ISUPDATE here is because as soon as
5689 : : * a transaction ends, all its locks are gone and meaningless, and
5690 : : * thus we can ignore them; whereas its updates persist. In the
5691 : : * TransactionIdIsInProgress case, above, we don't need to check
5692 : : * because we know the lock is still "alive" and thus a conflict needs
5693 : : * always be checked.
5694 : : */
4395 5695 [ + + ]: 2029 : if (!ISUPDATE_from_mxstatus(status))
2461 andres@anarazel.de 5696 : 2020 : return TM_Ok;
5697 : :
4403 alvherre@alvh.no-ip. 5698 [ + + ]: 9 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5699 : 9 : LOCKMODE_from_mxstatus(wantedstatus)))
5700 : : {
5701 : : /* bummer */
1759 5702 [ + + ]: 8 : if (!ItemPointerEquals(&tup->t_self, &tup->t_data->t_ctid))
2461 andres@anarazel.de 5703 : 6 : return TM_Updated;
5704 : : else
5705 : 2 : return TM_Deleted;
5706 : : }
5707 : :
5708 : 1 : return TM_Ok;
5709 : : }
5710 : :
5711 : : /* Not in progress, not aborted, not committed -- must have crashed */
2461 andres@anarazel.de 5712 :UBC 0 : return TM_Ok;
5713 : : }
5714 : :
5715 : :
5716 : : /*
5717 : : * Recursive part of heap_lock_updated_tuple
5718 : : *
5719 : : * Fetch the tuple pointed to by tid in rel, and mark it as locked by the given
5720 : : * xid with the given mode; if this tuple is updated, recurse to lock the new
5721 : : * version as well.
5722 : : */
5723 : : static TM_Result
48 peter@eisentraut.org 5724 :GNC 2210 : heap_lock_updated_tuple_rec(Relation rel, const ItemPointerData *tid, TransactionId xid,
5725 : : LockTupleMode mode)
5726 : : {
5727 : : TM_Result result;
5728 : : ItemPointerData tupid;
5729 : : HeapTupleData mytup;
5730 : : Buffer buf;
5731 : : uint16 new_infomask,
5732 : : new_infomask2,
5733 : : old_infomask,
5734 : : old_infomask2;
5735 : : TransactionId xmax,
5736 : : new_xmax;
4403 alvherre@alvh.no-ip. 5737 :CBC 2210 : TransactionId priorXmax = InvalidTransactionId;
3439 andres@anarazel.de 5738 : 2210 : bool cleared_all_frozen = false;
5739 : : bool pinned_desired_page;
5740 : 2210 : Buffer vmbuffer = InvalidBuffer;
5741 : : BlockNumber block;
5742 : :
4711 alvherre@alvh.no-ip. 5743 : 2210 : ItemPointerCopy(tid, &tupid);
5744 : :
5745 : : for (;;)
5746 : : {
5747 : 2213 : new_infomask = 0;
5748 : 2213 : new_xmax = InvalidTransactionId;
3439 andres@anarazel.de 5749 : 2213 : block = ItemPointerGetBlockNumber(&tupid);
4711 alvherre@alvh.no-ip. 5750 : 2213 : ItemPointerCopy(&tupid, &(mytup.t_self));
5751 : :
1344 tgl@sss.pgh.pa.us 5752 [ + - ]: 2213 : if (!heap_fetch(rel, SnapshotAny, &mytup, &buf, false))
5753 : : {
5754 : : /*
5755 : : * if we fail to find the updated version of the tuple, it's
5756 : : * because it was vacuumed/pruned away after its creator
5757 : : * transaction aborted. So behave as if we got to the end of the
5758 : : * chain, and there's no further tuple to lock: return success to
5759 : : * caller.
5760 : : */
2461 andres@anarazel.de 5761 :UBC 0 : result = TM_Ok;
2847 tgl@sss.pgh.pa.us 5762 : 0 : goto out_unlocked;
5763 : : }
5764 : :
4711 alvherre@alvh.no-ip. 5765 :CBC 2213 : l4:
5766 [ - + ]: 2221 : CHECK_FOR_INTERRUPTS();
5767 : :
5768 : : /*
5769 : : * Before locking the buffer, pin the visibility map page if it
5770 : : * appears to be necessary. Since we haven't got the lock yet,
5771 : : * someone else might be in the middle of changing this, so we'll need
5772 : : * to recheck after we have the lock.
5773 : : */
3439 andres@anarazel.de 5774 [ - + ]: 2221 : if (PageIsAllVisible(BufferGetPage(buf)))
5775 : : {
3439 andres@anarazel.de 5776 :UBC 0 : visibilitymap_pin(rel, block, &vmbuffer);
2847 tgl@sss.pgh.pa.us 5777 : 0 : pinned_desired_page = true;
5778 : : }
5779 : : else
2847 tgl@sss.pgh.pa.us 5780 :CBC 2221 : pinned_desired_page = false;
5781 : :
4711 alvherre@alvh.no-ip. 5782 : 2221 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5783 : :
5784 : : /*
5785 : : * If we didn't pin the visibility map page and the page has become
5786 : : * all visible while we were busy locking the buffer, we'll have to
5787 : : * unlock and re-lock, to avoid holding the buffer lock across I/O.
5788 : : * That's a bit unfortunate, but hopefully shouldn't happen often.
5789 : : *
5790 : : * Note: in some paths through this function, we will reach here
5791 : : * holding a pin on a vm page that may or may not be the one matching
5792 : : * this page. If this page isn't all-visible, we won't use the vm
5793 : : * page, but we hold onto such a pin till the end of the function.
5794 : : */
2847 tgl@sss.pgh.pa.us 5795 [ + - - + ]: 2221 : if (!pinned_desired_page && PageIsAllVisible(BufferGetPage(buf)))
5796 : : {
3422 andres@anarazel.de 5797 :UBC 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5798 : 0 : visibilitymap_pin(rel, block, &vmbuffer);
5799 : 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5800 : : }
5801 : :
5802 : : /*
5803 : : * Check the tuple XMIN against prior XMAX, if any. If we reached the
5804 : : * end of the chain, we're done, so return success.
5805 : : */
4403 alvherre@alvh.no-ip. 5806 [ + + - + ]:CBC 2224 : if (TransactionIdIsValid(priorXmax) &&
2967 5807 : 3 : !TransactionIdEquals(HeapTupleHeaderGetXmin(mytup.t_data),
5808 : : priorXmax))
5809 : : {
2461 andres@anarazel.de 5810 :UBC 0 : result = TM_Ok;
3439 5811 : 0 : goto out_locked;
5812 : : }
5813 : :
5814 : : /*
5815 : : * Also check Xmin: if this tuple was created by an aborted
5816 : : * (sub)transaction, then we already locked the last live one in the
5817 : : * chain, thus we're done, so return success.
5818 : : */
3386 alvherre@alvh.no-ip. 5819 [ + + ]:CBC 2221 : if (TransactionIdDidAbort(HeapTupleHeaderGetXmin(mytup.t_data)))
5820 : : {
2461 andres@anarazel.de 5821 : 24 : result = TM_Ok;
2847 tgl@sss.pgh.pa.us 5822 : 24 : goto out_locked;
5823 : : }
5824 : :
4711 alvherre@alvh.no-ip. 5825 : 2197 : old_infomask = mytup.t_data->t_infomask;
4403 5826 : 2197 : old_infomask2 = mytup.t_data->t_infomask2;
4711 5827 : 2197 : xmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5828 : :
5829 : : /*
5830 : : * If this tuple version has been updated or locked by some concurrent
5831 : : * transaction(s), what we do depends on whether our lock mode
5832 : : * conflicts with what those other transactions hold, and also on the
5833 : : * status of them.
5834 : : */
4403 5835 [ + + ]: 2197 : if (!(old_infomask & HEAP_XMAX_INVALID))
5836 : : {
5837 : : TransactionId rawxmax;
5838 : : bool needwait;
5839 : :
5840 : 2138 : rawxmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5841 [ + + ]: 2138 : if (old_infomask & HEAP_XMAX_IS_MULTI)
5842 : : {
5843 : : int nmembers;
5844 : : int i;
5845 : : MultiXactMember *members;
5846 : :
5847 : : /*
5848 : : * We don't need a test for pg_upgrade'd tuples: this is only
5849 : : * applied to tuples after the first in an update chain. Said
5850 : : * first tuple in the chain may well be locked-in-9.2-and-
5851 : : * pg_upgraded, but that one was already locked by our caller,
5852 : : * not us; and any subsequent ones cannot be because our
5853 : : * caller must necessarily have obtained a snapshot later than
5854 : : * the pg_upgrade itself.
5855 : : */
3463 5856 [ - + ]: 2109 : Assert(!HEAP_LOCKED_UPGRADED(mytup.t_data->t_infomask));
5857 : :
4159 5858 : 2109 : nmembers = GetMultiXactIdMembers(rawxmax, &members, false,
3101 tgl@sss.pgh.pa.us 5859 : 2109 : HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
4403 alvherre@alvh.no-ip. 5860 [ + + ]: 40854 : for (i = 0; i < nmembers; i++)
5861 : : {
3439 andres@anarazel.de 5862 : 38745 : result = test_lockmode_for_conflict(members[i].status,
5863 : 38745 : members[i].xid,
5864 : : mode,
5865 : : &mytup,
5866 : : &needwait);
5867 : :
5868 : : /*
5869 : : * If the tuple was already locked by ourselves in a
5870 : : * previous iteration of this (say heap_lock_tuple was
5871 : : * forced to restart the locking loop because of a change
5872 : : * in xmax), then we hold the lock already on this tuple
5873 : : * version and we don't need to do anything; and this is
5874 : : * not an error condition either. We just need to skip
5875 : : * this tuple and continue locking the next version in the
5876 : : * update chain.
5877 : : */
2461 5878 [ - + ]: 38745 : if (result == TM_SelfModified)
5879 : : {
3066 alvherre@alvh.no-ip. 5880 :UBC 0 : pfree(members);
5881 : 0 : goto next;
5882 : : }
5883 : :
4403 alvherre@alvh.no-ip. 5884 [ - + ]:CBC 38745 : if (needwait)
5885 : : {
4403 alvherre@alvh.no-ip. 5886 :UBC 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
4291 5887 : 0 : XactLockTableWait(members[i].xid, rel,
5888 : : &mytup.t_self,
5889 : : XLTW_LockUpdated);
4403 5890 : 0 : pfree(members);
5891 : 0 : goto l4;
5892 : : }
2461 andres@anarazel.de 5893 [ - + ]:CBC 38745 : if (result != TM_Ok)
5894 : : {
4403 alvherre@alvh.no-ip. 5895 :UBC 0 : pfree(members);
3439 andres@anarazel.de 5896 : 0 : goto out_locked;
5897 : : }
5898 : : }
4403 alvherre@alvh.no-ip. 5899 [ + - ]:CBC 2109 : if (members)
5900 : 2109 : pfree(members);
5901 : : }
5902 : : else
5903 : : {
5904 : : MultiXactStatus status;
5905 : :
5906 : : /*
5907 : : * For a non-multi Xmax, we first need to compute the
5908 : : * corresponding MultiXactStatus by using the infomask bits.
5909 : : */
5910 [ + + ]: 29 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5911 : : {
5912 [ + - ]: 10 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5913 : 10 : status = MultiXactStatusForKeyShare;
4403 alvherre@alvh.no-ip. 5914 [ # # ]:UBC 0 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5915 : 0 : status = MultiXactStatusForShare;
5916 [ # # ]: 0 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5917 : : {
5918 [ # # ]: 0 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5919 : 0 : status = MultiXactStatusForUpdate;
5920 : : else
5921 : 0 : status = MultiXactStatusForNoKeyUpdate;
5922 : : }
5923 : : else
5924 : : {
5925 : : /*
5926 : : * LOCK_ONLY present alone (a pg_upgraded tuple marked
5927 : : * as share-locked in the old cluster) shouldn't be
5928 : : * seen in the middle of an update chain.
5929 : : */
5930 [ # # ]: 0 : elog(ERROR, "invalid lock status in tuple");
5931 : : }
5932 : : }
5933 : : else
5934 : : {
5935 : : /* it's an update, but which kind? */
4403 alvherre@alvh.no-ip. 5936 [ + + ]:CBC 19 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5937 : 14 : status = MultiXactStatusUpdate;
5938 : : else
5939 : 5 : status = MultiXactStatusNoKeyUpdate;
5940 : : }
5941 : :
3439 andres@anarazel.de 5942 : 29 : result = test_lockmode_for_conflict(status, rawxmax, mode,
5943 : : &mytup, &needwait);
5944 : :
5945 : : /*
5946 : : * If the tuple was already locked by ourselves in a previous
5947 : : * iteration of this (say heap_lock_tuple was forced to
5948 : : * restart the locking loop because of a change in xmax), then
5949 : : * we hold the lock already on this tuple version and we don't
5950 : : * need to do anything; and this is not an error condition
5951 : : * either. We just need to skip this tuple and continue
5952 : : * locking the next version in the update chain.
5953 : : */
2461 5954 [ - + ]: 29 : if (result == TM_SelfModified)
3066 alvherre@alvh.no-ip. 5955 :UBC 0 : goto next;
5956 : :
4403 alvherre@alvh.no-ip. 5957 [ + + ]:CBC 29 : if (needwait)
5958 : : {
5959 : 8 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
3969 heikki.linnakangas@i 5960 : 8 : XactLockTableWait(rawxmax, rel, &mytup.t_self,
5961 : : XLTW_LockUpdated);
4403 alvherre@alvh.no-ip. 5962 : 8 : goto l4;
5963 : : }
2461 andres@anarazel.de 5964 [ + + ]: 21 : if (result != TM_Ok)
5965 : : {
3439 5966 : 8 : goto out_locked;
5967 : : }
5968 : : }
5969 : : }
5970 : :
5971 : : /* compute the new Xmax and infomask values for the tuple ... */
4711 alvherre@alvh.no-ip. 5972 : 2181 : compute_new_xmax_infomask(xmax, old_infomask, mytup.t_data->t_infomask2,
5973 : : xid, mode, false,
5974 : : &new_xmax, &new_infomask, &new_infomask2);
5975 : :
3439 andres@anarazel.de 5976 [ - + - - ]: 2181 : if (PageIsAllVisible(BufferGetPage(buf)) &&
3439 andres@anarazel.de 5977 :UBC 0 : visibilitymap_clear(rel, block, vmbuffer,
5978 : : VISIBILITYMAP_ALL_FROZEN))
5979 : 0 : cleared_all_frozen = true;
5980 : :
4711 alvherre@alvh.no-ip. 5981 :CBC 2181 : START_CRIT_SECTION();
5982 : :
5983 : : /* ... and set them */
5984 : 2181 : HeapTupleHeaderSetXmax(mytup.t_data, new_xmax);
5985 : 2181 : mytup.t_data->t_infomask &= ~HEAP_XMAX_BITS;
5986 : 2181 : mytup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5987 : 2181 : mytup.t_data->t_infomask |= new_infomask;
5988 : 2181 : mytup.t_data->t_infomask2 |= new_infomask2;
5989 : :
5990 : 2181 : MarkBufferDirty(buf);
5991 : :
5992 : : /* XLOG stuff */
5993 [ + - + + : 2181 : if (RelationNeedsWAL(rel))
+ - + - ]
5994 : : {
5995 : : xl_heap_lock_updated xlrec;
5996 : : XLogRecPtr recptr;
3528 kgrittn@postgresql.o 5997 : 2181 : Page page = BufferGetPage(buf);
5998 : :
4045 heikki.linnakangas@i 5999 : 2181 : XLogBeginInsert();
6000 : 2181 : XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
6001 : :
6002 : 2181 : xlrec.offnum = ItemPointerGetOffsetNumber(&mytup.t_self);
4711 alvherre@alvh.no-ip. 6003 : 2181 : xlrec.xmax = new_xmax;
6004 : 2181 : xlrec.infobits_set = compute_infobits(new_infomask, new_infomask2);
3439 andres@anarazel.de 6005 : 2181 : xlrec.flags =
6006 : 2181 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
6007 : :
309 peter@eisentraut.org 6008 : 2181 : XLogRegisterData(&xlrec, SizeOfHeapLockUpdated);
6009 : :
4045 heikki.linnakangas@i 6010 : 2181 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_LOCK_UPDATED);
6011 : :
4711 alvherre@alvh.no-ip. 6012 : 2181 : PageSetLSN(page, recptr);
6013 : : }
6014 : :
6015 [ - + ]: 2181 : END_CRIT_SECTION();
6016 : :
3066 6017 : 2181 : next:
6018 : : /* if we find the end of update chain, we're done. */
4711 6019 [ + - + - ]: 4362 : if (mytup.t_data->t_infomask & HEAP_XMAX_INVALID ||
2811 andres@anarazel.de 6020 [ + + ]: 4362 : HeapTupleHeaderIndicatesMovedPartitions(mytup.t_data) ||
4585 bruce@momjian.us 6021 [ + + ]: 2185 : ItemPointerEquals(&mytup.t_self, &mytup.t_data->t_ctid) ||
4711 alvherre@alvh.no-ip. 6022 : 4 : HeapTupleHeaderIsOnlyLocked(mytup.t_data))
6023 : : {
2461 andres@anarazel.de 6024 : 2178 : result = TM_Ok;
3439 6025 : 2178 : goto out_locked;
6026 : : }
6027 : :
6028 : : /* tail recursion */
4403 alvherre@alvh.no-ip. 6029 : 3 : priorXmax = HeapTupleHeaderGetUpdateXid(mytup.t_data);
4711 6030 : 3 : ItemPointerCopy(&(mytup.t_data->t_ctid), &tupid);
6031 : 3 : UnlockReleaseBuffer(buf);
6032 : : }
6033 : :
6034 : : result = TM_Ok;
6035 : :
3439 andres@anarazel.de 6036 : 2210 : out_locked:
6037 : 2210 : UnlockReleaseBuffer(buf);
6038 : :
2847 tgl@sss.pgh.pa.us 6039 : 2210 : out_unlocked:
3439 andres@anarazel.de 6040 [ - + ]: 2210 : if (vmbuffer != InvalidBuffer)
3439 andres@anarazel.de 6041 :UBC 0 : ReleaseBuffer(vmbuffer);
6042 : :
3439 andres@anarazel.de 6043 :CBC 2210 : return result;
6044 : : }
6045 : :
6046 : : /*
6047 : : * heap_lock_updated_tuple
6048 : : * Follow update chain when locking an updated tuple, acquiring locks (row
6049 : : * marks) on the updated versions.
6050 : : *
6051 : : * The initial tuple is assumed to be already locked.
6052 : : *
6053 : : * This function doesn't check visibility, it just unconditionally marks the
6054 : : * tuple(s) as locked. If any tuple in the updated chain is being deleted
6055 : : * concurrently (or updated with the key being modified), sleep until the
6056 : : * transaction doing it is finished.
6057 : : *
6058 : : * Note that we don't acquire heavyweight tuple locks on the tuples we walk
6059 : : * when we have to wait for other transactions to release them, as opposed to
6060 : : * what heap_lock_tuple does. The reason is that having more than one
6061 : : * transaction walking the chain is probably uncommon enough that risk of
6062 : : * starvation is not likely: one of the preconditions for being here is that
6063 : : * the snapshot in use predates the update that created this tuple (because we
6064 : : * started at an earlier version of the tuple), but at the same time such a
6065 : : * transaction cannot be using repeatable read or serializable isolation
6066 : : * levels, because that would lead to a serializability failure.
6067 : : */
6068 : : static TM_Result
48 peter@eisentraut.org 6069 :GNC 2226 : heap_lock_updated_tuple(Relation rel, HeapTuple tuple, const ItemPointerData *ctid,
6070 : : TransactionId xid, LockTupleMode mode)
6071 : : {
6072 : : /*
6073 : : * If the tuple has not been updated, or has moved into another partition
6074 : : * (effectively a delete) stop here.
6075 : : */
2811 andres@anarazel.de 6076 [ + + ]:CBC 2226 : if (!HeapTupleHeaderIndicatesMovedPartitions(tuple->t_data) &&
6077 [ + + ]: 2224 : !ItemPointerEquals(&tuple->t_self, ctid))
6078 : : {
6079 : : /*
6080 : : * If this is the first possibly-multixact-able operation in the
6081 : : * current transaction, set my per-backend OldestMemberMXactId
6082 : : * setting. We can be certain that the transaction will never become a
6083 : : * member of any older MultiXactIds than that. (We have to do this
6084 : : * even if we end up just using our own TransactionId below, since
6085 : : * some other backend could incorporate our XID into a MultiXact
6086 : : * immediately afterwards.)
6087 : : */
4711 alvherre@alvh.no-ip. 6088 : 2210 : MultiXactIdSetOldestMember();
6089 : :
6090 : 2210 : return heap_lock_updated_tuple_rec(rel, ctid, xid, mode);
6091 : : }
6092 : :
6093 : : /* nothing to lock */
2461 andres@anarazel.de 6094 : 16 : return TM_Ok;
6095 : : }
6096 : :
6097 : : /*
6098 : : * heap_finish_speculative - mark speculative insertion as successful
6099 : : *
6100 : : * To successfully finish a speculative insertion we have to clear speculative
6101 : : * token from tuple. To do so the t_ctid field, which will contain a
6102 : : * speculative token value, is modified in place to point to the tuple itself,
6103 : : * which is characteristic of a newly inserted ordinary tuple.
6104 : : *
6105 : : * NB: It is not ok to commit without either finishing or aborting a
6106 : : * speculative insertion. We could treat speculative tuples of committed
6107 : : * transactions implicitly as completed, but then we would have to be prepared
6108 : : * to deal with speculative tokens on committed tuples. That wouldn't be
6109 : : * difficult - no-one looks at the ctid field of a tuple with invalid xmax -
6110 : : * but clearing the token at completion isn't very expensive either.
6111 : : * An explicit confirmation WAL record also makes logical decoding simpler.
6112 : : */
6113 : : void
48 peter@eisentraut.org 6114 :GNC 2074 : heap_finish_speculative(Relation relation, const ItemPointerData *tid)
6115 : : {
6116 : : Buffer buffer;
6117 : : Page page;
6118 : : OffsetNumber offnum;
6119 : : ItemId lp;
6120 : : HeapTupleHeader htup;
6121 : :
2461 andres@anarazel.de 6122 :CBC 2074 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
3876 6123 : 2074 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
110 peter@eisentraut.org 6124 :GNC 2074 : page = BufferGetPage(buffer);
6125 : :
2461 andres@anarazel.de 6126 :CBC 2074 : offnum = ItemPointerGetOffsetNumber(tid);
2 tgl@sss.pgh.pa.us 6127 [ + - - + ]:GNC 2074 : if (offnum < 1 || offnum > PageGetMaxOffsetNumber(page))
2 tgl@sss.pgh.pa.us 6128 [ # # ]:UNC 0 : elog(ERROR, "offnum out of range");
2 tgl@sss.pgh.pa.us 6129 :GNC 2074 : lp = PageGetItemId(page, offnum);
6130 [ - + ]: 2074 : if (!ItemIdIsNormal(lp))
3681 andres@anarazel.de 6131 [ # # ]:UBC 0 : elog(ERROR, "invalid lp");
6132 : :
3876 andres@anarazel.de 6133 :CBC 2074 : htup = (HeapTupleHeader) PageGetItem(page, lp);
6134 : :
6135 : : /* NO EREPORT(ERROR) from here till changes are logged */
6136 : 2074 : START_CRIT_SECTION();
6137 : :
2461 6138 [ - + ]: 2074 : Assert(HeapTupleHeaderIsSpeculative(htup));
6139 : :
3876 6140 : 2074 : MarkBufferDirty(buffer);
6141 : :
6142 : : /*
6143 : : * Replace the speculative insertion token with a real t_ctid, pointing to
6144 : : * itself like it does on regular tuples.
