Age Owner Branch data TLA Line data Source code
1 : : /*-------------------------------------------------------------------------
2 : : *
3 : : * heapam.c
4 : : * heap access method code
5 : : *
6 : : * Portions Copyright (c) 1996-2024, PostgreSQL Global Development Group
7 : : * Portions Copyright (c) 1994, Regents of the University of California
8 : : *
9 : : *
10 : : * IDENTIFICATION
11 : : * src/backend/access/heap/heapam.c
12 : : *
13 : : *
14 : : * INTERFACE ROUTINES
15 : : * heap_beginscan - begin relation scan
16 : : * heap_rescan - restart a relation scan
17 : : * heap_endscan - end relation scan
18 : : * heap_getnext - retrieve next tuple in scan
19 : : * heap_fetch - retrieve tuple with given tid
20 : : * heap_insert - insert tuple into a relation
21 : : * heap_multi_insert - insert multiple tuples into a relation
22 : : * heap_delete - delete a tuple from a relation
23 : : * heap_update - replace a tuple in a relation with another tuple
24 : : *
25 : : * NOTES
26 : : * This file contains the heap_ routines which implement
27 : : * the POSTGRES heap access method used for all POSTGRES
28 : : * relations.
29 : : *
30 : : *-------------------------------------------------------------------------
31 : : */
32 : : #include "postgres.h"
33 : :
34 : : #include "access/bufmask.h"
35 : : #include "access/heapam.h"
36 : : #include "access/heapam_xlog.h"
37 : : #include "access/heaptoast.h"
38 : : #include "access/hio.h"
39 : : #include "access/multixact.h"
40 : : #include "access/parallel.h"
41 : : #include "access/relscan.h"
42 : : #include "access/subtrans.h"
43 : : #include "access/syncscan.h"
44 : : #include "access/sysattr.h"
45 : : #include "access/tableam.h"
46 : : #include "access/transam.h"
47 : : #include "access/valid.h"
48 : : #include "access/visibilitymap.h"
49 : : #include "access/xact.h"
50 : : #include "access/xlog.h"
51 : : #include "access/xloginsert.h"
52 : : #include "access/xlogutils.h"
53 : : #include "catalog/catalog.h"
54 : : #include "commands/vacuum.h"
55 : : #include "miscadmin.h"
56 : : #include "pgstat.h"
57 : : #include "port/atomics.h"
58 : : #include "port/pg_bitutils.h"
59 : : #include "storage/bufmgr.h"
60 : : #include "storage/freespace.h"
61 : : #include "storage/lmgr.h"
62 : : #include "storage/predicate.h"
63 : : #include "storage/procarray.h"
64 : : #include "storage/standby.h"
65 : : #include "utils/datum.h"
66 : : #include "utils/inval.h"
67 : : #include "utils/relcache.h"
68 : : #include "utils/snapmgr.h"
69 : : #include "utils/spccache.h"
70 : :
71 : :
72 : : static HeapTuple heap_prepare_insert(Relation relation, HeapTuple tup,
73 : : TransactionId xid, CommandId cid, int options);
74 : : static XLogRecPtr log_heap_update(Relation reln, Buffer oldbuf,
75 : : Buffer newbuf, HeapTuple oldtup,
76 : : HeapTuple newtup, HeapTuple old_key_tuple,
77 : : bool all_visible_cleared, bool new_all_visible_cleared);
78 : : static Bitmapset *HeapDetermineColumnsInfo(Relation relation,
79 : : Bitmapset *interesting_cols,
80 : : Bitmapset *external_cols,
81 : : HeapTuple oldtup, HeapTuple newtup,
82 : : bool *has_external);
83 : : static bool heap_acquire_tuplock(Relation relation, ItemPointer tid,
84 : : LockTupleMode mode, LockWaitPolicy wait_policy,
85 : : bool *have_tuple_lock);