6145 : : */
2461 6146 : 2074 : htup->t_ctid = *tid;
6147 : :
6148 : : /* XLOG stuff */
3876 6149 [ + + + + : 2074 : if (RelationNeedsWAL(relation))
+ - + - ]
6150 : : {
6151 : : xl_heap_confirm xlrec;
6152 : : XLogRecPtr recptr;
6153 : :
2461 6154 : 2065 : xlrec.offnum = ItemPointerGetOffsetNumber(tid);
6155 : :
3876 6156 : 2065 : XLogBeginInsert();
6157 : :
6158 : : /* We want the same filtering on this as on a plain insert */
3282 6159 : 2065 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
6160 : :
309 peter@eisentraut.org 6161 : 2065 : XLogRegisterData(&xlrec, SizeOfHeapConfirm);
3876 andres@anarazel.de 6162 : 2065 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6163 : :
6164 : 2065 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_CONFIRM);
6165 : :
6166 : 2065 : PageSetLSN(page, recptr);
6167 : : }
6168 : :
6169 [ - + ]: 2074 : END_CRIT_SECTION();
6170 : :
6171 : 2074 : UnlockReleaseBuffer(buffer);
6172 : 2074 : }
6173 : :
6174 : : /*
6175 : : * heap_abort_speculative - kill a speculatively inserted tuple
6176 : : *
6177 : : * Marks a tuple that was speculatively inserted in the same command as dead,
6178 : : * by setting its xmin as invalid. That makes it immediately appear as dead
6179 : : * to all transactions, including our own. In particular, it makes
6180 : : * HeapTupleSatisfiesDirty() regard the tuple as dead, so that another backend
6181 : : * inserting a duplicate key value won't unnecessarily wait for our whole
6182 : : * transaction to finish (it'll just wait for our speculative insertion to
6183 : : * finish).
6184 : : *
6185 : : * Killing the tuple prevents "unprincipled deadlocks", which are deadlocks
6186 : : * that arise due to a mutual dependency that is not user visible. By
6187 : : * definition, unprincipled deadlocks cannot be prevented by the user
6188 : : * reordering lock acquisition in client code, because the implementation level
6189 : : * lock acquisitions are not under the user's direct control. If speculative
6190 : : * inserters did not take this precaution, then under high concurrency they
6191 : : * could deadlock with each other, which would not be acceptable.
6192 : : *
6193 : : * This is somewhat redundant with heap_delete, but we prefer to have a
6194 : : * dedicated routine with stripped down requirements. Note that this is also
6195 : : * used to delete the TOAST tuples created during speculative insertion.
6196 : : *
6197 : : * This routine does not affect logical decoding as it only looks at
6198 : : * confirmation records.
6199 : : */
6200 : : void
48 peter@eisentraut.org 6201 :GNC 10 : heap_abort_speculative(Relation relation, const ItemPointerData *tid)
6202 : : {
3876 andres@anarazel.de 6203 :CBC 10 : TransactionId xid = GetCurrentTransactionId();
6204 : : ItemId lp;
6205 : : HeapTupleData tp;
6206 : : Page page;
6207 : : BlockNumber block;
6208 : : Buffer buffer;
6209 : :
6210 [ - + ]: 10 : Assert(ItemPointerIsValid(tid));
6211 : :
6212 : 10 : block = ItemPointerGetBlockNumber(tid);
6213 : 10 : buffer = ReadBuffer(relation, block);
3528 kgrittn@postgresql.o 6214 : 10 : page = BufferGetPage(buffer);
6215 : :
3876 andres@anarazel.de 6216 : 10 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6217 : :
6218 : : /*
6219 : : * Page can't be all visible, we just inserted into it, and are still
6220 : : * running.
6221 : : */
6222 [ - + ]: 10 : Assert(!PageIsAllVisible(page));
6223 : :
6224 : 10 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
6225 [ - + ]: 10 : Assert(ItemIdIsNormal(lp));
6226 : :
6227 : 10 : tp.t_tableOid = RelationGetRelid(relation);
6228 : 10 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
6229 : 10 : tp.t_len = ItemIdGetLength(lp);
6230 : 10 : tp.t_self = *tid;
6231 : :
6232 : : /*
6233 : : * Sanity check that the tuple really is a speculatively inserted tuple,
6234 : : * inserted by us.
6235 : : */
6236 [ - + ]: 10 : if (tp.t_data->t_choice.t_heap.t_xmin != xid)
3876 andres@anarazel.de 6237 [ # # ]:UBC 0 : elog(ERROR, "attempted to kill a tuple inserted by another transaction");
3409 andres@anarazel.de 6238 [ + + - + ]:CBC 10 : if (!(IsToastRelation(relation) || HeapTupleHeaderIsSpeculative(tp.t_data)))
3876 andres@anarazel.de 6239 [ # # ]:UBC 0 : elog(ERROR, "attempted to kill a non-speculative tuple");
3876 andres@anarazel.de 6240 [ - + ]:CBC 10 : Assert(!HeapTupleHeaderIsHeapOnly(tp.t_data));
6241 : :
6242 : : /*
6243 : : * No need to check for serializable conflicts here. There is never a
6244 : : * need for a combo CID, either. No need to extract replica identity, or
6245 : : * do anything special with infomask bits.
6246 : : */
6247 : :
6248 : 10 : START_CRIT_SECTION();
6249 : :
6250 : : /*
6251 : : * The tuple will become DEAD immediately. Flag that this page is a
6252 : : * candidate for pruning by setting xmin to TransactionXmin. While not
6253 : : * immediately prunable, it is the oldest xid we can cheaply determine
6254 : : * that's safe against wraparound / being older than the table's
6255 : : * relfrozenxid. To defend against the unlikely case of a new relation
6256 : : * having a newer relfrozenxid than our TransactionXmin, use relfrozenxid
6257 : : * if so (vacuum can't subsequently move relfrozenxid to beyond
6258 : : * TransactionXmin, so there's no race here).
6259 : : */
2082 6260 [ - + ]: 10 : Assert(TransactionIdIsValid(TransactionXmin));
6261 : : {
597 noah@leadboat.com 6262 : 10 : TransactionId relfrozenxid = relation->rd_rel->relfrozenxid;
6263 : : TransactionId prune_xid;
6264 : :
6265 [ - + ]: 10 : if (TransactionIdPrecedes(TransactionXmin, relfrozenxid))
597 noah@leadboat.com 6266 :UBC 0 : prune_xid = relfrozenxid;
6267 : : else
597 noah@leadboat.com 6268 :CBC 10 : prune_xid = TransactionXmin;
6269 [ - + + + : 10 : PageSetPrunable(page, prune_xid);
- + ]
6270 : : }
6271 : :
6272 : : /* store transaction information of xact deleting the tuple */
3876 andres@anarazel.de 6273 : 10 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
6274 : 10 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6275 : :
6276 : : /*
6277 : : * Set the tuple header xmin to InvalidTransactionId. This makes the
6278 : : * tuple immediately invisible everyone. (In particular, to any
6279 : : * transactions waiting on the speculative token, woken up later.)
6280 : : */
6281 : 10 : HeapTupleHeaderSetXmin(tp.t_data, InvalidTransactionId);
6282 : :
6283 : : /* Clear the speculative insertion token too */
6284 : 10 : tp.t_data->t_ctid = tp.t_self;
6285 : :
6286 : 10 : MarkBufferDirty(buffer);
6287 : :
6288 : : /*
6289 : : * XLOG stuff
6290 : : *
6291 : : * The WAL records generated here match heap_delete(). The same recovery
6292 : : * routines are used.
6293 : : */
6294 [ + - - + : 10 : if (RelationNeedsWAL(relation))
- - - - ]
6295 : : {
6296 : : xl_heap_delete xlrec;
6297 : : XLogRecPtr recptr;
6298 : :
6299 : 10 : xlrec.flags = XLH_DELETE_IS_SUPER;
6300 : 20 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
6301 : 10 : tp.t_data->t_infomask2);
6302 : 10 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
6303 : 10 : xlrec.xmax = xid;
6304 : :
6305 : 10 : XLogBeginInsert();
309 peter@eisentraut.org 6306 : 10 : XLogRegisterData(&xlrec, SizeOfHeapDelete);
3876 andres@anarazel.de 6307 : 10 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6308 : :
6309 : : /* No replica identity & replication origin logged */
6310 : :
6311 : 10 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
6312 : :
6313 : 10 : PageSetLSN(page, recptr);
6314 : : }
6315 : :
6316 [ - + ]: 10 : END_CRIT_SECTION();
6317 : :
6318 : 10 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6319 : :
6320 [ + + ]: 10 : if (HeapTupleHasExternal(&tp))
6321 : : {
3409 6322 [ - + ]: 1 : Assert(!IsToastRelation(relation));
2266 rhaas@postgresql.org 6323 : 1 : heap_toast_delete(relation, &tp, true);
6324 : : }
6325 : :
6326 : : /*
6327 : : * Never need to mark tuple for invalidation, since catalogs don't support
6328 : : * speculative insertion
6329 : : */
6330 : :
6331 : : /* Now we can release the buffer */
3876 andres@anarazel.de 6332 : 10 : ReleaseBuffer(buffer);
6333 : :
6334 : : /* count deletion, as we counted the insertion too */
6335 : 10 : pgstat_count_heap_delete(relation);
6336 : 10 : }
6337 : :
6338 : : /*
6339 : : * heap_inplace_lock - protect inplace update from concurrent heap_update()
6340 : : *
6341 : : * Evaluate whether the tuple's state is compatible with a no-key update.
6342 : : * Current transaction rowmarks are fine, as is KEY SHARE from any
6343 : : * transaction. If compatible, return true with the buffer exclusive-locked,
6344 : : * and the caller must release that by calling
6345 : : * heap_inplace_update_and_unlock(), calling heap_inplace_unlock(), or raising
6346 : : * an error. Otherwise, call release_callback(arg), wait for blocking
6347 : : * transactions to end, and return false.
6348 : : *
6349 : : * Since this is intended for system catalogs and SERIALIZABLE doesn't cover
6350 : : * DDL, this doesn't guarantee any particular predicate locking.
6351 : : *
6352 : : * heap_delete() is a rarer source of blocking transactions (xwait). We'll
6353 : : * wait for such a transaction just like for the normal heap_update() case.
6354 : : * Normal concurrent DROP commands won't cause that, because all inplace
6355 : : * updaters take some lock that conflicts with DROP. An explicit SQL "DELETE
6356 : : * FROM pg_class" can cause it. By waiting, if the concurrent transaction
6357 : : * executed both "DELETE FROM pg_class" and "INSERT INTO pg_class", our caller
6358 : : * can find the successor tuple.
6359 : : *
6360 : : * Readers of inplace-updated fields expect changes to those fields are
6361 : : * durable. For example, vac_truncate_clog() reads datfrozenxid from
6362 : : * pg_database tuples via catalog snapshots. A future snapshot must not
6363 : : * return a lower datfrozenxid for the same database OID (lower in the
6364 : : * FullTransactionIdPrecedes() sense). We achieve that since no update of a
6365 : : * tuple can start while we hold a lock on its buffer. In cases like
6366 : : * BEGIN;GRANT;CREATE INDEX;COMMIT we're inplace-updating a tuple visible only
6367 : : * to this transaction. ROLLBACK then is one case where it's okay to lose
6368 : : * inplace updates. (Restoring relhasindex=false on ROLLBACK is fine, since
6369 : : * any concurrent CREATE INDEX would have blocked, then inplace-updated the
6370 : : * committed tuple.)
6371 : : *
6372 : : * In principle, we could avoid waiting by overwriting every tuple in the
6373 : : * updated tuple chain. Reader expectations permit updating a tuple only if
6374 : : * it's aborted, is the tail of the chain, or we already updated the tuple
6375 : : * referenced in its t_ctid. Hence, we would need to overwrite the tuples in
6376 : : * order from tail to head. That would imply either (a) mutating all tuples
6377 : : * in one critical section or (b) accepting a chance of partial completion.
6378 : : * Partial completion of a relfrozenxid update would have the weird
6379 : : * consequence that the table's next VACUUM could see the table's relfrozenxid
6380 : : * move forward between vacuum_get_cutoffs() and finishing.
6381 : : */
6382 : : bool
449 noah@leadboat.com 6383 : 94161 : heap_inplace_lock(Relation relation,
6384 : : HeapTuple oldtup_ptr, Buffer buffer,
6385 : : void (*release_callback) (void *), void *arg)
6386 : : {
6387 : 94161 : HeapTupleData oldtup = *oldtup_ptr; /* minimize diff vs. heap_update() */
6388 : : TM_Result result;
6389 : : bool ret;
6390 : :
6391 : : #ifdef USE_ASSERT_CHECKING
6392 [ + + ]: 94161 : if (RelationGetRelid(relation) == RelationRelationId)
6393 : 93207 : check_inplace_rel_lock(oldtup_ptr);
6394 : : #endif
6395 : :
6396 [ - + ]: 94161 : Assert(BufferIsValid(buffer));
6397 : :
6398 : : /*
6399 : : * Register shared cache invals if necessary. Other sessions may finish
6400 : : * inplace updates of this tuple between this step and LockTuple(). Since
6401 : : * inplace updates don't change cache keys, that's harmless.
6402 : : *
6403 : : * While it's tempting to register invals only after confirming we can
6404 : : * return true, the following obstacle precludes reordering steps that
6405 : : * way. Registering invals might reach a CatalogCacheInitializeCache()
6406 : : * that locks "buffer". That would hang indefinitely if running after our
6407 : : * own LockBuffer(). Hence, we must register invals before LockBuffer().
6408 : : */
2 6409 : 94161 : CacheInvalidateHeapTupleInplace(relation, oldtup_ptr);
6410 : :
449 6411 : 94161 : LockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
6412 : 94161 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
6413 : :
6414 : : /*----------
6415 : : * Interpret HeapTupleSatisfiesUpdate() like heap_update() does, except:
6416 : : *
6417 : : * - wait unconditionally
6418 : : * - already locked tuple above, since inplace needs that unconditionally
6419 : : * - don't recheck header after wait: simpler to defer to next iteration
6420 : : * - don't try to continue even if the updater aborts: likewise
6421 : : * - no crosscheck
6422 : : */
6423 : 94161 : result = HeapTupleSatisfiesUpdate(&oldtup, GetCurrentCommandId(false),
6424 : : buffer);
6425 : :
6426 [ - + ]: 94161 : if (result == TM_Invisible)
6427 : : {
6428 : : /* no known way this can happen */
3884 rhaas@postgresql.org 6429 [ # # ]:UBC 0 : ereport(ERROR,
6430 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6431 : : errmsg_internal("attempted to overwrite invisible tuple")));
6432 : : }
449 noah@leadboat.com 6433 [ - + ]:CBC 94161 : else if (result == TM_SelfModified)
6434 : : {
6435 : : /*
6436 : : * CREATE INDEX might reach this if an expression is silly enough to
6437 : : * call e.g. SELECT ... FROM pg_class FOR SHARE. C code of other SQL
6438 : : * statements might get here after a heap_update() of the same row, in
6439 : : * the absence of an intervening CommandCounterIncrement().
6440 : : */
449 noah@leadboat.com 6441 [ # # ]:UBC 0 : ereport(ERROR,
6442 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
6443 : : errmsg("tuple to be updated was already modified by an operation triggered by the current command")));
6444 : : }
449 noah@leadboat.com 6445 [ + + ]:CBC 94161 : else if (result == TM_BeingModified)
6446 : : {
6447 : : TransactionId xwait;
6448 : : uint16 infomask;
6449 : :
6450 : 56 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
6451 : 56 : infomask = oldtup.t_data->t_infomask;
6452 : :
6453 [ + + ]: 56 : if (infomask & HEAP_XMAX_IS_MULTI)
6454 : : {
6455 : 5 : LockTupleMode lockmode = LockTupleNoKeyExclusive;
6456 : 5 : MultiXactStatus mxact_status = MultiXactStatusNoKeyUpdate;
6457 : : int remain;
6458 : :
6459 [ + + ]: 5 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
6460 : : lockmode, NULL))
6461 : : {
6462 : 2 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
414 6463 : 2 : release_callback(arg);
449 6464 : 2 : ret = false;
6465 : 2 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
6466 : : relation, &oldtup.t_self, XLTW_Update,
6467 : : &remain);
6468 : : }
6469 : : else
6470 : 3 : ret = true;
6471 : : }
6472 [ + + ]: 51 : else if (TransactionIdIsCurrentTransactionId(xwait))
6473 : 1 : ret = true;
6474 [ + + ]: 50 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
6475 : 1 : ret = true;
6476 : : else
6477 : : {
6478 : 49 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
414 6479 : 49 : release_callback(arg);
449 6480 : 49 : ret = false;
6481 : 49 : XactLockTableWait(xwait, relation, &oldtup.t_self,
6482 : : XLTW_Update);
6483 : : }
6484 : : }
6485 : : else
6486 : : {
6487 : 94105 : ret = (result == TM_Ok);
6488 [ - + ]: 94105 : if (!ret)
6489 : : {
449 noah@leadboat.com 6490 :UBC 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
414 6491 : 0 : release_callback(arg);
6492 : : }
6493 : : }
6494 : :
6495 : : /*
6496 : : * GetCatalogSnapshot() relies on invalidation messages to know when to
6497 : : * take a new snapshot. COMMIT of xwait is responsible for sending the
6498 : : * invalidation. We're not acquiring heavyweight locks sufficient to
6499 : : * block if not yet sent, so we must take a new snapshot to ensure a later
6500 : : * attempt has a fair chance. While we don't need this if xwait aborted,
6501 : : * don't bother optimizing that.
6502 : : */
449 noah@leadboat.com 6503 [ + + ]:CBC 94161 : if (!ret)
6504 : : {
6505 : 51 : UnlockTuple(relation, &oldtup.t_self, InplaceUpdateTupleLock);
410 6506 : 51 : ForgetInplace_Inval();
449 6507 : 51 : InvalidateCatalogSnapshot();
6508 : : }
6509 : 94161 : return ret;
6510 : : }
6511 : :
6512 : : /*
6513 : : * heap_inplace_update_and_unlock - core of systable_inplace_update_finish
6514 : : *
6515 : : * The tuple cannot change size, and therefore its header fields and null
6516 : : * bitmap (if any) don't change either.
6517 : : *
6518 : : * Since we hold LOCKTAG_TUPLE, no updater has a local copy of this tuple.
6519 : : */
6520 : : void
6521 : 65225 : heap_inplace_update_and_unlock(Relation relation,
6522 : : HeapTuple oldtup, HeapTuple tuple,
6523 : : Buffer buffer)
6524 : : {
6525 : 65225 : HeapTupleHeader htup = oldtup->t_data;
6526 : : uint32 oldlen;
6527 : : uint32 newlen;
6528 : : char *dst;
6529 : : char *src;
418 6530 : 65225 : int nmsgs = 0;
6531 : 65225 : SharedInvalidationMessage *invalMessages = NULL;
6532 : 65225 : bool RelcacheInitFileInval = false;
6533 : :
449 6534 [ - + ]: 65225 : Assert(ItemPointerEquals(&oldtup->t_self, &tuple->t_self));
6535 : 65225 : oldlen = oldtup->t_len - htup->t_hoff;
7161 tgl@sss.pgh.pa.us 6536 : 65225 : newlen = tuple->t_len - tuple->t_data->t_hoff;
6537 [ + - - + ]: 65225 : if (oldlen != newlen || htup->t_hoff != tuple->t_data->t_hoff)
3681 andres@anarazel.de 6538 [ # # ]:UBC 0 : elog(ERROR, "wrong tuple length");
6539 : :
418 noah@leadboat.com 6540 :CBC 65225 : dst = (char *) htup + htup->t_hoff;
6541 : 65225 : src = (char *) tuple->t_data + tuple->t_data->t_hoff;
6542 : :
6543 : : /* Like RecordTransactionCommit(), log only if needed */
6544 [ + + ]: 65225 : if (XLogStandbyInfoActive())
6545 : 50990 : nmsgs = inplaceGetInvalidationMessages(&invalMessages,
6546 : : &RelcacheInitFileInval);
6547 : :
6548 : : /*
6549 : : * Unlink relcache init files as needed. If unlinking, acquire
6550 : : * RelCacheInitLock until after associated invalidations. By doing this
6551 : : * in advance, if we checkpoint and then crash between inplace
6552 : : * XLogInsert() and inval, we don't rely on StartupXLOG() ->
6553 : : * RelationCacheInitFileRemove(). That uses elevel==LOG, so replay would
6554 : : * neglect to PANIC on EIO.
6555 : : */
6556 : 65225 : PreInplace_Inval();
6557 : :
6558 : : /*----------
6559 : : * NO EREPORT(ERROR) from here till changes are complete
6560 : : *
6561 : : * Our buffer lock won't stop a reader having already pinned and checked
6562 : : * visibility for this tuple. Hence, we write WAL first, then mutate the
6563 : : * buffer. Like in MarkBufferDirtyHint() or RecordTransactionCommit(),
6564 : : * checkpoint delay makes that acceptable. With the usual order of
6565 : : * changes, a crash after memcpy() and before XLogInsert() could allow
6566 : : * datfrozenxid to overtake relfrozenxid:
6567 : : *
6568 : : * ["D" is a VACUUM (ONLY_DATABASE_STATS)]
6569 : : * ["R" is a VACUUM tbl]
6570 : : * D: vac_update_datfrozenxid() -> systable_beginscan(pg_class)
6571 : : * D: systable_getnext() returns pg_class tuple of tbl
6572 : : * R: memcpy() into pg_class tuple of tbl
6573 : : * D: raise pg_database.datfrozenxid, XLogInsert(), finish
6574 : : * [crash]
6575 : : * [recovery restores datfrozenxid w/o relfrozenxid]
6576 : : *
6577 : : * Mimic MarkBufferDirtyHint() subroutine XLogSaveBufferForHint().
6578 : : * Specifically, use DELAY_CHKPT_START, and copy the buffer to the stack.
6579 : : * The stack copy facilitates a FPI of the post-mutation block before we
6580 : : * accept other sessions seeing it. DELAY_CHKPT_START allows us to
6581 : : * XLogInsert() before MarkBufferDirty(). Since XLogSaveBufferForHint()
6582 : : * can operate under BUFFER_LOCK_SHARED, it can't avoid DELAY_CHKPT_START.
6583 : : * This function, however, likely could avoid it with the following order
6584 : : * of operations: MarkBufferDirty(), XLogInsert(), memcpy(). Opt to use
6585 : : * DELAY_CHKPT_START here, too, as a way to have fewer distinct code
6586 : : * patterns to analyze. Inplace update isn't so frequent that it should
6587 : : * pursue the small optimization of skipping DELAY_CHKPT_START.
6588 : : */
6589 [ - + ]: 65225 : Assert((MyProc->delayChkptFlags & DELAY_CHKPT_START) == 0);
6590 : 65225 : START_CRIT_SECTION();
6591 : 65225 : MyProc->delayChkptFlags |= DELAY_CHKPT_START;
6592 : :
6593 : : /* XLOG stuff */
5483 rhaas@postgresql.org 6594 [ + - + + : 65225 : if (RelationNeedsWAL(relation))
+ - + + ]
6595 : : {
6596 : : xl_heap_inplace xlrec;
6597 : : PGAlignedBlock copied_buffer;
418 noah@leadboat.com 6598 : 65217 : char *origdata = (char *) BufferGetBlock(buffer);
6599 : 65217 : Page page = BufferGetPage(buffer);
6600 : 65217 : uint16 lower = ((PageHeader) page)->pd_lower;
6601 : 65217 : uint16 upper = ((PageHeader) page)->pd_upper;
6602 : : uintptr_t dst_offset_in_block;
6603 : : RelFileLocator rlocator;
6604 : : ForkNumber forkno;
6605 : : BlockNumber blkno;
6606 : : XLogRecPtr recptr;
6607 : :
4045 heikki.linnakangas@i 6608 : 65217 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
418 noah@leadboat.com 6609 : 65217 : xlrec.dbId = MyDatabaseId;
6610 : 65217 : xlrec.tsId = MyDatabaseTableSpace;
6611 : 65217 : xlrec.relcacheInitFileInval = RelcacheInitFileInval;
6612 : 65217 : xlrec.nmsgs = nmsgs;
6613 : :
4045 heikki.linnakangas@i 6614 : 65217 : XLogBeginInsert();
309 peter@eisentraut.org 6615 : 65217 : XLogRegisterData(&xlrec, MinSizeOfHeapInplace);
418 noah@leadboat.com 6616 [ + + ]: 65217 : if (nmsgs != 0)
309 peter@eisentraut.org 6617 : 36047 : XLogRegisterData(invalMessages,
6618 : : nmsgs * sizeof(SharedInvalidationMessage));
6619 : :
6620 : : /* register block matching what buffer will look like after changes */
418 noah@leadboat.com 6621 : 65217 : memcpy(copied_buffer.data, origdata, lower);
6622 : 65217 : memcpy(copied_buffer.data + upper, origdata + upper, BLCKSZ - upper);
6623 : 65217 : dst_offset_in_block = dst - origdata;
6624 : 65217 : memcpy(copied_buffer.data + dst_offset_in_block, src, newlen);
6625 : 65217 : BufferGetTag(buffer, &rlocator, &forkno, &blkno);
6626 [ - + ]: 65217 : Assert(forkno == MAIN_FORKNUM);
6627 : 65217 : XLogRegisterBlock(0, &rlocator, forkno, blkno, copied_buffer.data,
6628 : : REGBUF_STANDARD);
6629 : 65217 : XLogRegisterBufData(0, src, newlen);
6630 : :
6631 : : /* inplace updates aren't decoded atm, don't log the origin */
6632 : :
4045 heikki.linnakangas@i 6633 : 65217 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_INPLACE);
6634 : :
418 noah@leadboat.com 6635 : 65217 : PageSetLSN(page, recptr);
6636 : : }
6637 : :
6638 : 65225 : memcpy(dst, src, newlen);
6639 : :
6640 : 65225 : MarkBufferDirty(buffer);
6641 : :
6642 : 65225 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6643 : :
6644 : : /*
6645 : : * Send invalidations to shared queue. SearchSysCacheLocked1() assumes we
6646 : : * do this before UnlockTuple().
6647 : : */
6648 : 65225 : AtInplace_Inval();
6649 : :
6650 : 65225 : MyProc->delayChkptFlags &= ~DELAY_CHKPT_START;
7161 tgl@sss.pgh.pa.us 6651 [ - + ]: 65225 : END_CRIT_SECTION();
418 noah@leadboat.com 6652 : 65225 : UnlockTuple(relation, &tuple->t_self, InplaceUpdateTupleLock);
6653 : :
6654 : 65225 : AcceptInvalidationMessages(); /* local processing of just-sent inval */
6655 : :
6656 : : /*
6657 : : * Queue a transactional inval, for logical decoding and for third-party
6658 : : * code that might have been relying on it since long before inplace
6659 : : * update adopted immediate invalidation. See README.tuplock section
6660 : : * "Reading inplace-updated columns" for logical decoding details.
6661 : : */
7161 tgl@sss.pgh.pa.us 6662 [ + + ]: 65225 : if (!IsBootstrapProcessingMode())
5237 6663 : 50282 : CacheInvalidateHeapTuple(relation, tuple, NULL);
7161 6664 : 65225 : }
6665 : :
6666 : : /*
6667 : : * heap_inplace_unlock - reverse of heap_inplace_lock
6668 : : */
6669 : : void
449 noah@leadboat.com 6670 : 28885 : heap_inplace_unlock(Relation relation,
6671 : : HeapTuple oldtup, Buffer buffer)
6672 : : {
6673 : 28885 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6674 : 28885 : UnlockTuple(relation, &oldtup->t_self, InplaceUpdateTupleLock);
410 6675 : 28885 : ForgetInplace_Inval();
449 6676 : 28885 : }
6677 : :
6678 : : #define FRM_NOOP 0x0001
6679 : : #define FRM_INVALIDATE_XMAX 0x0002
6680 : : #define FRM_RETURN_IS_XID 0x0004
6681 : : #define FRM_RETURN_IS_MULTI 0x0008
6682 : : #define FRM_MARK_COMMITTED 0x0010
6683 : :
6684 : : /*
6685 : : * FreezeMultiXactId
6686 : : * Determine what to do during freezing when a tuple is marked by a
6687 : : * MultiXactId.
6688 : : *
6689 : : * "flags" is an output value; it's used to tell caller what to do on return.
6690 : : * "pagefrz" is an input/output value, used to manage page level freezing.
6691 : : *
6692 : : * Possible values that we can set in "flags":
6693 : : * FRM_NOOP
6694 : : * don't do anything -- keep existing Xmax
6695 : : * FRM_INVALIDATE_XMAX
6696 : : * mark Xmax as InvalidTransactionId and set XMAX_INVALID flag.
6697 : : * FRM_RETURN_IS_XID
6698 : : * The Xid return value is a single update Xid to set as xmax.
6699 : : * FRM_MARK_COMMITTED
6700 : : * Xmax can be marked as HEAP_XMAX_COMMITTED
6701 : : * FRM_RETURN_IS_MULTI
6702 : : * The return value is a new MultiXactId to set as new Xmax.
6703 : : * (caller must obtain proper infomask bits using GetMultiXactIdHintBits)
6704 : : *
6705 : : * Caller delegates control of page freezing to us. In practice we always
6706 : : * force freezing of caller's page unless FRM_NOOP processing is indicated.
6707 : : * We help caller ensure that XIDs < FreezeLimit and MXIDs < MultiXactCutoff
6708 : : * can never be left behind. We freely choose when and how to process each
6709 : : * Multi, without ever violating the cutoff postconditions for freezing.
6710 : : *
6711 : : * It's useful to remove Multis on a proactive timeline (relative to freezing
6712 : : * XIDs) to keep MultiXact member SLRU buffer misses to a minimum. It can also
6713 : : * be cheaper in the short run, for us, since we too can avoid SLRU buffer
6714 : : * misses through eager processing.
6715 : : *
6716 : : * NB: Creates a _new_ MultiXactId when FRM_RETURN_IS_MULTI is set, though only
6717 : : * when FreezeLimit and/or MultiXactCutoff cutoffs leave us with no choice.
6718 : : * This can usually be put off, which is usually enough to avoid it altogether.
6719 : : * Allocating new multis during VACUUM should be avoided on general principle;
6720 : : * only VACUUM can advance relminmxid, so allocating new Multis here comes with
6721 : : * its own special risks.
6722 : : *
6723 : : * NB: Caller must maintain "no freeze" NewRelfrozenXid/NewRelminMxid trackers
6724 : : * using heap_tuple_should_freeze when we haven't forced page-level freezing.
6725 : : *
6726 : : * NB: Caller should avoid needlessly calling heap_tuple_should_freeze when we
6727 : : * have already forced page-level freezing, since that might incur the same
6728 : : * SLRU buffer misses that we specifically intended to avoid by freezing.
6729 : : */
6730 : : static TransactionId
4384 alvherre@alvh.no-ip. 6731 : 6 : FreezeMultiXactId(MultiXactId multi, uint16 t_infomask,
6732 : : const struct VacuumCutoffs *cutoffs, uint16 *flags,
6733 : : HeapPageFreeze *pagefrz)
6734 : : {
6735 : : TransactionId newxmax;
6736 : : MultiXactMember *members;
6737 : : int nmembers;
6738 : : bool need_replace;
6739 : : int nnewmembers;
6740 : : MultiXactMember *newmembers;
6741 : : bool has_lockers;
6742 : : TransactionId update_xid;
6743 : : bool update_committed;
6744 : : TransactionId FreezePageRelfrozenXid;
6745 : :
6746 : 6 : *flags = 0;
6747 : :
6748 : : /* We should only be called in Multis */
6749 [ - + ]: 6 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
6750 : :
3463 6751 [ + - - + ]: 12 : if (!MultiXactIdIsValid(multi) ||
6752 : 6 : HEAP_LOCKED_UPGRADED(t_infomask))
6753 : : {
4384 alvherre@alvh.no-ip. 6754 :UBC 0 : *flags |= FRM_INVALIDATE_XMAX;
1085 pg@bowt.ie 6755 : 0 : pagefrz->freeze_required = true;
4384 alvherre@alvh.no-ip. 6756 : 0 : return InvalidTransactionId;
6757 : : }
1091 pg@bowt.ie 6758 [ - + ]:CBC 6 : else if (MultiXactIdPrecedes(multi, cutoffs->relminmxid))
2956 andres@anarazel.de 6759 [ # # ]:UBC 0 : ereport(ERROR,
6760 : : (errcode(ERRCODE_DATA_CORRUPTED),
6761 : : errmsg_internal("found multixact %u from before relminmxid %u",
6762 : : multi, cutoffs->relminmxid)));
1085 pg@bowt.ie 6763 [ + + ]:CBC 6 : else if (MultiXactIdPrecedes(multi, cutoffs->OldestMxact))
6764 : : {
6765 : : TransactionId update_xact;
6766 : :
6767 : : /*
6768 : : * This old multi cannot possibly have members still running, but
6769 : : * verify just in case. If it was a locker only, it can be removed
6770 : : * without any further consideration; but if it contained an update,
6771 : : * we might need to preserve it.
6772 : : */
2956 andres@anarazel.de 6773 [ - + ]: 4 : if (MultiXactIdIsRunning(multi,
6774 : 4 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask)))
2956 andres@anarazel.de 6775 [ # # ]:UBC 0 : ereport(ERROR,
6776 : : (errcode(ERRCODE_DATA_CORRUPTED),
6777 : : errmsg_internal("multixact %u from before multi freeze cutoff %u found to be still running",
6778 : : multi, cutoffs->OldestMxact)));
6779 : :
4384 alvherre@alvh.no-ip. 6780 [ + - ]:CBC 4 : if (HEAP_XMAX_IS_LOCKED_ONLY(t_infomask))
6781 : : {
6782 : 4 : *flags |= FRM_INVALIDATE_XMAX;
1085 pg@bowt.ie 6783 : 4 : pagefrz->freeze_required = true;
6784 : 4 : return InvalidTransactionId;
6785 : : }
6786 : :
6787 : : /* replace multi with single XID for its updater? */
1085 pg@bowt.ie 6788 :UBC 0 : update_xact = MultiXactIdGetUpdateXid(multi, t_infomask);
6789 [ # # ]: 0 : if (TransactionIdPrecedes(update_xact, cutoffs->relfrozenxid))
6790 [ # # ]: 0 : ereport(ERROR,
6791 : : (errcode(ERRCODE_DATA_CORRUPTED),
6792 : : errmsg_internal("multixact %u contains update XID %u from before relfrozenxid %u",
6793 : : multi, update_xact,
6794 : : cutoffs->relfrozenxid)));
6795 [ # # ]: 0 : else if (TransactionIdPrecedes(update_xact, cutoffs->OldestXmin))
6796 : : {
6797 : : /*
6798 : : * Updater XID has to have aborted (otherwise the tuple would have
6799 : : * been pruned away instead, since updater XID is < OldestXmin).
6800 : : * Just remove xmax.
6801 : : */
1079 6802 [ # # ]: 0 : if (TransactionIdDidCommit(update_xact))
1085 6803 [ # # ]: 0 : ereport(ERROR,
6804 : : (errcode(ERRCODE_DATA_CORRUPTED),
6805 : : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6806 : : multi, update_xact,
6807 : : cutoffs->OldestXmin)));
6808 : 0 : *flags |= FRM_INVALIDATE_XMAX;
6809 : 0 : pagefrz->freeze_required = true;
6810 : 0 : return InvalidTransactionId;
6811 : : }
6812 : :
6813 : : /* Have to keep updater XID as new xmax */
6814 : 0 : *flags |= FRM_RETURN_IS_XID;
6815 : 0 : pagefrz->freeze_required = true;
6816 : 0 : return update_xact;
6817 : : }
6818 : :
6819 : : /*
6820 : : * Some member(s) of this Multi may be below FreezeLimit xid cutoff, so we
6821 : : * need to walk the whole members array to figure out what to do, if
6822 : : * anything.
6823 : : */
6824 : : nmembers =
3463 alvherre@alvh.no-ip. 6825 :CBC 2 : GetMultiXactIdMembers(multi, &members, false,
4159 6826 : 2 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask));
4384 6827 [ - + ]: 2 : if (nmembers <= 0)
6828 : : {
6829 : : /* Nothing worth keeping */
4384 alvherre@alvh.no-ip. 6830 :UBC 0 : *flags |= FRM_INVALIDATE_XMAX;
1085 pg@bowt.ie 6831 : 0 : pagefrz->freeze_required = true;
4384 alvherre@alvh.no-ip. 6832 : 0 : return InvalidTransactionId;
6833 : : }
6834 : :
6835 : : /*
6836 : : * The FRM_NOOP case is the only case where we might need to ratchet back
6837 : : * FreezePageRelfrozenXid or FreezePageRelminMxid. It is also the only
6838 : : * case where our caller might ratchet back its NoFreezePageRelfrozenXid
6839 : : * or NoFreezePageRelminMxid "no freeze" trackers to deal with a multi.
6840 : : * FRM_NOOP handling should result in the NewRelfrozenXid/NewRelminMxid
6841 : : * trackers managed by VACUUM being ratcheting back by xmax to the degree
6842 : : * required to make it safe to leave xmax undisturbed, independent of
6843 : : * whether or not page freezing is triggered somewhere else.
6844 : : *
6845 : : * Our policy is to force freezing in every case other than FRM_NOOP,
6846 : : * which obviates the need to maintain either set of trackers, anywhere.
6847 : : * Every other case will reliably execute a freeze plan for xmax that
6848 : : * either replaces xmax with an XID/MXID >= OldestXmin/OldestMxact, or
6849 : : * sets xmax to an InvalidTransactionId XID, rendering xmax fully frozen.
6850 : : * (VACUUM's NewRelfrozenXid/NewRelminMxid trackers are initialized with
6851 : : * OldestXmin/OldestMxact, so later values never need to be tracked here.)
6852 : : */
4384 alvherre@alvh.no-ip. 6853 :CBC 2 : need_replace = false;
1085 pg@bowt.ie 6854 : 2 : FreezePageRelfrozenXid = pagefrz->FreezePageRelfrozenXid;
1091 6855 [ + + ]: 4 : for (int i = 0; i < nmembers; i++)
6856 : : {
6857 : 3 : TransactionId xid = members[i].xid;
6858 : :
6859 [ - + ]: 3 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6860 : :
6861 [ + + ]: 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
6862 : : {
6863 : : /* Can't violate the FreezeLimit postcondition */
4384 alvherre@alvh.no-ip. 6864 : 1 : need_replace = true;
6865 : 1 : break;
6866 : : }
1085 pg@bowt.ie 6867 [ - + ]: 2 : if (TransactionIdPrecedes(xid, FreezePageRelfrozenXid))
1085 pg@bowt.ie 6868 :UBC 0 : FreezePageRelfrozenXid = xid;
6869 : : }
6870 : :
6871 : : /* Can't violate the MultiXactCutoff postcondition, either */
1085 pg@bowt.ie 6872 [ + + ]:CBC 2 : if (!need_replace)
6873 : 1 : need_replace = MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff);
6874 : :
4384 alvherre@alvh.no-ip. 6875 [ + + ]: 2 : if (!need_replace)
6876 : : {
6877 : : /*
6878 : : * vacuumlazy.c might ratchet back NewRelminMxid, NewRelfrozenXid, or
6879 : : * both together to make it safe to retain this particular multi after
6880 : : * freezing its page
6881 : : */
6882 : 1 : *flags |= FRM_NOOP;
1085 pg@bowt.ie 6883 : 1 : pagefrz->FreezePageRelfrozenXid = FreezePageRelfrozenXid;
6884 [ - + ]: 1 : if (MultiXactIdPrecedes(multi, pagefrz->FreezePageRelminMxid))
1085 pg@bowt.ie 6885 :UBC 0 : pagefrz->FreezePageRelminMxid = multi;
4384 alvherre@alvh.no-ip. 6886 :CBC 1 : pfree(members);
1354 pg@bowt.ie 6887 : 1 : return multi;
6888 : : }
6889 : :
6890 : : /*
6891 : : * Do a more thorough second pass over the multi to figure out which
6892 : : * member XIDs actually need to be kept. Checking the precise status of
6893 : : * individual members might even show that we don't need to keep anything.
6894 : : * That is quite possible even though the Multi must be >= OldestMxact,
6895 : : * since our second pass only keeps member XIDs when it's truly necessary;
6896 : : * even member XIDs >= OldestXmin often won't be kept by second pass.
6897 : : */
4384 alvherre@alvh.no-ip. 6898 : 1 : nnewmembers = 0;
7 michael@paquier.xyz 6899 :GNC 1 : newmembers = palloc_array(MultiXactMember, nmembers);
4384 alvherre@alvh.no-ip. 6900 :CBC 1 : has_lockers = false;
6901 : 1 : update_xid = InvalidTransactionId;
6902 : 1 : update_committed = false;
6903 : :
6904 : : /*
6905 : : * Determine whether to keep each member xid, or to ignore it instead
6906 : : */
1091 pg@bowt.ie 6907 [ + + ]: 3 : for (int i = 0; i < nmembers; i++)
6908 : : {
6909 : 2 : TransactionId xid = members[i].xid;
6910 : 2 : MultiXactStatus mstatus = members[i].status;
6911 : :
6912 [ - + ]: 2 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6913 : :
6914 [ + - ]: 2 : if (!ISUPDATE_from_mxstatus(mstatus))
6915 : : {
6916 : : /*
6917 : : * Locker XID (not updater XID). We only keep lockers that are
6918 : : * still running.
6919 : : */
6920 [ + - + + ]: 4 : if (TransactionIdIsCurrentTransactionId(xid) ||
6921 : 2 : TransactionIdIsInProgress(xid))
6922 : : {
1085 6923 [ - + ]: 1 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
1085 pg@bowt.ie 6924 [ # # ]:UBC 0 : ereport(ERROR,
6925 : : (errcode(ERRCODE_DATA_CORRUPTED),
6926 : : errmsg_internal("multixact %u contains running locker XID %u from before removable cutoff %u",
6927 : : multi, xid,
6928 : : cutoffs->OldestXmin)));
1091 pg@bowt.ie 6929 :CBC 1 : newmembers[nnewmembers++] = members[i];
6930 : 1 : has_lockers = true;
6931 : : }
6932 : :
6933 : 2 : continue;
6934 : : }
6935 : :
6936 : : /*
6937 : : * Updater XID (not locker XID). Should we keep it?
6938 : : *
6939 : : * Since the tuple wasn't totally removed when vacuum pruned, the
6940 : : * update Xid cannot possibly be older than OldestXmin cutoff unless
6941 : : * the updater XID aborted. If the updater transaction is known
6942 : : * aborted or crashed then it's okay to ignore it, otherwise not.
6943 : : *
6944 : : * In any case the Multi should never contain two updaters, whatever
6945 : : * their individual commit status. Check for that first, in passing.
6946 : : */
1091 pg@bowt.ie 6947 [ # # ]:UBC 0 : if (TransactionIdIsValid(update_xid))
6948 [ # # ]: 0 : ereport(ERROR,
6949 : : (errcode(ERRCODE_DATA_CORRUPTED),
6950 : : errmsg_internal("multixact %u has two or more updating members",
6951 : : multi),
6952 : : errdetail_internal("First updater XID=%u second updater XID=%u.",
6953 : : update_xid, xid)));
6954 : :
6955 : : /*
6956 : : * As with all tuple visibility routines, it's critical to test
6957 : : * TransactionIdIsInProgress before TransactionIdDidCommit, because of
6958 : : * race conditions explained in detail in heapam_visibility.c.
6959 : : */
6960 [ # # # # ]: 0 : if (TransactionIdIsCurrentTransactionId(xid) ||
6961 : 0 : TransactionIdIsInProgress(xid))
6962 : 0 : update_xid = xid;
6963 [ # # ]: 0 : else if (TransactionIdDidCommit(xid))
6964 : : {
6965 : : /*
6966 : : * The transaction committed, so we can tell caller to set
6967 : : * HEAP_XMAX_COMMITTED. (We can only do this because we know the
6968 : : * transaction is not running.)
6969 : : */
6970 : 0 : update_committed = true;
6971 : 0 : update_xid = xid;
6972 : : }
6973 : : else
6974 : : {
6975 : : /*
6976 : : * Not in progress, not committed -- must be aborted or crashed;
6977 : : * we can ignore it.
6978 : : */
6979 : 0 : continue;
6980 : : }
6981 : :
6982 : : /*
6983 : : * We determined that updater must be kept -- add it to pending new
6984 : : * members list
6985 : : */
1085 6986 [ # # ]: 0 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6987 [ # # ]: 0 : ereport(ERROR,
6988 : : (errcode(ERRCODE_DATA_CORRUPTED),
6989 : : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6990 : : multi, xid, cutoffs->OldestXmin)));
1091 6991 : 0 : newmembers[nnewmembers++] = members[i];
6992 : : }
6993 : :
4384 alvherre@alvh.no-ip. 6994 :CBC 1 : pfree(members);
6995 : :
6996 : : /*
6997 : : * Determine what to do with caller's multi based on information gathered
6998 : : * during our second pass
6999 : : */
7000 [ - + ]: 1 : if (nnewmembers == 0)
7001 : : {
7002 : : /* Nothing worth keeping */
4384 alvherre@alvh.no-ip. 7003 :UBC 0 : *flags |= FRM_INVALIDATE_XMAX;
1091 pg@bowt.ie 7004 : 0 : newxmax = InvalidTransactionId;
7005 : : }
4384 alvherre@alvh.no-ip. 7006 [ - + - - ]:CBC 1 : else if (TransactionIdIsValid(update_xid) && !has_lockers)
7007 : : {
7008 : : /*
7009 : : * If there's a single member and it's an update, pass it back alone
7010 : : * without creating a new Multi. (XXX we could do this when there's a
7011 : : * single remaining locker, too, but that would complicate the API too
7012 : : * much; moreover, the case with the single updater is more
7013 : : * interesting, because those are longer-lived.)
7014 : : */
4384 alvherre@alvh.no-ip. 7015 [ # # ]:UBC 0 : Assert(nnewmembers == 1);
7016 : 0 : *flags |= FRM_RETURN_IS_XID;
7017 [ # # ]: 0 : if (update_committed)
7018 : 0 : *flags |= FRM_MARK_COMMITTED;
1091 pg@bowt.ie 7019 : 0 : newxmax = update_xid;
7020 : : }
7021 : : else
7022 : : {
7023 : : /*
7024 : : * Create a new multixact with the surviving members of the previous
7025 : : * one, to set as new Xmax in the tuple
7026 : : */
1091 pg@bowt.ie 7027 :CBC 1 : newxmax = MultiXactIdCreateFromMembers(nnewmembers, newmembers);
4384 alvherre@alvh.no-ip. 7028 : 1 : *flags |= FRM_RETURN_IS_MULTI;
7029 : : }
7030 : :
7031 : 1 : pfree(newmembers);
7032 : :
1085 pg@bowt.ie 7033 : 1 : pagefrz->freeze_required = true;
1091 7034 : 1 : return newxmax;
7035 : : }
7036 : :
7037 : : /*
7038 : : * heap_prepare_freeze_tuple
7039 : : *
7040 : : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7041 : : * are older than the OldestXmin and/or OldestMxact freeze cutoffs. If so,
7042 : : * setup enough state (in the *frz output argument) to enable caller to
7043 : : * process this tuple as part of freezing its page, and return true. Return
7044 : : * false if nothing can be changed about the tuple right now.
7045 : : *
7046 : : * Also sets *totally_frozen to true if the tuple will be totally frozen once
7047 : : * caller executes returned freeze plan (or if the tuple was already totally
7048 : : * frozen by an earlier VACUUM). This indicates that there are no remaining
7049 : : * XIDs or MultiXactIds that will need to be processed by a future VACUUM.
7050 : : *
7051 : : * VACUUM caller must assemble HeapTupleFreeze freeze plan entries for every
7052 : : * tuple that we returned true for, and then execute freezing. Caller must
7053 : : * initialize pagefrz fields for page as a whole before first call here for
7054 : : * each heap page.
7055 : : *
7056 : : * VACUUM caller decides on whether or not to freeze the page as a whole.
7057 : : * We'll often prepare freeze plans for a page that caller just discards.