86 : : static inline BlockNumber heapgettup_advance_block(HeapScanDesc scan,
87 : : BlockNumber block,
88 : : ScanDirection dir);
89 : : static pg_noinline BlockNumber heapgettup_initial_block(HeapScanDesc scan,
90 : : ScanDirection dir);
91 : : static void compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
92 : : uint16 old_infomask2, TransactionId add_to_xmax,
93 : : LockTupleMode mode, bool is_update,
94 : : TransactionId *result_xmax, uint16 *result_infomask,
95 : : uint16 *result_infomask2);
96 : : static TM_Result heap_lock_updated_tuple(Relation rel, HeapTuple tuple,
97 : : ItemPointer ctid, TransactionId xid,
98 : : LockTupleMode mode);
99 : : static void GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
100 : : uint16 *new_infomask2);
101 : : static TransactionId MultiXactIdGetUpdateXid(TransactionId xmax,
102 : : uint16 t_infomask);
103 : : static bool DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
104 : : LockTupleMode lockmode, bool *current_is_member);
105 : : static void MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
106 : : Relation rel, ItemPointer ctid, XLTW_Oper oper,
107 : : int *remaining);
108 : : static bool ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
109 : : uint16 infomask, Relation rel, int *remaining);
110 : : static void index_delete_sort(TM_IndexDeleteOp *delstate);
111 : : static int bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate);
112 : : static XLogRecPtr log_heap_new_cid(Relation relation, HeapTuple tup);
113 : : static HeapTuple ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
114 : : bool *copy);
115 : :
116 : :
117 : : /*
118 : : * Each tuple lock mode has a corresponding heavyweight lock, and one or two
119 : : * corresponding MultiXactStatuses (one to merely lock tuples, another one to
120 : : * update them). This table (and the macros below) helps us determine the
121 : : * heavyweight lock mode and MultiXactStatus values to use for any particular
122 : : * tuple lock strength.
123 : : *
124 : : * Don't look at lockstatus/updstatus directly! Use get_mxact_status_for_lock
125 : : * instead.
126 : : */
127 : : static const struct
128 : : {
129 : : LOCKMODE hwlock;
130 : : int lockstatus;
131 : : int updstatus;
132 : : }
133 : :
134 : : tupleLockExtraInfo[MaxLockTupleMode + 1] =
135 : : {
136 : : { /* LockTupleKeyShare */
137 : : AccessShareLock,
138 : : MultiXactStatusForKeyShare,
139 : : -1 /* KeyShare does not allow updating tuples */
140 : : },
141 : : { /* LockTupleShare */
142 : : RowShareLock,
143 : : MultiXactStatusForShare,
144 : : -1 /* Share does not allow updating tuples */
145 : : },
146 : : { /* LockTupleNoKeyExclusive */
147 : : ExclusiveLock,
148 : : MultiXactStatusForNoKeyUpdate,
149 : : MultiXactStatusNoKeyUpdate
150 : : },
151 : : { /* LockTupleExclusive */
152 : : AccessExclusiveLock,
153 : : MultiXactStatusForUpdate,
154 : : MultiXactStatusUpdate
155 : : }
156 : : };
157 : :
158 : : /* Get the LOCKMODE for a given MultiXactStatus */
159 : : #define LOCKMODE_from_mxstatus(status) \
160 : : (tupleLockExtraInfo[TUPLOCK_from_mxstatus((status))].hwlock)
161 : :
162 : : /*
163 : : * Acquire heavyweight locks on tuples, using a LockTupleMode strength value.
164 : : * This is more readable than having every caller translate it to lock.h's
165 : : * LOCKMODE.