7058 : : * However, VACUUM doesn't always get to make a choice; it must freeze when
7059 : : * pagefrz.freeze_required is set, to ensure that any XIDs < FreezeLimit (and
7060 : : * MXIDs < MultiXactCutoff) can never be left behind. We help to make sure
7061 : : * that VACUUM always follows that rule.
7062 : : *
7063 : : * We sometimes force freezing of xmax MultiXactId values long before it is
7064 : : * strictly necessary to do so just to ensure the FreezeLimit postcondition.
7065 : : * It's worth processing MultiXactIds proactively when it is cheap to do so,
7066 : : * and it's convenient to make that happen by piggy-backing it on the "force
7067 : : * freezing" mechanism. Conversely, we sometimes delay freezing MultiXactIds
7068 : : * because it is expensive right now (though only when it's still possible to
7069 : : * do so without violating the FreezeLimit/MultiXactCutoff postcondition).
7070 : : *
7071 : : * It is assumed that the caller has checked the tuple with
7072 : : * HeapTupleSatisfiesVacuum() and determined that it is not HEAPTUPLE_DEAD
7073 : : * (else we should be removing the tuple, not freezing it).
7074 : : *
7075 : : * NB: This function has side effects: it might allocate a new MultiXactId.
7076 : : * It will be set as tuple's new xmax when our *frz output is processed within
7077 : : * heap_execute_freeze_tuple later on. If the tuple is in a shared buffer
7078 : : * then caller had better have an exclusive lock on it already.
7079 : : */
7080 : : bool
2956 andres@anarazel.de 7081 : 4665109 : heap_prepare_freeze_tuple(HeapTupleHeader tuple,
7082 : : const struct VacuumCutoffs *cutoffs,
7083 : : HeapPageFreeze *pagefrz,
7084 : : HeapTupleFreeze *frz, bool *totally_frozen)
7085 : : {
1091 pg@bowt.ie 7086 : 4665109 : bool xmin_already_frozen = false,
7087 : 4665109 : xmax_already_frozen = false;
7088 : 4665109 : bool freeze_xmin = false,
7089 : 4665109 : replace_xvac = false,
7090 : 4665109 : replace_xmax = false,
7091 : 4665109 : freeze_xmax = false;
7092 : : TransactionId xid;
7093 : :
1079 7094 : 4665109 : frz->xmax = HeapTupleHeaderGetRawXmax(tuple);
4384 alvherre@alvh.no-ip. 7095 : 4665109 : frz->t_infomask2 = tuple->t_infomask2;
7096 : 4665109 : frz->t_infomask = tuple->t_infomask;
1079 pg@bowt.ie 7097 : 4665109 : frz->frzflags = 0;
7098 : 4665109 : frz->checkflags = 0;
7099 : :
7100 : : /*
7101 : : * Process xmin, while keeping track of whether it's already frozen, or
7102 : : * will become frozen iff our freeze plan is executed by caller (could be
7103 : : * neither).
7104 : : */
6982 tgl@sss.pgh.pa.us 7105 : 4665109 : xid = HeapTupleHeaderGetXmin(tuple);
2421 alvherre@alvh.no-ip. 7106 [ + + ]: 4665109 : if (!TransactionIdIsNormal(xid))
1091 pg@bowt.ie 7107 : 1429161 : xmin_already_frozen = true;
7108 : : else
7109 : : {
7110 [ - + ]: 3235948 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
2956 andres@anarazel.de 7111 [ # # ]:UBC 0 : ereport(ERROR,
7112 : : (errcode(ERRCODE_DATA_CORRUPTED),
7113 : : errmsg_internal("found xmin %u from before relfrozenxid %u",
7114 : : xid, cutoffs->relfrozenxid)));
7115 : :
7116 : : /* Will set freeze_xmin flags in freeze plan below */
1085 pg@bowt.ie 7117 :CBC 3235948 : freeze_xmin = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7118 : :
7119 : : /* Verify that xmin committed if and when freeze plan is executed */
1079 7120 [ + + ]: 3235948 : if (freeze_xmin)
7121 : 2713781 : frz->checkflags |= HEAP_FREEZE_CHECK_XMIN_COMMITTED;
7122 : : }
7123 : :
7124 : : /*
7125 : : * Old-style VACUUM FULL is gone, but we have to process xvac for as long
7126 : : * as we support having MOVED_OFF/MOVED_IN tuples in the database
7127 : : */
1091 7128 : 4665109 : xid = HeapTupleHeaderGetXvac(tuple);
7129 [ - + ]: 4665109 : if (TransactionIdIsNormal(xid))
7130 : : {
1091 pg@bowt.ie 7131 [ # # ]:UBC 0 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7132 [ # # ]: 0 : Assert(TransactionIdPrecedes(xid, cutoffs->OldestXmin));
7133 : :
7134 : : /*
7135 : : * For Xvac, we always freeze proactively. This allows totally_frozen
7136 : : * tracking to ignore xvac.
7137 : : */
1085 7138 : 0 : replace_xvac = pagefrz->freeze_required = true;
7139 : :
7140 : : /* Will set replace_xvac flags in freeze plan below */
7141 : : }
7142 : :
7143 : : /* Now process xmax */
1079 pg@bowt.ie 7144 :CBC 4665109 : xid = frz->xmax;
4402 alvherre@alvh.no-ip. 7145 [ + + ]: 4665109 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7146 : : {
7147 : : /* Raw xmax is a MultiXactId */
7148 : : TransactionId newxmax;
7149 : : uint16 flags;
7150 : :
7151 : : /*
7152 : : * We will either remove xmax completely (in the "freeze_xmax" path),
7153 : : * process xmax by replacing it (in the "replace_xmax" path), or
7154 : : * perform no-op xmax processing. The only constraint is that the
7155 : : * FreezeLimit/MultiXactCutoff postcondition must never be violated.
7156 : : */
1091 pg@bowt.ie 7157 : 6 : newxmax = FreezeMultiXactId(xid, tuple->t_infomask, cutoffs,
7158 : : &flags, pagefrz);
7159 : :
1085 7160 [ + + ]: 6 : if (flags & FRM_NOOP)
7161 : : {
7162 : : /*
7163 : : * xmax is a MultiXactId, and nothing about it changes for now.
7164 : : * This is the only case where 'freeze_required' won't have been
7165 : : * set for us by FreezeMultiXactId, as well as the only case where
7166 : : * neither freeze_xmax nor replace_xmax are set (given a multi).
7167 : : *
7168 : : * This is a no-op, but the call to FreezeMultiXactId might have
7169 : : * ratcheted back NewRelfrozenXid and/or NewRelminMxid trackers
7170 : : * for us (the "freeze page" variants, specifically). That'll
7171 : : * make it safe for our caller to freeze the page later on, while
7172 : : * leaving this particular xmax undisturbed.
7173 : : *
7174 : : * FreezeMultiXactId is _not_ responsible for the "no freeze"
7175 : : * NewRelfrozenXid/NewRelminMxid trackers, though -- that's our
7176 : : * job. A call to heap_tuple_should_freeze for this same tuple
7177 : : * will take place below if 'freeze_required' isn't set already.
7178 : : * (This repeats work from FreezeMultiXactId, but allows "no
7179 : : * freeze" tracker maintenance to happen in only one place.)
7180 : : */
7181 [ - + ]: 1 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->MultiXactCutoff));
7182 [ + - - + ]: 1 : Assert(MultiXactIdIsValid(newxmax) && xid == newxmax);
7183 : : }
7184 [ - + ]: 5 : else if (flags & FRM_RETURN_IS_XID)
7185 : : {
7186 : : /*
7187 : : * xmax will become an updater Xid (original MultiXact's updater
7188 : : * member Xid will be carried forward as a simple Xid in Xmax).
7189 : : */
1091 pg@bowt.ie 7190 [ # # ]:UBC 0 : Assert(!TransactionIdPrecedes(newxmax, cutoffs->OldestXmin));
7191 : :
7192 : : /*
7193 : : * NB -- some of these transformations are only valid because we
7194 : : * know the return Xid is a tuple updater (i.e. not merely a
7195 : : * locker.) Also note that the only reason we don't explicitly
7196 : : * worry about HEAP_KEYS_UPDATED is because it lives in
7197 : : * t_infomask2 rather than t_infomask.
7198 : : */
4384 alvherre@alvh.no-ip. 7199 : 0 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7200 : 0 : frz->xmax = newxmax;
7201 [ # # ]: 0 : if (flags & FRM_MARK_COMMITTED)
3086 teodor@sigaev.ru 7202 : 0 : frz->t_infomask |= HEAP_XMAX_COMMITTED;
1091 pg@bowt.ie 7203 : 0 : replace_xmax = true;
7204 : : }
4384 alvherre@alvh.no-ip. 7205 [ + + ]:CBC 5 : else if (flags & FRM_RETURN_IS_MULTI)
7206 : : {
7207 : : uint16 newbits;
7208 : : uint16 newbits2;
7209 : :
7210 : : /*
7211 : : * xmax is an old MultiXactId that we have to replace with a new
7212 : : * MultiXactId, to carry forward two or more original member XIDs.
7213 : : */
1091 pg@bowt.ie 7214 [ - + ]: 1 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->OldestMxact));
7215 : :
7216 : : /*
7217 : : * We can't use GetMultiXactIdHintBits directly on the new multi
7218 : : * here; that routine initializes the masks to all zeroes, which
7219 : : * would lose other bits we need. Doing it this way ensures all
7220 : : * unrelated bits remain untouched.
7221 : : */
4384 alvherre@alvh.no-ip. 7222 : 1 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7223 : 1 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7224 : 1 : GetMultiXactIdHintBits(newxmax, &newbits, &newbits2);
7225 : 1 : frz->t_infomask |= newbits;
7226 : 1 : frz->t_infomask2 |= newbits2;
7227 : 1 : frz->xmax = newxmax;
1091 pg@bowt.ie 7228 : 1 : replace_xmax = true;
7229 : : }
7230 : : else
7231 : : {
7232 : : /*
7233 : : * Freeze plan for tuple "freezes xmax" in the strictest sense:
7234 : : * it'll leave nothing in xmax (neither an Xid nor a MultiXactId).
7235 : : */
7236 [ - + ]: 4 : Assert(flags & FRM_INVALIDATE_XMAX);
1354 7237 [ - + ]: 4 : Assert(!TransactionIdIsValid(newxmax));
7238 : :
7239 : : /* Will set freeze_xmax flags in freeze plan below */
1091 7240 : 4 : freeze_xmax = true;
7241 : : }
7242 : :
7243 : : /* MultiXactId processing forces freezing (barring FRM_NOOP case) */
1085 7244 [ - + - - : 6 : Assert(pagefrz->freeze_required || (!freeze_xmax && !replace_xmax));
- - ]
7245 : : }
3472 rhaas@postgresql.org 7246 [ + + ]: 4665103 : else if (TransactionIdIsNormal(xid))
7247 : : {
7248 : : /* Raw xmax is normal XID */
1091 pg@bowt.ie 7249 [ - + ]: 277957 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
2956 andres@anarazel.de 7250 [ # # ]:UBC 0 : ereport(ERROR,
7251 : : (errcode(ERRCODE_DATA_CORRUPTED),
7252 : : errmsg_internal("found xmax %u from before relfrozenxid %u",
7253 : : xid, cutoffs->relfrozenxid)));
7254 : :
7255 : : /* Will set freeze_xmax flags in freeze plan below */
1079 pg@bowt.ie 7256 :CBC 277957 : freeze_xmax = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
7257 : :
7258 : : /*
7259 : : * Verify that xmax aborted if and when freeze plan is executed,
7260 : : * provided it's from an update. (A lock-only xmax can be removed
7261 : : * independent of this, since the lock is released at xact end.)
7262 : : */
7263 [ + + + + ]: 277957 : if (freeze_xmax && !HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask))
7264 : 1953 : frz->checkflags |= HEAP_FREEZE_CHECK_XMAX_ABORTED;
7265 : : }
1120 7266 [ + - ]: 4387146 : else if (!TransactionIdIsValid(xid))
7267 : : {
7268 : : /* Raw xmax is InvalidTransactionId XID */
7269 [ - + ]: 4387146 : Assert((tuple->t_infomask & HEAP_XMAX_IS_MULTI) == 0);
2784 alvherre@alvh.no-ip. 7270 : 4387146 : xmax_already_frozen = true;
7271 : : }
7272 : : else
2784 alvherre@alvh.no-ip. 7273 [ # # ]:UBC 0 : ereport(ERROR,
7274 : : (errcode(ERRCODE_DATA_CORRUPTED),
7275 : : errmsg_internal("found raw xmax %u (infomask 0x%04x) not invalid and not multi",
7276 : : xid, tuple->t_infomask)));
7277 : :
1091 pg@bowt.ie 7278 [ + + ]:CBC 4665109 : if (freeze_xmin)
7279 : : {
7280 [ - + ]: 2713781 : Assert(!xmin_already_frozen);
7281 : :
7282 : 2713781 : frz->t_infomask |= HEAP_XMIN_FROZEN;
7283 : : }
7284 [ - + ]: 4665109 : if (replace_xvac)
7285 : : {
7286 : : /*
7287 : : * If a MOVED_OFF tuple is not dead, the xvac transaction must have
7288 : : * failed; whereas a non-dead MOVED_IN tuple must mean the xvac
7289 : : * transaction succeeded.
7290 : : */
1085 pg@bowt.ie 7291 [ # # ]:UBC 0 : Assert(pagefrz->freeze_required);
1091 7292 [ # # ]: 0 : if (tuple->t_infomask & HEAP_MOVED_OFF)
7293 : 0 : frz->frzflags |= XLH_INVALID_XVAC;
7294 : : else
7295 : 0 : frz->frzflags |= XLH_FREEZE_XVAC;
7296 : : }
1091 pg@bowt.ie 7297 [ + + ]:CBC 4665109 : if (replace_xmax)
7298 : : {
7299 [ + - - + ]: 1 : Assert(!xmax_already_frozen && !freeze_xmax);
1085 7300 [ - + ]: 1 : Assert(pagefrz->freeze_required);
7301 : :
7302 : : /* Already set replace_xmax flags in freeze plan earlier */
7303 : : }
4402 alvherre@alvh.no-ip. 7304 [ + + ]: 4665109 : if (freeze_xmax)
7305 : : {
1091 pg@bowt.ie 7306 [ + - - + ]: 2925 : Assert(!xmax_already_frozen && !replace_xmax);
7307 : :
4384 alvherre@alvh.no-ip. 7308 : 2925 : frz->xmax = InvalidTransactionId;
7309 : :
7310 : : /*
7311 : : * The tuple might be marked either XMAX_INVALID or XMAX_COMMITTED +
7312 : : * LOCKED. Normalize to INVALID just to be sure no one gets confused.
7313 : : * Also get rid of the HEAP_KEYS_UPDATED bit.
7314 : : */
7315 : 2925 : frz->t_infomask &= ~HEAP_XMAX_BITS;
7316 : 2925 : frz->t_infomask |= HEAP_XMAX_INVALID;
7317 : 2925 : frz->t_infomask2 &= ~HEAP_HOT_UPDATED;
7318 : 2925 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
7319 : : }
7320 : :
7321 : : /*
7322 : : * Determine if this tuple is already totally frozen, or will become
7323 : : * totally frozen (provided caller executes freeze plans for the page)
7324 : : */
1091 pg@bowt.ie 7325 [ + + + + : 8805126 : *totally_frozen = ((freeze_xmin || xmin_already_frozen) &&
+ + ]
7326 [ + + ]: 4140017 : (freeze_xmax || xmax_already_frozen));
7327 : :
1085 7328 [ + + + + : 4665109 : if (!pagefrz->freeze_required && !(xmin_already_frozen &&
+ + ]
7329 : : xmax_already_frozen))
7330 : : {
7331 : : /*
7332 : : * So far no previous tuple from the page made freezing mandatory.
7333 : : * Does this tuple force caller to freeze the entire page?
7334 : : */
7335 : 2143870 : pagefrz->freeze_required =
7336 : 2143870 : heap_tuple_should_freeze(tuple, cutoffs,
7337 : : &pagefrz->NoFreezePageRelfrozenXid,
7338 : : &pagefrz->NoFreezePageRelminMxid);
7339 : : }
7340 : :
7341 : : /* Tell caller if this tuple has a usable freeze plan set in *frz */
1091 7342 [ + + + - : 4665109 : return freeze_xmin || replace_xvac || replace_xmax || freeze_xmax;
+ - + + ]
7343 : : }
7344 : :
7345 : : /*
7346 : : * Perform xmin/xmax XID status sanity checks before actually executing freeze
7347 : : * plans.
7348 : : *
7349 : : * heap_prepare_freeze_tuple doesn't perform these checks directly because
7350 : : * pg_xact lookups are relatively expensive. They shouldn't be repeated by
7351 : : * successive VACUUMs that each decide against freezing the same page.
7352 : : */
7353 : : void
623 heikki.linnakangas@i 7354 : 21586 : heap_pre_freeze_checks(Buffer buffer,
7355 : : HeapTupleFreeze *tuples, int ntuples)
7356 : : {
1128 pg@bowt.ie 7357 : 21586 : Page page = BufferGetPage(buffer);
7358 : :
1079 7359 [ + + ]: 977025 : for (int i = 0; i < ntuples; i++)
7360 : : {
7361 : 955439 : HeapTupleFreeze *frz = tuples + i;
7362 : 955439 : ItemId itemid = PageGetItemId(page, frz->offset);
7363 : : HeapTupleHeader htup;
7364 : :
7365 : 955439 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
7366 : :
7367 : : /* Deliberately avoid relying on tuple hint bits here */
7368 [ + + ]: 955439 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMIN_COMMITTED)
7369 : : {
7370 : 955438 : TransactionId xmin = HeapTupleHeaderGetRawXmin(htup);
7371 : :
7372 [ - + ]: 955438 : Assert(!HeapTupleHeaderXminFrozen(htup));
7373 [ - + ]: 955438 : if (unlikely(!TransactionIdDidCommit(xmin)))
1079 pg@bowt.ie 7374 [ # # ]:UBC 0 : ereport(ERROR,
7375 : : (errcode(ERRCODE_DATA_CORRUPTED),
7376 : : errmsg_internal("uncommitted xmin %u needs to be frozen",
7377 : : xmin)));
7378 : : }
7379 : :
7380 : : /*
7381 : : * TransactionIdDidAbort won't work reliably in the presence of XIDs
7382 : : * left behind by transactions that were in progress during a crash,
7383 : : * so we can only check that xmax didn't commit
7384 : : */
1079 pg@bowt.ie 7385 [ + + ]:CBC 955439 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMAX_ABORTED)
7386 : : {
7387 : 589 : TransactionId xmax = HeapTupleHeaderGetRawXmax(htup);
7388 : :
7389 [ - + ]: 589 : Assert(TransactionIdIsNormal(xmax));
7390 [ - + ]: 589 : if (unlikely(TransactionIdDidCommit(xmax)))
1079 pg@bowt.ie 7391 [ # # ]:UBC 0 : ereport(ERROR,
7392 : : (errcode(ERRCODE_DATA_CORRUPTED),
7393 : : errmsg_internal("cannot freeze committed xmax %u",
7394 : : xmax)));
7395 : : }
7396 : : }
623 heikki.linnakangas@i 7397 :CBC 21586 : }
7398 : :
7399 : : /*
7400 : : * Helper which executes freezing of one or more heap tuples on a page on
7401 : : * behalf of caller. Caller passes an array of tuple plans from
7402 : : * heap_prepare_freeze_tuple. Caller must set 'offset' in each plan for us.
7403 : : * Must be called in a critical section that also marks the buffer dirty and,
7404 : : * if needed, emits WAL.
7405 : : */
7406 : : void
7407 : 21586 : heap_freeze_prepared_tuples(Buffer buffer, HeapTupleFreeze *tuples, int ntuples)
7408 : : {
7409 : 21586 : Page page = BufferGetPage(buffer);
7410 : :
1128 pg@bowt.ie 7411 [ + + ]: 977025 : for (int i = 0; i < ntuples; i++)
7412 : : {
1079 7413 : 955439 : HeapTupleFreeze *frz = tuples + i;
7414 : 955439 : ItemId itemid = PageGetItemId(page, frz->offset);
7415 : : HeapTupleHeader htup;
7416 : :
1128 7417 : 955439 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
1079 7418 : 955439 : heap_execute_freeze_tuple(htup, frz);
7419 : : }
1128 7420 : 21586 : }
7421 : :
7422 : : /*
7423 : : * heap_freeze_tuple
7424 : : * Freeze tuple in place, without WAL logging.
7425 : : *
7426 : : * Useful for callers like CLUSTER that perform their own WAL logging.
7427 : : */
7428 : : bool
2956 andres@anarazel.de 7429 : 358489 : heap_freeze_tuple(HeapTupleHeader tuple,
7430 : : TransactionId relfrozenxid, TransactionId relminmxid,
7431 : : TransactionId FreezeLimit, TransactionId MultiXactCutoff)
7432 : : {
7433 : : HeapTupleFreeze frz;
7434 : : bool do_freeze;
7435 : : bool totally_frozen;
7436 : : struct VacuumCutoffs cutoffs;
7437 : : HeapPageFreeze pagefrz;
7438 : :
1091 pg@bowt.ie 7439 : 358489 : cutoffs.relfrozenxid = relfrozenxid;
7440 : 358489 : cutoffs.relminmxid = relminmxid;
7441 : 358489 : cutoffs.OldestXmin = FreezeLimit;
7442 : 358489 : cutoffs.OldestMxact = MultiXactCutoff;
7443 : 358489 : cutoffs.FreezeLimit = FreezeLimit;
7444 : 358489 : cutoffs.MultiXactCutoff = MultiXactCutoff;
7445 : :
1085 7446 : 358489 : pagefrz.freeze_required = true;
7447 : 358489 : pagefrz.FreezePageRelfrozenXid = FreezeLimit;
7448 : 358489 : pagefrz.FreezePageRelminMxid = MultiXactCutoff;
7449 : 358489 : pagefrz.NoFreezePageRelfrozenXid = FreezeLimit;
7450 : 358489 : pagefrz.NoFreezePageRelminMxid = MultiXactCutoff;
7451 : :
1091 7452 : 358489 : do_freeze = heap_prepare_freeze_tuple(tuple, &cutoffs,
7453 : : &pagefrz, &frz, &totally_frozen);
7454 : :
7455 : : /*
7456 : : * Note that because this is not a WAL-logged operation, we don't need to
7457 : : * fill in the offset in the freeze record.
7458 : : */
7459 : :
4384 alvherre@alvh.no-ip. 7460 [ + + ]: 358489 : if (do_freeze)
7461 : 276955 : heap_execute_freeze_tuple(tuple, &frz);
7462 : 358489 : return do_freeze;
7463 : : }
7464 : :
7465 : : /*
7466 : : * For a given MultiXactId, return the hint bits that should be set in the
7467 : : * tuple's infomask.
7468 : : *
7469 : : * Normally this should be called for a multixact that was just created, and
7470 : : * so is on our local cache, so the GetMembers call is fast.
7471 : : */
7472 : : static void
4711 7473 : 76803 : GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
7474 : : uint16 *new_infomask2)
7475 : : {
7476 : : int nmembers;
7477 : : MultiXactMember *members;
7478 : : int i;
4585 bruce@momjian.us 7479 : 76803 : uint16 bits = HEAP_XMAX_IS_MULTI;
7480 : 76803 : uint16 bits2 = 0;
7481 : 76803 : bool has_update = false;
7482 : 76803 : LockTupleMode strongest = LockTupleKeyShare;
7483 : :
7484 : : /*
7485 : : * We only use this in multis we just created, so they cannot be values
7486 : : * pre-pg_upgrade.
7487 : : */
4159 alvherre@alvh.no-ip. 7488 : 76803 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
7489 : :
4711 7490 [ + + ]: 1472621 : for (i = 0; i < nmembers; i++)
7491 : : {
7492 : : LockTupleMode mode;
7493 : :
7494 : : /*
7495 : : * Remember the strongest lock mode held by any member of the
7496 : : * multixact.
7497 : : */
4703 7498 : 1395818 : mode = TUPLOCK_from_mxstatus(members[i].status);
7499 [ + + ]: 1395818 : if (mode > strongest)
7500 : 2888 : strongest = mode;
7501 : :
7502 : : /* See what other bits we need */
4711 7503 [ + + + + : 1395818 : switch (members[i].status)
- ]
7504 : : {
7505 : 1393400 : case MultiXactStatusForKeyShare:
7506 : : case MultiXactStatusForShare:
7507 : : case MultiXactStatusForNoKeyUpdate:
7508 : 1393400 : break;
7509 : :
7510 : 52 : case MultiXactStatusForUpdate:
7511 : 52 : bits2 |= HEAP_KEYS_UPDATED;
7512 : 52 : break;
7513 : :
7514 : 2356 : case MultiXactStatusNoKeyUpdate:
7515 : 2356 : has_update = true;
7516 : 2356 : break;
7517 : :
7518 : 10 : case MultiXactStatusUpdate:
7519 : 10 : bits2 |= HEAP_KEYS_UPDATED;
7520 : 10 : has_update = true;
7521 : 10 : break;
7522 : : }
7523 : : }
7524 : :
4703 7525 [ + + + + ]: 76803 : if (strongest == LockTupleExclusive ||
7526 : : strongest == LockTupleNoKeyExclusive)
7527 : 2446 : bits |= HEAP_XMAX_EXCL_LOCK;
7528 [ + + ]: 74357 : else if (strongest == LockTupleShare)
7529 : 439 : bits |= HEAP_XMAX_SHR_LOCK;
7530 [ + - ]: 73918 : else if (strongest == LockTupleKeyShare)
7531 : 73918 : bits |= HEAP_XMAX_KEYSHR_LOCK;
7532 : :
4711 7533 [ + + ]: 76803 : if (!has_update)
7534 : 74437 : bits |= HEAP_XMAX_LOCK_ONLY;
7535 : :
7536 [ + - ]: 76803 : if (nmembers > 0)
7537 : 76803 : pfree(members);
7538 : :
7539 : 76803 : *new_infomask = bits;
7540 : 76803 : *new_infomask2 = bits2;
7541 : 76803 : }
7542 : :
7543 : : /*
7544 : : * MultiXactIdGetUpdateXid
7545 : : *
7546 : : * Given a multixact Xmax and corresponding infomask, which does not have the
7547 : : * HEAP_XMAX_LOCK_ONLY bit set, obtain and return the Xid of the updating
7548 : : * transaction.
7549 : : *
7550 : : * Caller is expected to check the status of the updating transaction, if
7551 : : * necessary.
7552 : : */
7553 : : static TransactionId
7554 : 159929 : MultiXactIdGetUpdateXid(TransactionId xmax, uint16 t_infomask)
7555 : : {
4585 bruce@momjian.us 7556 : 159929 : TransactionId update_xact = InvalidTransactionId;
7557 : : MultiXactMember *members;
7558 : : int nmembers;
7559 : :
4711 alvherre@alvh.no-ip. 7560 [ - + ]: 159929 : Assert(!(t_infomask & HEAP_XMAX_LOCK_ONLY));
7561 [ - + ]: 159929 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
7562 : :
7563 : : /*
7564 : : * Since we know the LOCK_ONLY bit is not set, this cannot be a multi from
7565 : : * pre-pg_upgrade.
7566 : : */
4159 7567 : 159929 : nmembers = GetMultiXactIdMembers(xmax, &members, false, false);
7568 : :
4711 7569 [ + - ]: 159929 : if (nmembers > 0)
7570 : : {
7571 : : int i;
7572 : :
7573 [ + + ]: 3191942 : for (i = 0; i < nmembers; i++)
7574 : : {
7575 : : /* Ignore lockers */
4401 7576 [ + + ]: 3032013 : if (!ISUPDATE_from_mxstatus(members[i].status))
4711 7577 : 2872084 : continue;
7578 : :
7579 : : /* there can be at most one updater */
7580 [ - + ]: 159929 : Assert(update_xact == InvalidTransactionId);
7581 : 159929 : update_xact = members[i].xid;
7582 : : #ifndef USE_ASSERT_CHECKING
7583 : :
7584 : : /*
7585 : : * in an assert-enabled build, walk the whole array to ensure
7586 : : * there's no other updater.
7587 : : */
7588 : : break;
7589 : : #endif
7590 : : }
7591 : :
7592 : 159929 : pfree(members);
7593 : : }
7594 : :
7595 : 159929 : return update_xact;
7596 : : }
7597 : :
7598 : : /*
7599 : : * HeapTupleGetUpdateXid
7600 : : * As above, but use a HeapTupleHeader
7601 : : *
7602 : : * See also HeapTupleHeaderGetUpdateXid, which can be used without previously
7603 : : * checking the hint bits.
7604 : : */
7605 : : TransactionId
328 peter@eisentraut.org 7606 : 159921 : HeapTupleGetUpdateXid(const HeapTupleHeaderData *tup)
7607 : : {
7608 : 159921 : return MultiXactIdGetUpdateXid(HeapTupleHeaderGetRawXmax(tup),
7609 : 159921 : tup->t_infomask);
7610 : : }
7611 : :
7612 : : /*
7613 : : * Does the given multixact conflict with the current transaction grabbing a
7614 : : * tuple lock of the given strength?
7615 : : *
7616 : : * The passed infomask pairs up with the given multixact in the tuple header.
7617 : : *
7618 : : * If current_is_member is not NULL, it is set to 'true' if the current
7619 : : * transaction is a member of the given multixact.
7620 : : */
7621 : : static bool
4009 alvherre@alvh.no-ip. 7622 : 218 : DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
7623 : : LockTupleMode lockmode, bool *current_is_member)
7624 : : {
7625 : : int nmembers;
7626 : : MultiXactMember *members;
3861 bruce@momjian.us 7627 : 218 : bool result = false;
7628 : 218 : LOCKMODE wanted = tupleLockExtraInfo[lockmode].hwlock;
7629 : :
3463 alvherre@alvh.no-ip. 7630 [ - + ]: 218 : if (HEAP_LOCKED_UPGRADED(infomask))
3463 alvherre@alvh.no-ip. 7631 :UBC 0 : return false;
7632 : :
3463 alvherre@alvh.no-ip. 7633 :CBC 218 : nmembers = GetMultiXactIdMembers(multi, &members, false,
4009 7634 : 218 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7635 [ + - ]: 218 : if (nmembers >= 0)
7636 : : {
7637 : : int i;
7638 : :
7639 [ + + ]: 2682 : for (i = 0; i < nmembers; i++)
7640 : : {
7641 : : TransactionId memxid;
7642 : : LOCKMODE memlockmode;
7643 : :
2374 7644 [ + + + + : 2471 : if (result && (current_is_member == NULL || *current_is_member))
+ - ]
7645 : : break;
7646 : :
7647 : 2464 : memlockmode = LOCKMODE_from_mxstatus(members[i].status);
7648 : :
7649 : : /* ignore members from current xact (but track their presence) */
2376 7650 : 2464 : memxid = members[i].xid;
7651 [ + + ]: 2464 : if (TransactionIdIsCurrentTransactionId(memxid))
7652 : : {
2374 7653 [ + + ]: 92 : if (current_is_member != NULL)
7654 : 78 : *current_is_member = true;
7655 : 92 : continue;
7656 : : }
7657 [ + + ]: 2372 : else if (result)
7658 : 8 : continue;
7659 : :
7660 : : /* ignore members that don't conflict with the lock we want */
7661 [ + + ]: 2364 : if (!DoLockModesConflict(memlockmode, wanted))
2376 7662 : 2325 : continue;
7663 : :
4009 7664 [ + + ]: 39 : if (ISUPDATE_from_mxstatus(members[i].status))
7665 : : {
7666 : : /* ignore aborted updaters */
7667 [ + + ]: 17 : if (TransactionIdDidAbort(memxid))
7668 : 1 : continue;
7669 : : }
7670 : : else
7671 : : {
7672 : : /* ignore lockers-only that are no longer in progress */
7673 [ + + ]: 22 : if (!TransactionIdIsInProgress(memxid))
7674 : 7 : continue;
7675 : : }
7676 : :
7677 : : /*
7678 : : * Whatever remains are either live lockers that conflict with our
7679 : : * wanted lock, and updaters that are not aborted. Those conflict
7680 : : * with what we want. Set up to return true, but keep going to
7681 : : * look for the current transaction among the multixact members,
7682 : : * if needed.
7683 : : */
7684 : 31 : result = true;
7685 : : }
7686 : 218 : pfree(members);
7687 : : }
7688 : :
7689 : 218 : return result;
7690 : : }
7691 : :
7692 : : /*
7693 : : * Do_MultiXactIdWait
7694 : : * Actual implementation for the two functions below.
7695 : : *
7696 : : * 'multi', 'status' and 'infomask' indicate what to sleep on (the status is
7697 : : * needed to ensure we only sleep on conflicting members, and the infomask is
7698 : : * used to optimize multixact access in case it's a lock-only multi); 'nowait'
7699 : : * indicates whether to use conditional lock acquisition, to allow callers to
7700 : : * fail if lock is unavailable. 'rel', 'ctid' and 'oper' are used to set up
7701 : : * context information for error messages. 'remaining', if not NULL, receives
7702 : : * the number of members that are still running, including any (non-aborted)
7703 : : * subtransactions of our own transaction. 'logLockFailure' indicates whether
7704 : : * to log details when a lock acquisition fails with 'nowait' enabled.
7705 : : *
7706 : : * We do this by sleeping on each member using XactLockTableWait. Any
7707 : : * members that belong to the current backend are *not* waited for, however;
7708 : : * this would not merely be useless but would lead to Assert failure inside
7709 : : * XactLockTableWait. By the time this returns, it is certain that all
7710 : : * transactions *of other backends* that were members of the MultiXactId
7711 : : * that conflict with the requested status are dead (and no new ones can have
7712 : : * been added, since it is not legal to add members to an existing
7713 : : * MultiXactId).
7714 : : *
7715 : : * But by the time we finish sleeping, someone else may have changed the Xmax
7716 : : * of the containing tuple, so the caller needs to iterate on us somehow.
7717 : : *
7718 : : * Note that in case we return false, the number of remaining members is
7719 : : * not to be trusted.
7720 : : */
7721 : : static bool
4711 7722 : 58 : Do_MultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7723 : : uint16 infomask, bool nowait,
7724 : : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7725 : : int *remaining, bool logLockFailure)
7726 : : {
7727 : 58 : bool result = true;
7728 : : MultiXactMember *members;
7729 : : int nmembers;
7730 : 58 : int remain = 0;
7731 : :
7732 : : /* for pre-pg_upgrade tuples, no need to sleep at all */
3463 7733 [ + - ]: 58 : nmembers = HEAP_LOCKED_UPGRADED(infomask) ? -1 :
7734 : 58 : GetMultiXactIdMembers(multi, &members, false,
7735 : 58 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7736 : :
4711 7737 [ + - ]: 58 : if (nmembers >= 0)
7738 : : {
7739 : : int i;
7740 : :
7741 [ + + ]: 187 : for (i = 0; i < nmembers; i++)
7742 : : {
7743 : 133 : TransactionId memxid = members[i].xid;
7744 : 133 : MultiXactStatus memstatus = members[i].status;
7745 : :
7746 [ + + ]: 133 : if (TransactionIdIsCurrentTransactionId(memxid))
7747 : : {
7748 : 24 : remain++;
7749 : 24 : continue;
7750 : : }
7751 : :
7752 [ + + ]: 109 : if (!DoLockModesConflict(LOCKMODE_from_mxstatus(memstatus),
7753 : 109 : LOCKMODE_from_mxstatus(status)))
7754 : : {
7755 [ + + + - ]: 22 : if (remaining && TransactionIdIsInProgress(memxid))
7756 : 8 : remain++;
7757 : 22 : continue;
7758 : : }
7759 : :
7760 : : /*
7761 : : * This member conflicts with our multi, so we have to sleep (or
7762 : : * return failure, if asked to avoid waiting.)
7763 : : *
7764 : : * Note that we don't set up an error context callback ourselves,
7765 : : * but instead we pass the info down to XactLockTableWait. This
7766 : : * might seem a bit wasteful because the context is set up and
7767 : : * tore down for each member of the multixact, but in reality it
7768 : : * should be barely noticeable, and it avoids duplicate code.
7769 : : */
7770 [ + + ]: 87 : if (nowait)
7771 : : {
278 fujii@postgresql.org 7772 : 4 : result = ConditionalXactLockTableWait(memxid, logLockFailure);
4711 alvherre@alvh.no-ip. 7773 [ + - ]: 4 : if (!result)
7774 : 4 : break;
7775 : : }
7776 : : else
4291 7777 : 83 : XactLockTableWait(memxid, rel, ctid, oper);
7778 : : }
7779 : :
4711 7780 : 58 : pfree(members);
7781 : : }
7782 : :
7783 [ + + ]: 58 : if (remaining)
7784 : 10 : *remaining = remain;
7785 : :
7786 : 58 : return result;
7787 : : }
7788 : :
7789 : : /*
7790 : : * MultiXactIdWait
7791 : : * Sleep on a MultiXactId.
7792 : : *
7793 : : * By the time we finish sleeping, someone else may have changed the Xmax
7794 : : * of the containing tuple, so the caller needs to iterate on us somehow.
7795 : : *
7796 : : * We return (in *remaining, if not NULL) the number of members that are still
7797 : : * running, including any (non-aborted) subtransactions of our own transaction.
7798 : : */
7799 : : static void
4291 7800 : 54 : MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
7801 : : Relation rel, const ItemPointerData *ctid, XLTW_Oper oper,
7802 : : int *remaining)
7803 : : {
7804 : 54 : (void) Do_MultiXactIdWait(multi, status, infomask, false,
7805 : : rel, ctid, oper, remaining, false);
4711 7806 : 54 : }
7807 : :
7808 : : /*
7809 : : * ConditionalMultiXactIdWait
7810 : : * As above, but only lock if we can get the lock without blocking.
7811 : : *
7812 : : * By the time we finish sleeping, someone else may have changed the Xmax
7813 : : * of the containing tuple, so the caller needs to iterate on us somehow.
7814 : : *
7815 : : * If the multixact is now all gone, return true. Returns false if some
7816 : : * transactions might still be running.
7817 : : *
7818 : : * We return (in *remaining, if not NULL) the number of members that are still
7819 : : * running, including any (non-aborted) subtransactions of our own transaction.
7820 : : */
7821 : : static bool
7822 : 4 : ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7823 : : uint16 infomask, Relation rel, int *remaining,
7824 : : bool logLockFailure)
7825 : : {
4291 7826 : 4 : return Do_MultiXactIdWait(multi, status, infomask, true,
7827 : : rel, NULL, XLTW_None, remaining, logLockFailure);
7828 : : }
7829 : :
7830 : : /*
7831 : : * heap_tuple_needs_eventual_freeze
7832 : : *
7833 : : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7834 : : * will eventually require freezing (if tuple isn't removed by pruning first).
7835 : : */
7836 : : bool
3578 rhaas@postgresql.org 7837 : 2309317 : heap_tuple_needs_eventual_freeze(HeapTupleHeader tuple)
7838 : : {
7839 : : TransactionId xid;
7840 : :
7841 : : /*
7842 : : * If xmin is a normal transaction ID, this tuple is definitely not
7843 : : * frozen.
7844 : : */
7845 : 2309317 : xid = HeapTupleHeaderGetXmin(tuple);
7846 [ + + ]: 2309317 : if (TransactionIdIsNormal(xid))
7847 : 15552 : return true;
7848 : :
7849 : : /*
7850 : : * If xmax is a valid xact or multixact, this tuple is also not frozen.
7851 : : */
7852 [ + + ]: 2293765 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7853 : : {
7854 : : MultiXactId multi;
7855 : :
7856 : 2 : multi = HeapTupleHeaderGetRawXmax(tuple);
7857 [ + - ]: 2 : if (MultiXactIdIsValid(multi))
7858 : 2 : return true;
7859 : : }
7860 : : else
7861 : : {
7862 : 2293763 : xid = HeapTupleHeaderGetRawXmax(tuple);
7863 [ + + ]: 2293763 : if (TransactionIdIsNormal(xid))
7864 : 13 : return true;
7865 : : }
7866 : :
7867 [ - + ]: 2293750 : if (tuple->t_infomask & HEAP_MOVED)
7868 : : {
3578 rhaas@postgresql.org 7869 :UBC 0 : xid = HeapTupleHeaderGetXvac(tuple);
7870 [ # # ]: 0 : if (TransactionIdIsNormal(xid))
7871 : 0 : return true;
7872 : : }
7873 : :
3578 rhaas@postgresql.org 7874 :CBC 2293750 : return false;
7875 : : }
7876 : :
7877 : : /*
7878 : : * heap_tuple_should_freeze
7879 : : *
7880 : : * Return value indicates if heap_prepare_freeze_tuple sibling function would
7881 : : * (or should) force freezing of the heap page that contains caller's tuple.
7882 : : * Tuple header XIDs/MXIDs < FreezeLimit/MultiXactCutoff trigger freezing.
7883 : : * This includes (xmin, xmax, xvac) fields, as well as MultiXact member XIDs.
7884 : : *
7885 : : * The *NoFreezePageRelfrozenXid and *NoFreezePageRelminMxid input/output
7886 : : * arguments help VACUUM track the oldest extant XID/MXID remaining in rel.
7887 : : * Our working assumption is that caller won't decide to freeze this tuple.
7888 : : * It's up to caller to only ratchet back its own top-level trackers after the
7889 : : * point that it fully commits to not freezing the tuple/page in question.
7890 : : */
7891 : : bool
1085 pg@bowt.ie 7892 : 2144137 : heap_tuple_should_freeze(HeapTupleHeader tuple,
7893 : : const struct VacuumCutoffs *cutoffs,
7894 : : TransactionId *NoFreezePageRelfrozenXid,
7895 : : MultiXactId *NoFreezePageRelminMxid)
7896 : : {
7897 : : TransactionId xid;
7898 : : MultiXactId multi;
1091 7899 : 2144137 : bool freeze = false;
7900 : :
7901 : : /* First deal with xmin */
5154 rhaas@postgresql.org 7902 : 2144137 : xid = HeapTupleHeaderGetXmin(tuple);
1354 pg@bowt.ie 7903 [ + + ]: 2144137 : if (TransactionIdIsNormal(xid))
7904 : : {
1091 7905 [ - + ]: 2103980 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
1085 7906 [ + + ]: 2103980 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7907 : 21677 : *NoFreezePageRelfrozenXid = xid;
1091 7908 [ + + ]: 2103980 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7909 : 19040 : freeze = true;
7910 : : }
7911 : :
7912 : : /* Now deal with xmax */
1354 7913 : 2144137 : xid = InvalidTransactionId;
7914 : 2144137 : multi = InvalidMultiXactId;
7915 [ + + ]: 2144137 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
4402 alvherre@alvh.no-ip. 7916 : 2 : multi = HeapTupleHeaderGetRawXmax(tuple);
7917 : : else
1354 pg@bowt.ie 7918 : 2144135 : xid = HeapTupleHeaderGetRawXmax(tuple);
7919 : :
7920 [ + + ]: 2144137 : if (TransactionIdIsNormal(xid))
7921 : : {
1091 7922 [ - + ]: 262430 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7923 : : /* xmax is a non-permanent XID */
1085 7924 [ + + ]: 262430 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7925 : 4 : *NoFreezePageRelfrozenXid = xid;
1091 7926 [ + + ]: 262430 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7927 : 3 : freeze = true;
7928 : : }
1354 7929 [ + + ]: 1881707 : else if (!MultiXactIdIsValid(multi))
7930 : : {
7931 : : /* xmax is a permanent XID or invalid MultiXactId/XID */
7932 : : }
7933 [ - + ]: 2 : else if (HEAP_LOCKED_UPGRADED(tuple->t_infomask))
7934 : : {
7935 : : /* xmax is a pg_upgrade'd MultiXact, which can't have updater XID */
1085 pg@bowt.ie 7936 [ # # ]:UBC 0 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7937 : 0 : *NoFreezePageRelminMxid = multi;
7938 : : /* heap_prepare_freeze_tuple always freezes pg_upgrade'd xmax */
1091 7939 : 0 : freeze = true;
7940 : : }
7941 : : else
7942 : : {
7943 : : /* xmax is a MultiXactId that may have an updater XID */
7944 : : MultiXactMember *members;
7945 : : int nmembers;
7946 : :
1091 pg@bowt.ie 7947 [ - + ]:CBC 2 : Assert(MultiXactIdPrecedesOrEquals(cutoffs->relminmxid, multi));
1085 7948 [ + - ]: 2 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7949 : 2 : *NoFreezePageRelminMxid = multi;
1091 7950 [ + - ]: 2 : if (MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff))
7951 : 2 : freeze = true;
7952 : :
7953 : : /* need to check whether any member of the mxact is old */
1354 7954 : 2 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7955 : 2 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask));
7956 : :
7957 [ + + ]: 5 : for (int i = 0; i < nmembers; i++)
7958 : : {
7959 : 3 : xid = members[i].xid;
1091 7960 [ - + ]: 3 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
1085 7961 [ - + ]: 3 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
1085 pg@bowt.ie 7962 :UBC 0 : *NoFreezePageRelfrozenXid = xid;
1091 pg@bowt.ie 7963 [ - + ]:CBC 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
1091 pg@bowt.ie 7964 :UBC 0 : freeze = true;
7965 : : }
1354 pg@bowt.ie 7966 [ + + ]:CBC 2 : if (nmembers > 0)
7967 : 1 : pfree(members);
7968 : : }
7969 : :
5154 rhaas@postgresql.org 7970 [ - + ]: 2144137 : if (tuple->t_infomask & HEAP_MOVED)
7971 : : {
5154 rhaas@postgresql.org 7972 :UBC 0 : xid = HeapTupleHeaderGetXvac(tuple);
1354 pg@bowt.ie 7973 [ # # ]: 0 : if (TransactionIdIsNormal(xid))
7974 : : {
1091 7975 [ # # ]: 0 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
1085 7976 [ # # ]: 0 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7977 : 0 : *NoFreezePageRelfrozenXid = xid;
7978 : : /* heap_prepare_freeze_tuple forces xvac freezing */
1091 7979 : 0 : freeze = true;
7980 : : }
7981 : : }
7982 : :
1091 pg@bowt.ie 7983 :CBC 2144137 : return freeze;
7984 : : }
7985 : :
7986 : : /*
7987 : : * Maintain snapshotConflictHorizon for caller by ratcheting forward its value
7988 : : * using any committed XIDs contained in 'tuple', an obsolescent heap tuple
7989 : : * that caller is in the process of physically removing, e.g. via HOT pruning
7990 : : * or index deletion.
7991 : : *
7992 : : * Caller must initialize its value to InvalidTransactionId, which is
7993 : : * generally interpreted as "definitely no need for a recovery conflict".
7994 : : * Final value must reflect all heap tuples that caller will physically remove
7995 : : * (or remove TID references to) via its ongoing pruning/deletion operation.
7996 : : * ResolveRecoveryConflictWithSnapshot() is passed the final value (taken from
7997 : : * caller's WAL record) by REDO routine when it replays caller's operation.
7998 : : */
7999 : : void
1126 8000 : 1566612 : HeapTupleHeaderAdvanceConflictHorizon(HeapTupleHeader tuple,
8001 : : TransactionId *snapshotConflictHorizon)
8002 : : {
5842 simon@2ndQuadrant.co 8003 : 1566612 : TransactionId xmin = HeapTupleHeaderGetXmin(tuple);
4711 alvherre@alvh.no-ip. 8004 : 1566612 : TransactionId xmax = HeapTupleHeaderGetUpdateXid(tuple);
5842 simon@2ndQuadrant.co 8005 : 1566612 : TransactionId xvac = HeapTupleHeaderGetXvac(tuple);
8006 : :
5791 tgl@sss.pgh.pa.us 8007 [ - + ]: 1566612 : if (tuple->t_infomask & HEAP_MOVED)
8008 : : {
1126 pg@bowt.ie 8009 [ # # ]:UBC 0 : if (TransactionIdPrecedes(*snapshotConflictHorizon, xvac))
8010 : 0 : *snapshotConflictHorizon = xvac;
8011 : : }
8012 : :
8013 : : /*
8014 : : * Ignore tuples inserted by an aborted transaction or if the tuple was
8015 : : * updated/deleted by the inserting transaction.