166 : : */
167 : : #define LockTupleTuplock(rel, tup, mode) \
168 : : LockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
169 : : #define UnlockTupleTuplock(rel, tup, mode) \
170 : : UnlockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
171 : : #define ConditionalLockTupleTuplock(rel, tup, mode) \
172 : : ConditionalLockTuple((rel), (tup), tupleLockExtraInfo[mode].hwlock)
173 : :
174 : : #ifdef USE_PREFETCH
175 : : /*
176 : : * heap_index_delete_tuples and index_delete_prefetch_buffer use this
177 : : * structure to coordinate prefetching activity
178 : : */
179 : : typedef struct
180 : : {
181 : : BlockNumber cur_hblkno;
182 : : int next_item;
183 : : int ndeltids;
184 : : TM_IndexDelete *deltids;
185 : : } IndexDeletePrefetchState;
186 : : #endif
187 : :
188 : : /* heap_index_delete_tuples bottom-up index deletion costing constants */
189 : : #define BOTTOMUP_MAX_NBLOCKS 6
190 : : #define BOTTOMUP_TOLERANCE_NBLOCKS 3
191 : :
192 : : /*
193 : : * heap_index_delete_tuples uses this when determining which heap blocks it
194 : : * must visit to help its bottom-up index deletion caller
195 : : */
196 : : typedef struct IndexDeleteCounts
197 : : {
198 : : int16 npromisingtids; /* Number of "promising" TIDs in group */
199 : : int16 ntids; /* Number of TIDs in group */
200 : : int16 ifirsttid; /* Offset to group's first deltid */
201 : : } IndexDeleteCounts;
202 : :
203 : : /*
204 : : * This table maps tuple lock strength values for each particular
205 : : * MultiXactStatus value.
206 : : */
207 : : static const int MultiXactStatusLock[MaxMultiXactStatus + 1] =
208 : : {
209 : : LockTupleKeyShare, /* ForKeyShare */
210 : : LockTupleShare, /* ForShare */
211 : : LockTupleNoKeyExclusive, /* ForNoKeyUpdate */
212 : : LockTupleExclusive, /* ForUpdate */
213 : : LockTupleNoKeyExclusive, /* NoKeyUpdate */
214 : : LockTupleExclusive /* Update */
215 : : };
216 : :
217 : : /* Get the LockTupleMode for a given MultiXactStatus */
218 : : #define TUPLOCK_from_mxstatus(status) \
219 : : (MultiXactStatusLock[(status)])
220 : :
221 : : /* ----------------------------------------------------------------
222 : : * heap support routines
223 : : * ----------------------------------------------------------------
224 : : */
225 : :
226 : : /*
227 : : * Streaming read API callback for parallel sequential scans. Returns the next
228 : : * block the caller wants from the read stream or InvalidBlockNumber when done.
229 : : */
230 : : static BlockNumber
6 tmunro@postgresql.or 231 :GNC 100170 : heap_scan_stream_read_next_parallel(ReadStream *stream,
232 : : void *callback_private_data,
233 : : void *per_buffer_data)
234 : : {
235 : 100170 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
236 : :
237 [ - + ]: 100170 : Assert(ScanDirectionIsForward(scan->rs_dir));
238 [ - + ]: 100170 : Assert(scan->rs_base.rs_parallel);
239 : :
240 [ + + ]: 100170 : if (unlikely(!scan->rs_inited))
241 : : {
242 : : /* parallel scan */
243 : 1544 : table_block_parallelscan_startblock_init(scan->rs_base.rs_rd,
244 : 1544 : scan->rs_parallelworkerdata,
245 : 1544 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
246 : :
247 : : /* may return InvalidBlockNumber if there are no more blocks */
248 : 3088 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
249 : 1544 : scan->rs_parallelworkerdata,
250 : 1544 : (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel);
251 : 1544 : scan->rs_inited = true;
252 : : }
253 : : else
254 : : {
255 : 98626 : scan->rs_prefetch_block = table_block_parallelscan_nextpage(scan->rs_base.rs_rd,
256 : 98626 : scan->rs_parallelworkerdata, (ParallelBlockTableScanDesc)
257 : 98626 : scan->rs_base.rs_parallel);
258 : : }
259 : :
260 : 100170 : return scan->rs_prefetch_block;
261 : : }
262 : :
263 : : /*
264 : : * Streaming read API callback for serial sequential and TID range scans.