8016 : : *
8017 : : * Look for a committed hint bit, or if no xmin bit is set, check clog.
8018 : : */
4378 rhaas@postgresql.org 8019 [ + + ]:CBC 1566612 : if (HeapTupleHeaderXminCommitted(tuple) ||
8020 [ + + + - ]: 106711 : (!HeapTupleHeaderXminInvalid(tuple) && TransactionIdDidCommit(xmin)))
8021 : : {
5487 simon@2ndQuadrant.co 8022 [ + + + + ]: 2813310 : if (xmax != xmin &&
1126 pg@bowt.ie 8023 : 1326464 : TransactionIdFollows(xmax, *snapshotConflictHorizon))
8024 : 101146 : *snapshotConflictHorizon = xmax;
8025 : : }
5842 simon@2ndQuadrant.co 8026 : 1566612 : }
8027 : :
8028 : : #ifdef USE_PREFETCH
8029 : : /*
8030 : : * Helper function for heap_index_delete_tuples. Issues prefetch requests for
8031 : : * prefetch_count buffers. The prefetch_state keeps track of all the buffers
8032 : : * we can prefetch, and which have already been prefetched; each call to this
8033 : : * function picks up where the previous call left off.
8034 : : *
8035 : : * Note: we expect the deltids array to be sorted in an order that groups TIDs
8036 : : * by heap block, with all TIDs for each block appearing together in exactly
8037 : : * one group.
8038 : : */
8039 : : static void
1799 pg@bowt.ie 8040 : 20048 : index_delete_prefetch_buffer(Relation rel,
8041 : : IndexDeletePrefetchState *prefetch_state,
8042 : : int prefetch_count)
8043 : : {
2458 andres@anarazel.de 8044 : 20048 : BlockNumber cur_hblkno = prefetch_state->cur_hblkno;
8045 : 20048 : int count = 0;
8046 : : int i;
1799 pg@bowt.ie 8047 : 20048 : int ndeltids = prefetch_state->ndeltids;
8048 : 20048 : TM_IndexDelete *deltids = prefetch_state->deltids;
8049 : :
2458 andres@anarazel.de 8050 : 20048 : for (i = prefetch_state->next_item;
1799 pg@bowt.ie 8051 [ + + + + ]: 689653 : i < ndeltids && count < prefetch_count;
2458 andres@anarazel.de 8052 : 669605 : i++)
8053 : : {
1799 pg@bowt.ie 8054 : 669605 : ItemPointer htid = &deltids[i].tid;
8055 : :
2458 andres@anarazel.de 8056 [ + + + + ]: 1333329 : if (cur_hblkno == InvalidBlockNumber ||
8057 : 663724 : ItemPointerGetBlockNumber(htid) != cur_hblkno)
8058 : : {
8059 : 18299 : cur_hblkno = ItemPointerGetBlockNumber(htid);
8060 : 18299 : PrefetchBuffer(rel, MAIN_FORKNUM, cur_hblkno);
8061 : 18299 : count++;
8062 : : }
8063 : : }
8064 : :
8065 : : /*
8066 : : * Save the prefetch position so that next time we can continue from that
8067 : : * position.
8068 : : */
8069 : 20048 : prefetch_state->next_item = i;
8070 : 20048 : prefetch_state->cur_hblkno = cur_hblkno;
8071 : 20048 : }
8072 : : #endif
8073 : :
8074 : : /*
8075 : : * Helper function for heap_index_delete_tuples. Checks for index corruption
8076 : : * involving an invalid TID in index AM caller's index page.
8077 : : *
8078 : : * This is an ideal place for these checks. The index AM must hold a buffer
8079 : : * lock on the index page containing the TIDs we examine here, so we don't
8080 : : * have to worry about concurrent VACUUMs at all. We can be sure that the
8081 : : * index is corrupt when htid points directly to an LP_UNUSED item or
8082 : : * heap-only tuple, which is not the case during standard index scans.
8083 : : */
8084 : : static inline void
1504 pg@bowt.ie 8085 : 555378 : index_delete_check_htid(TM_IndexDeleteOp *delstate,
8086 : : Page page, OffsetNumber maxoff,
8087 : : const ItemPointerData *htid, TM_IndexStatus *istatus)
8088 : : {
8089 : 555378 : OffsetNumber indexpagehoffnum = ItemPointerGetOffsetNumber(htid);
8090 : : ItemId iid;
8091 : :
8092 [ + - + - : 555378 : Assert(OffsetNumberIsValid(istatus->idxoffnum));
- + ]
8093 : :
8094 [ - + ]: 555378 : if (unlikely(indexpagehoffnum > maxoff))
1504 pg@bowt.ie 8095 [ # # ]:UBC 0 : ereport(ERROR,
8096 : : (errcode(ERRCODE_INDEX_CORRUPTED),
8097 : : 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\"",
8098 : : ItemPointerGetBlockNumber(htid),
8099 : : indexpagehoffnum,
8100 : : istatus->idxoffnum, delstate->iblknum,
8101 : : RelationGetRelationName(delstate->irel))));
8102 : :
1504 pg@bowt.ie 8103 :CBC 555378 : iid = PageGetItemId(page, indexpagehoffnum);
8104 [ - + ]: 555378 : if (unlikely(!ItemIdIsUsed(iid)))
1504 pg@bowt.ie 8105 [ # # ]:UBC 0 : ereport(ERROR,
8106 : : (errcode(ERRCODE_INDEX_CORRUPTED),
8107 : : errmsg_internal("heap tid from index tuple (%u,%u) points to unused heap page item at offset %u of block %u in index \"%s\"",
8108 : : ItemPointerGetBlockNumber(htid),
8109 : : indexpagehoffnum,
8110 : : istatus->idxoffnum, delstate->iblknum,
8111 : : RelationGetRelationName(delstate->irel))));
8112 : :
1504 pg@bowt.ie 8113 [ + + ]:CBC 555378 : if (ItemIdHasStorage(iid))
8114 : : {
8115 : : HeapTupleHeader htup;
8116 : :
8117 [ - + ]: 328532 : Assert(ItemIdIsNormal(iid));
8118 : 328532 : htup = (HeapTupleHeader) PageGetItem(page, iid);
8119 : :
8120 [ - + ]: 328532 : if (unlikely(HeapTupleHeaderIsHeapOnly(htup)))
1504 pg@bowt.ie 8121 [ # # ]:UBC 0 : ereport(ERROR,
8122 : : (errcode(ERRCODE_INDEX_CORRUPTED),
8123 : : errmsg_internal("heap tid from index tuple (%u,%u) points to heap-only tuple at offset %u of block %u in index \"%s\"",
8124 : : ItemPointerGetBlockNumber(htid),
8125 : : indexpagehoffnum,
8126 : : istatus->idxoffnum, delstate->iblknum,
8127 : : RelationGetRelationName(delstate->irel))));
8128 : : }
1504 pg@bowt.ie 8129 :CBC 555378 : }
8130 : :
8131 : : /*
8132 : : * heapam implementation of tableam's index_delete_tuples interface.
8133 : : *
8134 : : * This helper function is called by index AMs during index tuple deletion.
8135 : : * See tableam header comments for an explanation of the interface implemented
8136 : : * here and a general theory of operation. Note that each call here is either
8137 : : * a simple index deletion call, or a bottom-up index deletion call.
8138 : : *
8139 : : * It's possible for this to generate a fair amount of I/O, since we may be
8140 : : * deleting hundreds of tuples from a single index block. To amortize that
8141 : : * cost to some degree, this uses prefetching and combines repeat accesses to
8142 : : * the same heap block.
8143 : : */
8144 : : TransactionId
1799 8145 : 5881 : heap_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
8146 : : {
8147 : : /* Initial assumption is that earlier pruning took care of conflict */
1126 8148 : 5881 : TransactionId snapshotConflictHorizon = InvalidTransactionId;
1813 8149 : 5881 : BlockNumber blkno = InvalidBlockNumber;
2458 andres@anarazel.de 8150 : 5881 : Buffer buf = InvalidBuffer;
1813 pg@bowt.ie 8151 : 5881 : Page page = NULL;
8152 : 5881 : OffsetNumber maxoff = InvalidOffsetNumber;
8153 : : TransactionId priorXmax;
8154 : : #ifdef USE_PREFETCH
8155 : : IndexDeletePrefetchState prefetch_state;
8156 : : int prefetch_distance;
8157 : : #endif
8158 : : SnapshotData SnapshotNonVacuumable;
1799 8159 : 5881 : int finalndeltids = 0,
8160 : 5881 : nblocksaccessed = 0;
8161 : :
8162 : : /* State that's only used in bottom-up index deletion case */
8163 : 5881 : int nblocksfavorable = 0;
8164 : 5881 : int curtargetfreespace = delstate->bottomupfreespace,
8165 : 5881 : lastfreespace = 0,
8166 : 5881 : actualfreespace = 0;
8167 : 5881 : bool bottomup_final_block = false;
8168 : :
8169 : 5881 : InitNonVacuumableSnapshot(SnapshotNonVacuumable, GlobalVisTestFor(rel));
8170 : :
8171 : : /* Sort caller's deltids array by TID for further processing */
8172 : 5881 : index_delete_sort(delstate);
8173 : :
8174 : : /*
8175 : : * Bottom-up case: resort deltids array in an order attuned to where the
8176 : : * greatest number of promising TIDs are to be found, and determine how
8177 : : * many blocks from the start of sorted array should be considered
8178 : : * favorable. This will also shrink the deltids array in order to
8179 : : * eliminate completely unfavorable blocks up front.
8180 : : */
8181 [ + + ]: 5881 : if (delstate->bottomup)
8182 : 1981 : nblocksfavorable = bottomup_sort_and_shrink(delstate);
8183 : :
8184 : : #ifdef USE_PREFETCH
8185 : : /* Initialize prefetch state. */
2458 andres@anarazel.de 8186 : 5881 : prefetch_state.cur_hblkno = InvalidBlockNumber;
8187 : 5881 : prefetch_state.next_item = 0;
1799 pg@bowt.ie 8188 : 5881 : prefetch_state.ndeltids = delstate->ndeltids;
8189 : 5881 : prefetch_state.deltids = delstate->deltids;
8190 : :
8191 : : /*
8192 : : * Determine the prefetch distance that we will attempt to maintain.
8193 : : *
8194 : : * Since the caller holds a buffer lock somewhere in rel, we'd better make
8195 : : * sure that isn't a catalog relation before we call code that does
8196 : : * syscache lookups, to avoid risk of deadlock.
8197 : : */
2451 tmunro@postgresql.or 8198 [ + + ]: 5881 : if (IsCatalogRelation(rel))
2102 8199 : 4284 : prefetch_distance = maintenance_io_concurrency;
8200 : : else
8201 : : prefetch_distance =
8202 : 1597 : get_tablespace_maintenance_io_concurrency(rel->rd_rel->reltablespace);
8203 : :
8204 : : /* Cap initial prefetch distance for bottom-up deletion caller */
1799 pg@bowt.ie 8205 [ + + ]: 5881 : if (delstate->bottomup)
8206 : : {
8207 [ - + ]: 1981 : Assert(nblocksfavorable >= 1);
8208 [ - + ]: 1981 : Assert(nblocksfavorable <= BOTTOMUP_MAX_NBLOCKS);
8209 : 1981 : prefetch_distance = Min(prefetch_distance, nblocksfavorable);
8210 : : }
8211 : :
8212 : : /* Start prefetching. */
8213 : 5881 : index_delete_prefetch_buffer(rel, &prefetch_state, prefetch_distance);
8214 : : #endif
8215 : :
8216 : : /* Iterate over deltids, determine which to delete, check their horizon */
8217 [ - + ]: 5881 : Assert(delstate->ndeltids > 0);
8218 [ + + ]: 561259 : for (int i = 0; i < delstate->ndeltids; i++)
8219 : : {
8220 : 557359 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8221 : 557359 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8222 : 557359 : ItemPointer htid = &ideltid->tid;
8223 : : OffsetNumber offnum;
8224 : :
8225 : : /*
8226 : : * Read buffer, and perform required extra steps each time a new block
8227 : : * is encountered. Avoid refetching if it's the same block as the one
8228 : : * from the last htid.
8229 : : */
1813 8230 [ + + + + ]: 1108837 : if (blkno == InvalidBlockNumber ||
8231 : 551478 : ItemPointerGetBlockNumber(htid) != blkno)
8232 : : {
8233 : : /*
8234 : : * Consider giving up early for bottom-up index deletion caller
8235 : : * first. (Only prefetch next-next block afterwards, when it
8236 : : * becomes clear that we're at least going to access the next
8237 : : * block in line.)
8238 : : *
8239 : : * Sometimes the first block frees so much space for bottom-up
8240 : : * caller that the deletion process can end without accessing any
8241 : : * more blocks. It is usually necessary to access 2 or 3 blocks
8242 : : * per bottom-up deletion operation, though.
8243 : : */
1799 8244 [ + + ]: 16148 : if (delstate->bottomup)
8245 : : {
8246 : : /*
8247 : : * We often allow caller to delete a few additional items
8248 : : * whose entries we reached after the point that space target
8249 : : * from caller was satisfied. The cost of accessing the page
8250 : : * was already paid at that point, so it made sense to finish
8251 : : * it off. When that happened, we finalize everything here
8252 : : * (by finishing off the whole bottom-up deletion operation
8253 : : * without needlessly paying the cost of accessing any more
8254 : : * blocks).
8255 : : */
8256 [ + + ]: 4360 : if (bottomup_final_block)
8257 : 133 : break;
8258 : :
8259 : : /*
8260 : : * Give up when we didn't enable our caller to free any
8261 : : * additional space as a result of processing the page that we
8262 : : * just finished up with. This rule is the main way in which
8263 : : * we keep the cost of bottom-up deletion under control.
8264 : : */
8265 [ + + + + ]: 4227 : if (nblocksaccessed >= 1 && actualfreespace == lastfreespace)
8266 : 1848 : break;
8267 : 2379 : lastfreespace = actualfreespace; /* for next time */
8268 : :
8269 : : /*
8270 : : * Deletion operation (which is bottom-up) will definitely
8271 : : * access the next block in line. Prepare for that now.
8272 : : *
8273 : : * Decay target free space so that we don't hang on for too
8274 : : * long with a marginal case. (Space target is only truly
8275 : : * helpful when it allows us to recognize that we don't need
8276 : : * to access more than 1 or 2 blocks to satisfy caller due to
8277 : : * agreeable workload characteristics.)
8278 : : *
8279 : : * We are a bit more patient when we encounter contiguous
8280 : : * blocks, though: these are treated as favorable blocks. The
8281 : : * decay process is only applied when the next block in line
8282 : : * is not a favorable/contiguous block. This is not an
8283 : : * exception to the general rule; we still insist on finding
8284 : : * at least one deletable item per block accessed. See
8285 : : * bottomup_nblocksfavorable() for full details of the theory
8286 : : * behind favorable blocks and heap block locality in general.
8287 : : *
8288 : : * Note: The first block in line is always treated as a
8289 : : * favorable block, so the earliest possible point that the
8290 : : * decay can be applied is just before we access the second
8291 : : * block in line. The Assert() verifies this for us.
8292 : : */
8293 [ + + - + ]: 2379 : Assert(nblocksaccessed > 0 || nblocksfavorable > 0);
8294 [ + + ]: 2379 : if (nblocksfavorable > 0)
8295 : 2153 : nblocksfavorable--;
8296 : : else
8297 : 226 : curtargetfreespace /= 2;
8298 : : }
8299 : :
8300 : : /* release old buffer */
8301 [ + + ]: 14167 : if (BufferIsValid(buf))
8302 : 8286 : UnlockReleaseBuffer(buf);
8303 : :
8304 : 14167 : blkno = ItemPointerGetBlockNumber(htid);
1813 8305 : 14167 : buf = ReadBuffer(rel, blkno);
1799 8306 : 14167 : nblocksaccessed++;
8307 [ + + - + ]: 14167 : Assert(!delstate->bottomup ||
8308 : : nblocksaccessed <= BOTTOMUP_MAX_NBLOCKS);
8309 : :
8310 : : #ifdef USE_PREFETCH
8311 : :
8312 : : /*
8313 : : * To maintain the prefetch distance, prefetch one more page for
8314 : : * each page we read.
8315 : : */
8316 : 14167 : index_delete_prefetch_buffer(rel, &prefetch_state, 1);
8317 : : #endif
8318 : :
1813 8319 : 14167 : LockBuffer(buf, BUFFER_LOCK_SHARE);
8320 : :
8321 : 14167 : page = BufferGetPage(buf);
8322 : 14167 : maxoff = PageGetMaxOffsetNumber(page);
8323 : : }
8324 : :
8325 : : /*
8326 : : * In passing, detect index corruption involving an index page with a
8327 : : * TID that points to a location in the heap that couldn't possibly be
8328 : : * correct. We only do this with actual TIDs from caller's index page
8329 : : * (not items reached by traversing through a HOT chain).
8330 : : */
1504 8331 : 555378 : index_delete_check_htid(delstate, page, maxoff, htid, istatus);
8332 : :
1799 8333 [ + + ]: 555378 : if (istatus->knowndeletable)
8334 [ + - - + ]: 138272 : Assert(!delstate->bottomup && !istatus->promising);
8335 : : else
8336 : : {
8337 : 417106 : ItemPointerData tmp = *htid;
8338 : : HeapTupleData heapTuple;
8339 : :
8340 : : /* Are any tuples from this HOT chain non-vacuumable? */
8341 [ + + ]: 417106 : if (heap_hot_search_buffer(&tmp, rel, buf, &SnapshotNonVacuumable,
8342 : : &heapTuple, NULL, true))
8343 : 250167 : continue; /* can't delete entry */
8344 : :
8345 : : /* Caller will delete, since whole HOT chain is vacuumable */
8346 : 166939 : istatus->knowndeletable = true;
8347 : :
8348 : : /* Maintain index free space info for bottom-up deletion case */
8349 [ + + ]: 166939 : if (delstate->bottomup)
8350 : : {
8351 [ - + ]: 8229 : Assert(istatus->freespace > 0);
8352 : 8229 : actualfreespace += istatus->freespace;
8353 [ + + ]: 8229 : if (actualfreespace >= curtargetfreespace)
8354 : 2137 : bottomup_final_block = true;
8355 : : }
8356 : : }
8357 : :
8358 : : /*
8359 : : * Maintain snapshotConflictHorizon value for deletion operation as a
8360 : : * whole by advancing current value using heap tuple headers. This is
8361 : : * loosely based on the logic for pruning a HOT chain.
8362 : : */
1813 8363 : 305211 : offnum = ItemPointerGetOffsetNumber(htid);
8364 : 305211 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
8365 : : for (;;)
2458 andres@anarazel.de 8366 : 20964 : {
8367 : : ItemId lp;
8368 : : HeapTupleHeader htup;
8369 : :
8370 : : /* Sanity check (pure paranoia) */
1547 pg@bowt.ie 8371 [ - + ]: 326175 : if (offnum < FirstOffsetNumber)
1547 pg@bowt.ie 8372 :UBC 0 : break;
8373 : :
8374 : : /*
8375 : : * An offset past the end of page's line pointer array is possible
8376 : : * when the array was truncated
8377 : : */
1547 pg@bowt.ie 8378 [ - + ]:CBC 326175 : if (offnum > maxoff)
1813 pg@bowt.ie 8379 :UBC 0 : break;
8380 : :
1813 pg@bowt.ie 8381 :CBC 326175 : lp = PageGetItemId(page, offnum);
8382 [ + + ]: 326175 : if (ItemIdIsRedirected(lp))
8383 : : {
8384 : 9398 : offnum = ItemIdGetRedirect(lp);
8385 : 9398 : continue;
8386 : : }
8387 : :
8388 : : /*
8389 : : * We'll often encounter LP_DEAD line pointers (especially with an
8390 : : * entry marked knowndeletable by our caller up front). No heap
8391 : : * tuple headers get examined for an htid that leads us to an
8392 : : * LP_DEAD item. This is okay because the earlier pruning
8393 : : * operation that made the line pointer LP_DEAD in the first place
8394 : : * must have considered the original tuple header as part of
8395 : : * generating its own snapshotConflictHorizon value.
8396 : : *
8397 : : * Relying on XLOG_HEAP2_PRUNE_VACUUM_SCAN records like this is
8398 : : * the same strategy that index vacuuming uses in all cases. Index
8399 : : * VACUUM WAL records don't even have a snapshotConflictHorizon
8400 : : * field of their own for this reason.
8401 : : */
8402 [ + + ]: 316777 : if (!ItemIdIsNormal(lp))
8403 : 201051 : break;
8404 : :
8405 : 115726 : htup = (HeapTupleHeader) PageGetItem(page, lp);
8406 : :
8407 : : /*
8408 : : * Check the tuple XMIN against prior XMAX, if any
8409 : : */
8410 [ + + - + ]: 127292 : if (TransactionIdIsValid(priorXmax) &&
8411 : 11566 : !TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
1813 pg@bowt.ie 8412 :UBC 0 : break;
8413 : :
1126 pg@bowt.ie 8414 :CBC 115726 : HeapTupleHeaderAdvanceConflictHorizon(htup,
8415 : : &snapshotConflictHorizon);
8416 : :
8417 : : /*
8418 : : * If the tuple is not HOT-updated, then we are at the end of this
8419 : : * HOT-chain. No need to visit later tuples from the same update
8420 : : * chain (they get their own index entries) -- just move on to
8421 : : * next htid from index AM caller.
8422 : : */
1813 8423 [ + + ]: 115726 : if (!HeapTupleHeaderIsHotUpdated(htup))
8424 : 104160 : break;
8425 : :
8426 : : /* Advance to next HOT chain member */
8427 [ - + ]: 11566 : Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blkno);
8428 : 11566 : offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
8429 : 11566 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
8430 : : }
8431 : :
8432 : : /* Enable further/final shrinking of deltids for caller */
1799 8433 : 305211 : finalndeltids = i + 1;
8434 : : }
8435 : :
8436 : 5881 : UnlockReleaseBuffer(buf);
8437 : :
8438 : : /*
8439 : : * Shrink deltids array to exclude non-deletable entries at the end. This
8440 : : * is not just a minor optimization. Final deltids array size might be
8441 : : * zero for a bottom-up caller. Index AM is explicitly allowed to rely on
8442 : : * ndeltids being zero in all cases with zero total deletable entries.
8443 : : */
8444 [ + + - + ]: 5881 : Assert(finalndeltids > 0 || delstate->bottomup);
8445 : 5881 : delstate->ndeltids = finalndeltids;
8446 : :
1126 8447 : 5881 : return snapshotConflictHorizon;
8448 : : }
8449 : :
8450 : : /*
8451 : : * Specialized inlineable comparison function for index_delete_sort()
8452 : : */
8453 : : static inline int
1799 8454 : 12893858 : index_delete_sort_cmp(TM_IndexDelete *deltid1, TM_IndexDelete *deltid2)
8455 : : {
8456 : 12893858 : ItemPointer tid1 = &deltid1->tid;
8457 : 12893858 : ItemPointer tid2 = &deltid2->tid;
8458 : :
8459 : : {
8460 : 12893858 : BlockNumber blk1 = ItemPointerGetBlockNumber(tid1);
8461 : 12893858 : BlockNumber blk2 = ItemPointerGetBlockNumber(tid2);
8462 : :
8463 [ + + ]: 12893858 : if (blk1 != blk2)
8464 [ + + ]: 5249917 : return (blk1 < blk2) ? -1 : 1;
8465 : : }
8466 : : {
8467 : 7643941 : OffsetNumber pos1 = ItemPointerGetOffsetNumber(tid1);
8468 : 7643941 : OffsetNumber pos2 = ItemPointerGetOffsetNumber(tid2);
8469 : :
8470 [ + - ]: 7643941 : if (pos1 != pos2)
8471 [ + + ]: 7643941 : return (pos1 < pos2) ? -1 : 1;
8472 : : }
8473 : :
1510 pg@bowt.ie 8474 :UBC 0 : Assert(false);
8475 : :
8476 : : return 0;
8477 : : }
8478 : :
8479 : : /*
8480 : : * Sort deltids array from delstate by TID. This prepares it for further
8481 : : * processing by heap_index_delete_tuples().
8482 : : *
8483 : : * This operation becomes a noticeable consumer of CPU cycles with some
8484 : : * workloads, so we go to the trouble of specialization/micro optimization.
8485 : : * We use shellsort for this because it's easy to specialize, compiles to
8486 : : * relatively few instructions, and is adaptive to presorted inputs/subsets
8487 : : * (which are typical here).
8488 : : */
8489 : : static void
1799 pg@bowt.ie 8490 :CBC 5881 : index_delete_sort(TM_IndexDeleteOp *delstate)
8491 : : {
8492 : 5881 : TM_IndexDelete *deltids = delstate->deltids;
8493 : 5881 : int ndeltids = delstate->ndeltids;
8494 : :
8495 : : /*
8496 : : * Shellsort gap sequence (taken from Sedgewick-Incerpi paper).
8497 : : *
8498 : : * This implementation is fast with array sizes up to ~4500. This covers
8499 : : * all supported BLCKSZ values.
8500 : : */
8501 : 5881 : const int gaps[9] = {1968, 861, 336, 112, 48, 21, 7, 3, 1};
8502 : :
8503 : : /* Think carefully before changing anything here -- keep swaps cheap */
8504 : : StaticAssertDecl(sizeof(TM_IndexDelete) <= 8,
8505 : : "element size exceeds 8 bytes");
8506 : :
8507 [ + + ]: 58810 : for (int g = 0; g < lengthof(gaps); g++)
8508 : : {
406 dgustafsson@postgres 8509 [ + + ]: 7685016 : for (int hi = gaps[g], i = hi; i < ndeltids; i++)
8510 : : {
1799 pg@bowt.ie 8511 : 7632087 : TM_IndexDelete d = deltids[i];
8512 : 7632087 : int j = i;
8513 : :
8514 [ + + + + ]: 13267658 : while (j >= hi && index_delete_sort_cmp(&deltids[j - hi], &d) >= 0)
8515 : : {
8516 : 5635571 : deltids[j] = deltids[j - hi];
8517 : 5635571 : j -= hi;
8518 : : }
8519 : 7632087 : deltids[j] = d;
8520 : : }
8521 : : }
8522 : 5881 : }
8523 : :
8524 : : /*
8525 : : * Returns how many blocks should be considered favorable/contiguous for a
8526 : : * bottom-up index deletion pass. This is a number of heap blocks that starts
8527 : : * from and includes the first block in line.
8528 : : *
8529 : : * There is always at least one favorable block during bottom-up index
8530 : : * deletion. In the worst case (i.e. with totally random heap blocks) the
8531 : : * first block in line (the only favorable block) can be thought of as a
8532 : : * degenerate array of contiguous blocks that consists of a single block.
8533 : : * heap_index_delete_tuples() will expect this.
8534 : : *
8535 : : * Caller passes blockgroups, a description of the final order that deltids
8536 : : * will be sorted in for heap_index_delete_tuples() bottom-up index deletion
8537 : : * processing. Note that deltids need not actually be sorted just yet (caller
8538 : : * only passes deltids to us so that we can interpret blockgroups).
8539 : : *
8540 : : * You might guess that the existence of contiguous blocks cannot matter much,
8541 : : * since in general the main factor that determines which blocks we visit is
8542 : : * the number of promising TIDs, which is a fixed hint from the index AM.
8543 : : * We're not really targeting the general case, though -- the actual goal is
8544 : : * to adapt our behavior to a wide variety of naturally occurring conditions.
8545 : : * The effects of most of the heuristics we apply are only noticeable in the
8546 : : * aggregate, over time and across many _related_ bottom-up index deletion
8547 : : * passes.
8548 : : *
8549 : : * Deeming certain blocks favorable allows heapam to recognize and adapt to
8550 : : * workloads where heap blocks visited during bottom-up index deletion can be
8551 : : * accessed contiguously, in the sense that each newly visited block is the
8552 : : * neighbor of the block that bottom-up deletion just finished processing (or
8553 : : * close enough to it). It will likely be cheaper to access more favorable
8554 : : * blocks sooner rather than later (e.g. in this pass, not across a series of
8555 : : * related bottom-up passes). Either way it is probably only a matter of time
8556 : : * (or a matter of further correlated version churn) before all blocks that
8557 : : * appear together as a single large batch of favorable blocks get accessed by
8558 : : * _some_ bottom-up pass. Large batches of favorable blocks tend to either
8559 : : * appear almost constantly or not even once (it all depends on per-index
8560 : : * workload characteristics).
8561 : : *
8562 : : * Note that the blockgroups sort order applies a power-of-two bucketing
8563 : : * scheme that creates opportunities for contiguous groups of blocks to get
8564 : : * batched together, at least with workloads that are naturally amenable to
8565 : : * being driven by heap block locality. This doesn't just enhance the spatial
8566 : : * locality of bottom-up heap block processing in the obvious way. It also
8567 : : * enables temporal locality of access, since sorting by heap block number
8568 : : * naturally tends to make the bottom-up processing order deterministic.
8569 : : *
8570 : : * Consider the following example to get a sense of how temporal locality
8571 : : * might matter: There is a heap relation with several indexes, each of which
8572 : : * is low to medium cardinality. It is subject to constant non-HOT updates.
8573 : : * The updates are skewed (in one part of the primary key, perhaps). None of
8574 : : * the indexes are logically modified by the UPDATE statements (if they were
8575 : : * then bottom-up index deletion would not be triggered in the first place).
8576 : : * Naturally, each new round of index tuples (for each heap tuple that gets a
8577 : : * heap_update() call) will have the same heap TID in each and every index.
8578 : : * Since these indexes are low cardinality and never get logically modified,
8579 : : * heapam processing during bottom-up deletion passes will access heap blocks
8580 : : * in approximately sequential order. Temporal locality of access occurs due
8581 : : * to bottom-up deletion passes behaving very similarly across each of the
8582 : : * indexes at any given moment. This keeps the number of buffer misses needed
8583 : : * to visit heap blocks to a minimum.
8584 : : */
8585 : : static int
8586 : 1981 : bottomup_nblocksfavorable(IndexDeleteCounts *blockgroups, int nblockgroups,
8587 : : TM_IndexDelete *deltids)
8588 : : {
8589 : 1981 : int64 lastblock = -1;
8590 : 1981 : int nblocksfavorable = 0;
8591 : :
8592 [ - + ]: 1981 : Assert(nblockgroups >= 1);
8593 [ - + ]: 1981 : Assert(nblockgroups <= BOTTOMUP_MAX_NBLOCKS);
8594 : :
8595 : : /*
8596 : : * We tolerate heap blocks that will be accessed only slightly out of
8597 : : * physical order. Small blips occur when a pair of almost-contiguous
8598 : : * blocks happen to fall into different buckets (perhaps due only to a
8599 : : * small difference in npromisingtids that the bucketing scheme didn't
8600 : : * quite manage to ignore). We effectively ignore these blips by applying
8601 : : * a small tolerance. The precise tolerance we use is a little arbitrary,
8602 : : * but it works well enough in practice.
8603 : : */
8604 [ + + ]: 6410 : for (int b = 0; b < nblockgroups; b++)
8605 : : {
8606 : 6128 : IndexDeleteCounts *group = blockgroups + b;
8607 : 6128 : TM_IndexDelete *firstdtid = deltids + group->ifirsttid;
8608 : 6128 : BlockNumber block = ItemPointerGetBlockNumber(&firstdtid->tid);
8609 : :
8610 [ + + ]: 6128 : if (lastblock != -1 &&
8611 [ + + ]: 4147 : ((int64) block < lastblock - BOTTOMUP_TOLERANCE_NBLOCKS ||
8612 [ + + ]: 3591 : (int64) block > lastblock + BOTTOMUP_TOLERANCE_NBLOCKS))
8613 : : break;
8614 : :
8615 : 4429 : nblocksfavorable++;
8616 : 4429 : lastblock = block;
8617 : : }
8618 : :
8619 : : /* Always indicate that there is at least 1 favorable block */
8620 [ - + ]: 1981 : Assert(nblocksfavorable >= 1);
8621 : :
8622 : 1981 : return nblocksfavorable;
8623 : : }
8624 : :
8625 : : /*
8626 : : * qsort comparison function for bottomup_sort_and_shrink()
8627 : : */
8628 : : static int
8629 : 201804 : bottomup_sort_and_shrink_cmp(const void *arg1, const void *arg2)
8630 : : {
8631 : 201804 : const IndexDeleteCounts *group1 = (const IndexDeleteCounts *) arg1;
8632 : 201804 : const IndexDeleteCounts *group2 = (const IndexDeleteCounts *) arg2;
8633 : :
8634 : : /*
8635 : : * Most significant field is npromisingtids (which we invert the order of
8636 : : * so as to sort in desc order).
8637 : : *
8638 : : * Caller should have already normalized npromisingtids fields into
8639 : : * power-of-two values (buckets).
8640 : : */
8641 [ + + ]: 201804 : if (group1->npromisingtids > group2->npromisingtids)
8642 : 8847 : return -1;
8643 [ + + ]: 192957 : if (group1->npromisingtids < group2->npromisingtids)
8644 : 10921 : return 1;
8645 : :
8646 : : /*
8647 : : * Tiebreak: desc ntids sort order.
8648 : : *
8649 : : * We cannot expect power-of-two values for ntids fields. We should
8650 : : * behave as if they were already rounded up for us instead.
8651 : : */
8652 [ + + ]: 182036 : if (group1->ntids != group2->ntids)
8653 : : {
8654 : 127414 : uint32 ntids1 = pg_nextpower2_32((uint32) group1->ntids);
8655 : 127414 : uint32 ntids2 = pg_nextpower2_32((uint32) group2->ntids);
8656 : :
8657 [ + + ]: 127414 : if (ntids1 > ntids2)
8658 : 20882 : return -1;
8659 [ + + ]: 106532 : if (ntids1 < ntids2)
8660 : 24843 : return 1;
8661 : : }
8662 : :
8663 : : /*
8664 : : * Tiebreak: asc offset-into-deltids-for-block (offset to first TID for
8665 : : * block in deltids array) order.
8666 : : *
8667 : : * This is equivalent to sorting in ascending heap block number order
8668 : : * (among otherwise equal subsets of the array). This approach allows us
8669 : : * to avoid accessing the out-of-line TID. (We rely on the assumption
8670 : : * that the deltids array was sorted in ascending heap TID order when
8671 : : * these offsets to the first TID from each heap block group were formed.)
8672 : : */
8673 [ + + ]: 136311 : if (group1->ifirsttid > group2->ifirsttid)
8674 : 67261 : return 1;
8675 [ + - ]: 69050 : if (group1->ifirsttid < group2->ifirsttid)
8676 : 69050 : return -1;
8677 : :
1799 pg@bowt.ie 8678 :UBC 0 : pg_unreachable();
8679 : :
8680 : : return 0;
8681 : : }
8682 : :
8683 : : /*
8684 : : * heap_index_delete_tuples() helper function for bottom-up deletion callers.
8685 : : *
8686 : : * Sorts deltids array in the order needed for useful processing by bottom-up
8687 : : * deletion. The array should already be sorted in TID order when we're
8688 : : * called. The sort process groups heap TIDs from deltids into heap block
8689 : : * groupings. Earlier/more-promising groups/blocks are usually those that are
8690 : : * known to have the most "promising" TIDs.
8691 : : *
8692 : : * Sets new size of deltids array (ndeltids) in state. deltids will only have
8693 : : * TIDs from the BOTTOMUP_MAX_NBLOCKS most promising heap blocks when we
8694 : : * return. This often means that deltids will be shrunk to a small fraction
8695 : : * of its original size (we eliminate many heap blocks from consideration for
8696 : : * caller up front).
8697 : : *
8698 : : * Returns the number of "favorable" blocks. See bottomup_nblocksfavorable()
8699 : : * for a definition and full details.
8700 : : */
8701 : : static int
1799 pg@bowt.ie 8702 :CBC 1981 : bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate)
8703 : : {
8704 : : IndexDeleteCounts *blockgroups;
8705 : : TM_IndexDelete *reordereddeltids;
8706 : 1981 : BlockNumber curblock = InvalidBlockNumber;
8707 : 1981 : int nblockgroups = 0;
8708 : 1981 : int ncopied = 0;
8709 : 1981 : int nblocksfavorable = 0;
8710 : :
8711 [ - + ]: 1981 : Assert(delstate->bottomup);
8712 [ - + ]: 1981 : Assert(delstate->ndeltids > 0);
8713 : :
8714 : : /* Calculate per-heap-block count of TIDs */
7 michael@paquier.xyz 8715 :GNC 1981 : blockgroups = palloc_array(IndexDeleteCounts, delstate->ndeltids);
1799 pg@bowt.ie 8716 [ + + ]:CBC 947172 : for (int i = 0; i < delstate->ndeltids; i++)
8717 : : {
8718 : 945191 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8719 : 945191 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8720 : 945191 : ItemPointer htid = &ideltid->tid;
8721 : 945191 : bool promising = istatus->promising;
8722 : :
8723 [ + + ]: 945191 : if (curblock != ItemPointerGetBlockNumber(htid))
8724 : : {
8725 : : /* New block group */
8726 : 39287 : nblockgroups++;
8727 : :
8728 [ + + - + ]: 39287 : Assert(curblock < ItemPointerGetBlockNumber(htid) ||
8729 : : !BlockNumberIsValid(curblock));
8730 : :
8731 : 39287 : curblock = ItemPointerGetBlockNumber(htid);
8732 : 39287 : blockgroups[nblockgroups - 1].ifirsttid = i;
8733 : 39287 : blockgroups[nblockgroups - 1].ntids = 1;
8734 : 39287 : blockgroups[nblockgroups - 1].npromisingtids = 0;
8735 : : }
8736 : : else
8737 : : {
8738 : 905904 : blockgroups[nblockgroups - 1].ntids++;
8739 : : }
8740 : :
8741 [ + + ]: 945191 : if (promising)
8742 : 122049 : blockgroups[nblockgroups - 1].npromisingtids++;
8743 : : }
8744 : :
8745 : : /*
8746 : : * We're about ready to sort block groups to determine the optimal order
8747 : : * for visiting heap blocks. But before we do, round the number of
8748 : : * promising tuples for each block group up to the next power-of-two,
8749 : : * unless it is very low (less than 4), in which case we round up to 4.
8750 : : * npromisingtids is far too noisy to trust when choosing between a pair
8751 : : * of block groups that both have very low values.
8752 : : *
8753 : : * This scheme divides heap blocks/block groups into buckets. Each bucket
8754 : : * contains blocks that have _approximately_ the same number of promising
8755 : : * TIDs as each other. The goal is to ignore relatively small differences
8756 : : * in the total number of promising entries, so that the whole process can
8757 : : * give a little weight to heapam factors (like heap block locality)
8758 : : * instead. This isn't a trade-off, really -- we have nothing to lose. It
8759 : : * would be foolish to interpret small differences in npromisingtids
8760 : : * values as anything more than noise.
8761 : : *
8762 : : * We tiebreak on nhtids when sorting block group subsets that have the
8763 : : * same npromisingtids, but this has the same issues as npromisingtids,
8764 : : * and so nhtids is subject to the same power-of-two bucketing scheme. The
8765 : : * only reason that we don't fix nhtids in the same way here too is that
8766 : : * we'll need accurate nhtids values after the sort. We handle nhtids
8767 : : * bucketization dynamically instead (in the sort comparator).
8768 : : *
8769 : : * See bottomup_nblocksfavorable() for a full explanation of when and how
8770 : : * heap locality/favorable blocks can significantly influence when and how
8771 : : * heap blocks are accessed.
8772 : : */
8773 [ + + ]: 41268 : for (int b = 0; b < nblockgroups; b++)
8774 : : {
8775 : 39287 : IndexDeleteCounts *group = blockgroups + b;
8776 : :
8777 : : /* Better off falling back on nhtids with low npromisingtids */
8778 [ + + ]: 39287 : if (group->npromisingtids <= 4)
8779 : 33801 : group->npromisingtids = 4;
8780 : : else
8781 : 5486 : group->npromisingtids =
8782 : 5486 : pg_nextpower2_32((uint32) group->npromisingtids);
8783 : : }
8784 : :
8785 : : /* Sort groups and rearrange caller's deltids array */
8786 : 1981 : qsort(blockgroups, nblockgroups, sizeof(IndexDeleteCounts),
8787 : : bottomup_sort_and_shrink_cmp);
8788 : 1981 : reordereddeltids = palloc(delstate->ndeltids * sizeof(TM_IndexDelete));
8789 : :
8790 : 1981 : nblockgroups = Min(BOTTOMUP_MAX_NBLOCKS, nblockgroups);
8791 : : /* Determine number of favorable blocks at the start of final deltids */
8792 : 1981 : nblocksfavorable = bottomup_nblocksfavorable(blockgroups, nblockgroups,
8793 : : delstate->deltids);
8794 : :
8795 [ + + ]: 13257 : for (int b = 0; b < nblockgroups; b++)
8796 : : {
8797 : 11276 : IndexDeleteCounts *group = blockgroups + b;
8798 : 11276 : TM_IndexDelete *firstdtid = delstate->deltids + group->ifirsttid;
8799 : :
8800 : 11276 : memcpy(reordereddeltids + ncopied, firstdtid,
8801 : 11276 : sizeof(TM_IndexDelete) * group->ntids);
8802 : 11276 : ncopied += group->ntids;
8803 : : }
8804 : :
8805 : : /* Copy final grouped and sorted TIDs back into start of caller's array */
8806 : 1981 : memcpy(delstate->deltids, reordereddeltids,
8807 : : sizeof(TM_IndexDelete) * ncopied);
8808 : 1981 : delstate->ndeltids = ncopied;
8809 : :
8810 : 1981 : pfree(reordereddeltids);
8811 : 1981 : pfree(blockgroups);
8812 : :
8813 : 1981 : return nblocksfavorable;
8814 : : }
8815 : :
8816 : : /*
8817 : : * Perform XLogInsert for a heap-visible operation. 'block' is the block
8818 : : * being marked all-visible, and vm_buffer is the buffer containing the
8819 : : * corresponding visibility map block. Both should have already been modified
8820 : : * and dirtied.
8821 : : *
8822 : : * snapshotConflictHorizon comes from the largest xmin on the page being
8823 : : * marked all-visible. REDO routine uses it to generate recovery conflicts.
8824 : : *
8825 : : * If checksums or wal_log_hints are enabled, we may also generate a full-page
8826 : : * image of heap_buffer. Otherwise, we optimize away the FPI (by specifying
8827 : : * REGBUF_NO_IMAGE for the heap buffer), in which case the caller should *not*
8828 : : * update the heap page's LSN.
8829 : : */
8830 : : XLogRecPtr
991 andres@anarazel.de 8831 : 36990 : log_heap_visible(Relation rel, Buffer heap_buffer, Buffer vm_buffer,
8832 : : TransactionId snapshotConflictHorizon, uint8 vmflags)
8833 : : {
8834 : : xl_heap_visible xlrec;
8835 : : XLogRecPtr recptr;
8836 : : uint8 flags;
8837 : :
4653 simon@2ndQuadrant.co 8838 [ - + ]: 36990 : Assert(BufferIsValid(heap_buffer));
8839 [ - + ]: 36990 : Assert(BufferIsValid(vm_buffer));
8840 : :
1126 pg@bowt.ie 8841 : 36990 : xlrec.snapshotConflictHorizon = snapshotConflictHorizon;
3578 rhaas@postgresql.org 8842 : 36990 : xlrec.flags = vmflags;
990 andres@anarazel.de 8843 [ + + + - : 36990 : if (RelationIsAccessibleInLogicalDecoding(rel))
- + - - -
- + + - +
- - - - -
- ]
8844 : 49 : xlrec.flags |= VISIBILITYMAP_XLOG_CATALOG_REL;
4045 heikki.linnakangas@i 8845 : 36990 : XLogBeginInsert();
309 peter@eisentraut.org 8846 : 36990 : XLogRegisterData(&xlrec, SizeOfHeapVisible);
8847 : :
4045 heikki.linnakangas@i 8848 : 36990 : XLogRegisterBuffer(0, vm_buffer, 0);
8849 : :
8850 : 36990 : flags = REGBUF_STANDARD;
8851 [ + + + - ]: 36990 : if (!XLogHintBitIsNeeded())
8852 : 3138 : flags |= REGBUF_NO_IMAGE;
8853 : 36990 : XLogRegisterBuffer(1, heap_buffer, flags);
8854 : :
8855 : 36990 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_VISIBLE);
8856 : :
5293 rhaas@postgresql.org 8857 : 36990 : return recptr;
8858 : : }
8859 : :
8860 : : /*
8861 : : * Perform XLogInsert for a heap-update operation. Caller must already
8862 : : * have modified the buffer(s) and marked them dirty.
8863 : : */
8864 : : static XLogRecPtr
4711 alvherre@alvh.no-ip. 8865 : 300589 : log_heap_update(Relation reln, Buffer oldbuf,
8866 : : Buffer newbuf, HeapTuple oldtup, HeapTuple newtup,
8867 : : HeapTuple old_key_tuple,
8868 : : bool all_visible_cleared, bool new_all_visible_cleared)
8869 : : {
8870 : : xl_heap_update xlrec;
8871 : : xl_heap_header xlhdr;
8872 : : xl_heap_header xlhdr_idx;
8873 : : uint8 info;
8874 : : uint16 prefix_suffix[2];
4298 heikki.linnakangas@i 8875 : 300589 : uint16 prefixlen = 0,
8876 : 300589 : suffixlen = 0;
8877 : : XLogRecPtr recptr;
3528 kgrittn@postgresql.o 8878 : 300589 : Page page = BufferGetPage(newbuf);
4390 rhaas@postgresql.org 8879 [ + + + - : 300589 : bool need_tuple_data = RelationIsLogicallyLogged(reln);
- + - - -
- + - +
+ ]
8880 : : bool init;
8881 : : int bufflags;
8882 : :
8883 : : /* Caller should not call me on a non-WAL-logged relation */
5483 8884 [ + - + + : 300589 : Assert(RelationNeedsWAL(reln));
+ - - + ]
8885 : :
4045 heikki.linnakangas@i 8886 : 300589 : XLogBeginInsert();
8887 : :
5791 tgl@sss.pgh.pa.us 8888 [ + + ]: 300589 : if (HeapTupleIsHeapOnly(newtup))
6663 8889 : 146950 : info = XLOG_HEAP_HOT_UPDATE;
8890 : : else
8891 : 153639 : info = XLOG_HEAP_UPDATE;
8892 : :
8893 : : /*
8894 : : * If the old and new tuple are on the same page, we only need to log the
8895 : : * parts of the new tuple that were changed. That saves on the amount of
8896 : : * WAL we need to write. Currently, we just count any unchanged bytes in
8897 : : * the beginning and end of the tuple. That's quick to check, and
8898 : : * perfectly covers the common case that only one field is updated.
8899 : : *
8900 : : * We could do this even if the old and new tuple are on different pages,
8901 : : * but only if we don't make a full-page image of the old page, which is
8902 : : * difficult to know in advance. Also, if the old tuple is corrupt for
8903 : : * some reason, it would allow the corruption to propagate the new page,
8904 : : * so it seems best to avoid. Under the general assumption that most
8905 : : * updates tend to create the new tuple version on the same page, there
8906 : : * isn't much to be gained by doing this across pages anyway.
8907 : : *
8908 : : * Skip this if we're taking a full-page image of the new page, as we
8909 : : * don't include the new tuple in the WAL record in that case. Also
8910 : : * disable if wal_level='logical', as logical decoding needs to be able to
8911 : : * read the new tuple in whole from the WAL record alone.