265 : : * Returns the next block the caller wants from the read stream or
266 : : * InvalidBlockNumber when done.
267 : : */
268 : : static BlockNumber
269 : 3191371 : heap_scan_stream_read_next_serial(ReadStream *stream,
270 : : void *callback_private_data,
271 : : void *per_buffer_data)
272 : : {
273 : 3191371 : HeapScanDesc scan = (HeapScanDesc) callback_private_data;
274 : :
275 [ + + ]: 3191371 : if (unlikely(!scan->rs_inited))
276 : : {
277 : 799536 : scan->rs_prefetch_block = heapgettup_initial_block(scan, scan->rs_dir);
278 : 799536 : scan->rs_inited = true;
279 : : }
280 : : else
281 : 2391835 : scan->rs_prefetch_block = heapgettup_advance_block(scan,
282 : : scan->rs_prefetch_block,
283 : : scan->rs_dir);
284 : :
285 : 3191371 : return scan->rs_prefetch_block;
286 : : }
287 : :
288 : : /* ----------------
289 : : * initscan - scan code common to heap_beginscan and heap_rescan
290 : : * ----------------
291 : : */
292 : : static void
3186 tgl@sss.pgh.pa.us 293 :CBC 818193 : initscan(HeapScanDesc scan, ScanKey key, bool keep_startblock)
294 : : {
1861 andres@anarazel.de 295 : 818193 : ParallelBlockTableScanDesc bpscan = NULL;
296 : : bool allow_strat;
297 : : bool allow_sync;
298 : :
299 : : /*
300 : : * Determine the number of blocks we have to scan.
301 : : *
302 : : * It is sufficient to do this once at scan start, since any tuples added
303 : : * while the scan is in progress will be invisible to my snapshot anyway.
304 : : * (That is not true when using a non-MVCC snapshot. However, we couldn't
305 : : * guarantee to return tuples added after scan start anyway, since they
306 : : * might go into pages we already scanned. To guarantee consistent
307 : : * results for a non-MVCC snapshot, the caller must hold some higher-level
308 : : * lock that ensures the interesting tuple(s) won't change.)
309 : : */
310 [ + + ]: 818193 : if (scan->rs_base.rs_parallel != NULL)
311 : : {
312 : 1983 : bpscan = (ParallelBlockTableScanDesc) scan->rs_base.rs_parallel;
313 : 1983 : scan->rs_nblocks = bpscan->phs_nblocks;
314 : : }
315 : : else
316 : 816210 : scan->rs_nblocks = RelationGetNumberOfBlocks(scan->rs_base.rs_rd);
317 : :
318 : : /*
319 : : * If the table is large relative to NBuffers, use a bulk-read access
320 : : * strategy and enable synchronized scanning (see syncscan.c). Although
321 : : * the thresholds for these features could be different, we make them the
322 : : * same so that there are only two behaviors to tune rather than four.
323 : : * (However, some callers need to be able to disable one or both of these
324 : : * behaviors, independently of the size of the table; also there is a GUC
325 : : * variable that can disable synchronized scanning.)
326 : : *
327 : : * Note that table_block_parallelscan_initialize has a very similar test;
328 : : * if you change this, consider changing that one, too.