8912 : : */
4298 heikki.linnakangas@i 8913 [ + + + + ]: 300589 : if (oldbuf == newbuf && !need_tuple_data &&
8914 [ + + ]: 147192 : !XLogCheckBufferNeedsBackup(newbuf))
8915 : : {
8916 : 146565 : char *oldp = (char *) oldtup->t_data + oldtup->t_data->t_hoff;
8917 : 146565 : char *newp = (char *) newtup->t_data + newtup->t_data->t_hoff;
8918 : 146565 : int oldlen = oldtup->t_len - oldtup->t_data->t_hoff;
8919 : 146565 : int newlen = newtup->t_len - newtup->t_data->t_hoff;
8920 : :
8921 : : /* Check for common prefix between old and new tuple */
8922 [ + + ]: 12348803 : for (prefixlen = 0; prefixlen < Min(oldlen, newlen); prefixlen++)
8923 : : {
8924 [ + + ]: 12322646 : if (newp[prefixlen] != oldp[prefixlen])
8925 : 120408 : break;
8926 : : }
8927 : :
8928 : : /*
8929 : : * Storing the length of the prefix takes 2 bytes, so we need to save
8930 : : * at least 3 bytes or there's no point.
8931 : : */
8932 [ + + ]: 146565 : if (prefixlen < 3)
8933 : 22100 : prefixlen = 0;
8934 : :
8935 : : /* Same for suffix */
8936 [ + + ]: 4771251 : for (suffixlen = 0; suffixlen < Min(oldlen, newlen) - prefixlen; suffixlen++)
8937 : : {
8938 [ + + ]: 4744838 : if (newp[newlen - suffixlen - 1] != oldp[oldlen - suffixlen - 1])
8939 : 120152 : break;
8940 : : }
8941 [ + + ]: 146565 : if (suffixlen < 3)
8942 : 36927 : suffixlen = 0;
8943 : : }
8944 : :
8945 : : /* Prepare main WAL data chain */
4390 rhaas@postgresql.org 8946 : 300589 : xlrec.flags = 0;
8947 [ + + ]: 300589 : if (all_visible_cleared)
3876 andres@anarazel.de 8948 : 1634 : xlrec.flags |= XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED;
4390 rhaas@postgresql.org 8949 [ + + ]: 300589 : if (new_all_visible_cleared)
3876 andres@anarazel.de 8950 : 1016 : xlrec.flags |= XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED;
4298 heikki.linnakangas@i 8951 [ + + ]: 300589 : if (prefixlen > 0)
3876 andres@anarazel.de 8952 : 124465 : xlrec.flags |= XLH_UPDATE_PREFIX_FROM_OLD;
4298 heikki.linnakangas@i 8953 [ + + ]: 300589 : if (suffixlen > 0)
3876 andres@anarazel.de 8954 : 109638 : xlrec.flags |= XLH_UPDATE_SUFFIX_FROM_OLD;
4045 heikki.linnakangas@i 8955 [ + + ]: 300589 : if (need_tuple_data)
8956 : : {
3876 andres@anarazel.de 8957 : 47015 : xlrec.flags |= XLH_UPDATE_CONTAINS_NEW_TUPLE;
4045 heikki.linnakangas@i 8958 [ + + ]: 47015 : if (old_key_tuple)
8959 : : {
8960 [ + + ]: 145 : if (reln->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
3876 andres@anarazel.de 8961 : 64 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_TUPLE;
8962 : : else
8963 : 81 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_KEY;
8964 : : }
8965 : : }
8966 : :
8967 : : /* If new tuple is the single and first tuple on page... */
4298 heikki.linnakangas@i 8968 [ + + + + ]: 304117 : if (ItemPointerGetOffsetNumber(&(newtup->t_self)) == FirstOffsetNumber &&
8969 : 3528 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
8970 : : {
8971 : 3233 : info |= XLOG_HEAP_INIT_PAGE;
4045 8972 : 3233 : init = true;
8973 : : }
8974 : : else
8975 : 297356 : init = false;
8976 : :
8977 : : /* Prepare WAL data for the old page */
8978 : 300589 : xlrec.old_offnum = ItemPointerGetOffsetNumber(&oldtup->t_self);
8979 : 300589 : xlrec.old_xmax = HeapTupleHeaderGetRawXmax(oldtup->t_data);
8980 : 601178 : xlrec.old_infobits_set = compute_infobits(oldtup->t_data->t_infomask,
8981 : 300589 : oldtup->t_data->t_infomask2);
8982 : :
8983 : : /* Prepare WAL data for the new page */
8984 : 300589 : xlrec.new_offnum = ItemPointerGetOffsetNumber(&newtup->t_self);
8985 : 300589 : xlrec.new_xmax = HeapTupleHeaderGetRawXmax(newtup->t_data);
8986 : :
8987 : 300589 : bufflags = REGBUF_STANDARD;
8988 [ + + ]: 300589 : if (init)
8989 : 3233 : bufflags |= REGBUF_WILL_INIT;
8990 [ + + ]: 300589 : if (need_tuple_data)
8991 : 47015 : bufflags |= REGBUF_KEEP_DATA;
8992 : :
8993 : 300589 : XLogRegisterBuffer(0, newbuf, bufflags);
8994 [ + + ]: 300589 : if (oldbuf != newbuf)
8995 : 141462 : XLogRegisterBuffer(1, oldbuf, REGBUF_STANDARD);
8996 : :
309 peter@eisentraut.org 8997 : 300589 : XLogRegisterData(&xlrec, SizeOfHeapUpdate);
8998 : :
8999 : : /*
9000 : : * Prepare WAL data for the new tuple.
9001 : : */
4298 heikki.linnakangas@i 9002 [ + + + + ]: 300589 : if (prefixlen > 0 || suffixlen > 0)
9003 : : {
9004 [ + + + + ]: 146104 : if (prefixlen > 0 && suffixlen > 0)
9005 : : {
9006 : 87999 : prefix_suffix[0] = prefixlen;
9007 : 87999 : prefix_suffix[1] = suffixlen;
309 peter@eisentraut.org 9008 : 87999 : XLogRegisterBufData(0, &prefix_suffix, sizeof(uint16) * 2);
9009 : : }
4298 heikki.linnakangas@i 9010 [ + + ]: 58105 : else if (prefixlen > 0)
9011 : : {
309 peter@eisentraut.org 9012 : 36466 : XLogRegisterBufData(0, &prefixlen, sizeof(uint16));
9013 : : }
9014 : : else
9015 : : {
9016 : 21639 : XLogRegisterBufData(0, &suffixlen, sizeof(uint16));
9017 : : }
9018 : : }
9019 : :
4045 heikki.linnakangas@i 9020 : 300589 : xlhdr.t_infomask2 = newtup->t_data->t_infomask2;
9021 : 300589 : xlhdr.t_infomask = newtup->t_data->t_infomask;
9022 : 300589 : xlhdr.t_hoff = newtup->t_data->t_hoff;
3952 tgl@sss.pgh.pa.us 9023 [ - + ]: 300589 : Assert(SizeofHeapTupleHeader + prefixlen + suffixlen <= newtup->t_len);
9024 : :
9025 : : /*
9026 : : * PG73FORMAT: write bitmap [+ padding] [+ oid] + data
9027 : : *
9028 : : * The 'data' doesn't include the common prefix or suffix.
9029 : : */
309 peter@eisentraut.org 9030 : 300589 : XLogRegisterBufData(0, &xlhdr, SizeOfHeapHeader);
4298 heikki.linnakangas@i 9031 [ + + ]: 300589 : if (prefixlen == 0)
9032 : : {
4045 9033 : 176124 : XLogRegisterBufData(0,
309 peter@eisentraut.org 9034 : 176124 : (char *) newtup->t_data + SizeofHeapTupleHeader,
3101 tgl@sss.pgh.pa.us 9035 : 176124 : newtup->t_len - SizeofHeapTupleHeader - suffixlen);
9036 : : }
9037 : : else
9038 : : {
9039 : : /*
9040 : : * Have to write the null bitmap and data after the common prefix as
9041 : : * two separate rdata entries.
9042 : : */
9043 : : /* bitmap [+ padding] [+ oid] */
3952 9044 [ + - ]: 124465 : if (newtup->t_data->t_hoff - SizeofHeapTupleHeader > 0)
9045 : : {
4045 heikki.linnakangas@i 9046 : 124465 : XLogRegisterBufData(0,
309 peter@eisentraut.org 9047 : 124465 : (char *) newtup->t_data + SizeofHeapTupleHeader,
3101 tgl@sss.pgh.pa.us 9048 : 124465 : newtup->t_data->t_hoff - SizeofHeapTupleHeader);
9049 : : }
9050 : :
9051 : : /* data after common prefix */
4045 heikki.linnakangas@i 9052 : 124465 : XLogRegisterBufData(0,
309 peter@eisentraut.org 9053 : 124465 : (char *) newtup->t_data + newtup->t_data->t_hoff + prefixlen,
3101 tgl@sss.pgh.pa.us 9054 : 124465 : newtup->t_len - newtup->t_data->t_hoff - prefixlen - suffixlen);
9055 : : }
9056 : :
9057 : : /* We need to log a tuple identity */
4045 heikki.linnakangas@i 9058 [ + + + + ]: 300589 : if (need_tuple_data && old_key_tuple)
9059 : : {
9060 : : /* don't really need this, but its more comfy to decode */
9061 : 145 : xlhdr_idx.t_infomask2 = old_key_tuple->t_data->t_infomask2;
9062 : 145 : xlhdr_idx.t_infomask = old_key_tuple->t_data->t_infomask;
9063 : 145 : xlhdr_idx.t_hoff = old_key_tuple->t_data->t_hoff;
9064 : :
309 peter@eisentraut.org 9065 : 145 : XLogRegisterData(&xlhdr_idx, SizeOfHeapHeader);
9066 : :
9067 : : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
3952 tgl@sss.pgh.pa.us 9068 : 145 : XLogRegisterData((char *) old_key_tuple->t_data + SizeofHeapTupleHeader,
9069 : 145 : old_key_tuple->t_len - SizeofHeapTupleHeader);
9070 : : }
9071 : :
9072 : : /* filtering by origin on a row level is much more efficient */
3282 andres@anarazel.de 9073 : 300589 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
9074 : :
4045 heikki.linnakangas@i 9075 : 300589 : recptr = XLogInsert(RM_HEAP_ID, info);
9076 : :
7280 neilc@samurai.com 9077 : 300589 : return recptr;
9078 : : }
9079 : :
9080 : : /*
9081 : : * Perform XLogInsert of an XLOG_HEAP2_NEW_CID record
9082 : : *
9083 : : * This is only used in wal_level >= WAL_LEVEL_LOGICAL, and only for catalog
9084 : : * tuples.
9085 : : */
9086 : : static XLogRecPtr
4390 rhaas@postgresql.org 9087 : 23990 : log_heap_new_cid(Relation relation, HeapTuple tup)
9088 : : {
9089 : : xl_heap_new_cid xlrec;
9090 : :
9091 : : XLogRecPtr recptr;
9092 : 23990 : HeapTupleHeader hdr = tup->t_data;
9093 : :
9094 [ - + ]: 23990 : Assert(ItemPointerIsValid(&tup->t_self));
9095 [ - + ]: 23990 : Assert(tup->t_tableOid != InvalidOid);
9096 : :
9097 : 23990 : xlrec.top_xid = GetTopTransactionId();
1260 9098 : 23990 : xlrec.target_locator = relation->rd_locator;
4045 heikki.linnakangas@i 9099 : 23990 : xlrec.target_tid = tup->t_self;
9100 : :
9101 : : /*
9102 : : * If the tuple got inserted & deleted in the same TX we definitely have a
9103 : : * combo CID, set cmin and cmax.
9104 : : */
4390 rhaas@postgresql.org 9105 [ + + ]: 23990 : if (hdr->t_infomask & HEAP_COMBOCID)
9106 : : {
9107 [ - + ]: 1973 : Assert(!(hdr->t_infomask & HEAP_XMAX_INVALID));
4378 9108 [ - + ]: 1973 : Assert(!HeapTupleHeaderXminInvalid(hdr));
4390 9109 : 1973 : xlrec.cmin = HeapTupleHeaderGetCmin(hdr);
9110 : 1973 : xlrec.cmax = HeapTupleHeaderGetCmax(hdr);
9111 : 1973 : xlrec.combocid = HeapTupleHeaderGetRawCommandId(hdr);
9112 : : }
9113 : : /* No combo CID, so only cmin or cmax can be set by this TX */
9114 : : else
9115 : : {
9116 : : /*
9117 : : * Tuple inserted.
9118 : : *
9119 : : * We need to check for LOCK ONLY because multixacts might be
9120 : : * transferred to the new tuple in case of FOR KEY SHARE updates in
9121 : : * which case there will be an xmax, although the tuple just got
9122 : : * inserted.
9123 : : */
9124 [ + + + + ]: 28746 : if (hdr->t_infomask & HEAP_XMAX_INVALID ||
9125 : 6729 : HEAP_XMAX_IS_LOCKED_ONLY(hdr->t_infomask))
9126 : : {
9127 : 15289 : xlrec.cmin = HeapTupleHeaderGetRawCommandId(hdr);
9128 : 15289 : xlrec.cmax = InvalidCommandId;
9129 : : }
9130 : : /* Tuple from a different tx updated or deleted. */
9131 : : else
9132 : : {
9133 : 6728 : xlrec.cmin = InvalidCommandId;
9134 : 6728 : xlrec.cmax = HeapTupleHeaderGetRawCommandId(hdr);
9135 : : }
9136 : 22017 : xlrec.combocid = InvalidCommandId;
9137 : : }
9138 : :
9139 : : /*
9140 : : * Note that we don't need to register the buffer here, because this
9141 : : * operation does not modify the page. The insert/update/delete that
9142 : : * called us certainly did, but that's WAL-logged separately.
9143 : : */
4045 heikki.linnakangas@i 9144 : 23990 : XLogBeginInsert();
309 peter@eisentraut.org 9145 : 23990 : XLogRegisterData(&xlrec, SizeOfHeapNewCid);
9146 : :
9147 : : /* will be looked at irrespective of origin */
9148 : :
4045 heikki.linnakangas@i 9149 : 23990 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_NEW_CID);
9150 : :
4390 rhaas@postgresql.org 9151 : 23990 : return recptr;
9152 : : }
9153 : :
9154 : : /*
9155 : : * Build a heap tuple representing the configured REPLICA IDENTITY to represent
9156 : : * the old tuple in an UPDATE or DELETE.
9157 : : *
9158 : : * Returns NULL if there's no need to log an identity or if there's no suitable
9159 : : * key defined.
9160 : : *
9161 : : * Pass key_required true if any replica identity columns changed value, or if
9162 : : * any of them have any external data. Delete must always pass true.
9163 : : *
9164 : : * *copy is set to true if the returned tuple is a modified copy rather than
9165 : : * the same tuple that was passed in.
9166 : : */
9167 : : static HeapTuple
1402 akapila@postgresql.o 9168 : 1766376 : ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
9169 : : bool *copy)
9170 : : {
4390 rhaas@postgresql.org 9171 : 1766376 : TupleDesc desc = RelationGetDescr(relation);
9172 : 1766376 : char replident = relation->rd_rel->relreplident;
9173 : : Bitmapset *idattrs;
9174 : : HeapTuple key_tuple;
9175 : : bool nulls[MaxHeapAttributeNumber];
9176 : : Datum values[MaxHeapAttributeNumber];
9177 : :
9178 : 1766376 : *copy = false;
9179 : :
9180 [ + + + + : 1766376 : if (!RelationIsLogicallyLogged(relation))
- + - - -
- + - +
+ ]
9181 : 1666092 : return NULL;
9182 : :
9183 [ + + ]: 100284 : if (replident == REPLICA_IDENTITY_NOTHING)
9184 : 231 : return NULL;
9185 : :
9186 [ + + ]: 100053 : if (replident == REPLICA_IDENTITY_FULL)
9187 : : {
9188 : : /*
9189 : : * When logging the entire old tuple, it very well could contain
9190 : : * toasted columns. If so, force them to be inlined.
9191 : : */
9192 [ + + ]: 194 : if (HeapTupleHasExternal(tp))
9193 : : {
9194 : 4 : *copy = true;
2298 tgl@sss.pgh.pa.us 9195 : 4 : tp = toast_flatten_tuple(tp, desc);
9196 : : }
4390 rhaas@postgresql.org 9197 : 194 : return tp;
9198 : : }
9199 : :
9200 : : /* if the key isn't required and we're only logging the key, we're done */
1402 akapila@postgresql.o 9201 [ + + ]: 99859 : if (!key_required)
4390 rhaas@postgresql.org 9202 : 46870 : return NULL;
9203 : :
9204 : : /* find out the replica identity columns */
2298 tgl@sss.pgh.pa.us 9205 : 52989 : idattrs = RelationGetIndexAttrBitmap(relation,
9206 : : INDEX_ATTR_BITMAP_IDENTITY_KEY);
9207 : :
9208 : : /*
9209 : : * If there's no defined replica identity columns, treat as !key_required.
9210 : : * (This case should not be reachable from heap_update, since that should
9211 : : * calculate key_required accurately. But heap_delete just passes
9212 : : * constant true for key_required, so we can hit this case in deletes.)
9213 : : */
9214 [ + + ]: 52989 : if (bms_is_empty(idattrs))
9215 : 6021 : return NULL;
9216 : :
9217 : : /*
9218 : : * Construct a new tuple containing only the replica identity columns,
9219 : : * with nulls elsewhere. While we're at it, assert that the replica
9220 : : * identity columns aren't null.
9221 : : */
9222 : 46968 : heap_deform_tuple(tp, desc, values, nulls);
9223 : :
9224 [ + + ]: 150898 : for (int i = 0; i < desc->natts; i++)
9225 : : {
9226 [ + + ]: 103930 : if (bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
9227 : : idattrs))
9228 [ - + ]: 46980 : Assert(!nulls[i]);
9229 : : else
9230 : 56950 : nulls[i] = true;
9231 : : }
9232 : :
4390 rhaas@postgresql.org 9233 : 46968 : key_tuple = heap_form_tuple(desc, values, nulls);
9234 : 46968 : *copy = true;
9235 : :
2298 tgl@sss.pgh.pa.us 9236 : 46968 : bms_free(idattrs);
9237 : :
9238 : : /*
9239 : : * If the tuple, which by here only contains indexed columns, still has
9240 : : * toasted columns, force them to be inlined. This is somewhat unlikely
9241 : : * since there's limits on the size of indexed columns, so we don't
9242 : : * duplicate toast_flatten_tuple()s functionality in the above loop over
9243 : : * the indexed columns, even if it would be more efficient.
9244 : : */
4390 rhaas@postgresql.org 9245 [ + + ]: 46968 : if (HeapTupleHasExternal(key_tuple))
9246 : : {
4243 bruce@momjian.us 9247 : 4 : HeapTuple oldtup = key_tuple;
9248 : :
2298 tgl@sss.pgh.pa.us 9249 : 4 : key_tuple = toast_flatten_tuple(oldtup, desc);
4390 rhaas@postgresql.org 9250 : 4 : heap_freetuple(oldtup);
9251 : : }
9252 : :
9253 : 46968 : return key_tuple;
9254 : : }
9255 : :
9256 : : /*
9257 : : * HeapCheckForSerializableConflictOut
9258 : : * We are reading a tuple. If it's not visible, there may be a
9259 : : * rw-conflict out with the inserter. Otherwise, if it is visible to us
9260 : : * but has been deleted, there may be a rw-conflict out with the deleter.
9261 : : *
9262 : : * We will determine the top level xid of the writing transaction with which
9263 : : * we may be in conflict, and ask CheckForSerializableConflictOut() to check
9264 : : * for overlap with our own transaction.
9265 : : *
9266 : : * This function should be called just about anywhere in heapam.c where a
9267 : : * tuple has been read. The caller must hold at least a shared lock on the
9268 : : * buffer, because this function might set hint bits on the tuple. There is
9269 : : * currently no known reason to call this function from an index AM.
9270 : : */
9271 : : void
2150 tmunro@postgresql.or 9272 : 30512187 : HeapCheckForSerializableConflictOut(bool visible, Relation relation,
9273 : : HeapTuple tuple, Buffer buffer,
9274 : : Snapshot snapshot)
9275 : : {
9276 : : TransactionId xid;
9277 : : HTSV_Result htsvResult;
9278 : :
9279 [ + + ]: 30512187 : if (!CheckForSerializableConflictOutNeeded(relation, snapshot))
9280 : 30486813 : return;
9281 : :
9282 : : /*
9283 : : * Check to see whether the tuple has been written to by a concurrent
9284 : : * transaction, either to create it not visible to us, or to delete it
9285 : : * while it is visible to us. The "visible" bool indicates whether the
9286 : : * tuple is visible to us, while HeapTupleSatisfiesVacuum checks what else
9287 : : * is going on with it.
9288 : : *
9289 : : * In the event of a concurrently inserted tuple that also happens to have
9290 : : * been concurrently updated (by a separate transaction), the xmin of the
9291 : : * tuple will be used -- not the updater's xid.
9292 : : */
9293 : 25374 : htsvResult = HeapTupleSatisfiesVacuum(tuple, TransactionXmin, buffer);
9294 [ + + + + : 25374 : switch (htsvResult)
- ]
9295 : : {
9296 : 24561 : case HEAPTUPLE_LIVE:
9297 [ + + ]: 24561 : if (visible)
9298 : 24548 : return;
9299 : 13 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9300 : 13 : break;
9301 : 361 : case HEAPTUPLE_RECENTLY_DEAD:
9302 : : case HEAPTUPLE_DELETE_IN_PROGRESS:
2015 pg@bowt.ie 9303 [ + + ]: 361 : if (visible)
9304 : 285 : xid = HeapTupleHeaderGetUpdateXid(tuple->t_data);
9305 : : else
9306 : 76 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9307 : :
9308 [ + + ]: 361 : if (TransactionIdPrecedes(xid, TransactionXmin))
9309 : : {
9310 : : /* This is like the HEAPTUPLE_DEAD case */
9311 [ - + ]: 67 : Assert(!visible);
9312 : 67 : return;
9313 : : }
2150 tmunro@postgresql.or 9314 : 294 : break;
9315 : 328 : case HEAPTUPLE_INSERT_IN_PROGRESS:
9316 : 328 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
9317 : 328 : break;
9318 : 124 : case HEAPTUPLE_DEAD:
2015 pg@bowt.ie 9319 [ - + ]: 124 : Assert(!visible);
2150 tmunro@postgresql.or 9320 : 124 : return;
2150 tmunro@postgresql.or 9321 :UBC 0 : default:
9322 : :
9323 : : /*
9324 : : * The only way to get to this default clause is if a new value is
9325 : : * added to the enum type without adding it to this switch
9326 : : * statement. That's a bug, so elog.
9327 : : */
9328 [ # # ]: 0 : elog(ERROR, "unrecognized return value from HeapTupleSatisfiesVacuum: %u", htsvResult);
9329 : :
9330 : : /*
9331 : : * In spite of having all enum values covered and calling elog on
9332 : : * this default, some compilers think this is a code path which
9333 : : * allows xid to be used below without initialization. Silence
9334 : : * that warning.
9335 : : */
9336 : : xid = InvalidTransactionId;
9337 : : }
9338 : :
2150 tmunro@postgresql.or 9339 [ - + ]:CBC 635 : Assert(TransactionIdIsValid(xid));
9340 [ - + ]: 635 : Assert(TransactionIdFollowsOrEquals(xid, TransactionXmin));
9341 : :
9342 : : /*
9343 : : * Find top level xid. Bail out if xid is too early to be a conflict, or
9344 : : * if it's our own xid.
9345 : : */
9346 [ + + ]: 635 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
9347 : 64 : return;
9348 : 571 : xid = SubTransGetTopmostTransaction(xid);
9349 [ - + ]: 571 : if (TransactionIdPrecedes(xid, TransactionXmin))
2150 tmunro@postgresql.or 9350 :UBC 0 : return;
9351 : :
2150 tmunro@postgresql.or 9352 :CBC 571 : CheckForSerializableConflictOut(relation, xid, snapshot);
9353 : : }
|