329 : : */
330 [ + + ]: 818191 : if (!RelationUsesLocalBuffers(scan->rs_base.rs_rd) &&
6154 tgl@sss.pgh.pa.us 331 [ + + ]: 812312 : scan->rs_nblocks > NBuffers / 4)
332 : : {
1792 andres@anarazel.de 333 : 10865 : allow_strat = (scan->rs_base.rs_flags & SO_ALLOW_STRAT) != 0;
334 : 10865 : allow_sync = (scan->rs_base.rs_flags & SO_ALLOW_SYNC) != 0;
335 : : }
336 : : else
5935 tgl@sss.pgh.pa.us 337 : 807326 : allow_strat = allow_sync = false;
338 : :
339 [ + + ]: 818191 : if (allow_strat)
340 : : {
341 : : /* During a rescan, keep the previous strategy object. */
6164 342 [ + + ]: 9617 : if (scan->rs_strategy == NULL)
343 : 9510 : scan->rs_strategy = GetAccessStrategy(BAS_BULKREAD);
344 : : }
345 : : else
346 : : {
347 [ - + ]: 808574 : if (scan->rs_strategy != NULL)
6164 tgl@sss.pgh.pa.us 348 :UBC 0 : FreeAccessStrategy(scan->rs_strategy);
6164 tgl@sss.pgh.pa.us 349 :CBC 808574 : scan->rs_strategy = NULL;
350 : : }
351 : :
1861 andres@anarazel.de 352 [ + + ]: 818191 : if (scan->rs_base.rs_parallel != NULL)
353 : : {
354 : : /* For parallel scan, believe whatever ParallelTableScanDesc says. */
1792 355 [ + + ]: 1983 : if (scan->rs_base.rs_parallel->phs_syncscan)
356 : 2 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
357 : : else
358 : 1981 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
359 : : }
3103 rhaas@postgresql.org 360 [ + + ]: 816208 : else if (keep_startblock)
361 : : {
362 : : /*
363 : : * When rescanning, we want to keep the previous startblock setting,
364 : : * so that rewinding a cursor doesn't generate surprising results.
365 : : * Reset the active syncscan setting, though.
366 : : */
1792 andres@anarazel.de 367 [ + + + + ]: 499442 : if (allow_sync && synchronize_seqscans)
1792 andres@anarazel.de 368 :GBC 20 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
369 : : else
1792 andres@anarazel.de 370 :CBC 499422 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
371 : : }
5422 tgl@sss.pgh.pa.us 372 [ + + + + ]: 316766 : else if (allow_sync && synchronize_seqscans)
373 : : {
1792 andres@anarazel.de 374 : 58 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
1861 375 : 58 : scan->rs_startblock = ss_get_location(scan->rs_base.rs_rd, scan->rs_nblocks);
376 : : }
377 : : else
378 : : {
1792 379 : 316708 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
6155 tgl@sss.pgh.pa.us 380 : 316708 : scan->rs_startblock = 0;
381 : : }
382 : :
3446 alvherre@alvh.no-ip. 383 : 818191 : scan->rs_numblocks = InvalidBlockNumber;
6714 tgl@sss.pgh.pa.us 384 : 818191 : scan->rs_inited = false;
8345 385 : 818191 : scan->rs_ctup.t_data = NULL;
6714 386 : 818191 : ItemPointerSetInvalid(&scan->rs_ctup.t_self);
8345 387 : 818191 : scan->rs_cbuf = InvalidBuffer;
6714 388 : 818191 : scan->rs_cblock = InvalidBlockNumber;
389 : :
390 : : /*
391 : : * Initialize to ForwardScanDirection because it is most common and
392 : : * because heap scans go forward before going backward (e.g. CURSORs).
393 : : */
6 tmunro@postgresql.or 394 :GNC 818191 : scan->rs_dir = ForwardScanDirection;
395 : 818191 : scan->rs_prefetch_block = InvalidBlockNumber;
396 : :
397 : : /* page-at-a-time fields are always invalid when not rs_inited */
398 : :
399 : : /*
400 : : * copy the scan key, if appropriate
401 : : */
773 tgl@sss.pgh.pa.us 402 [ + + + + ]:CBC 818191 : if (key != NULL && scan->rs_base.rs_nkeys > 0)
1861 andres@anarazel.de 403 : 181972 : memcpy(scan->rs_base.rs_key, key, scan->rs_base.rs_nkeys * sizeof(ScanKeyData));
404 : :
405 : : /*
406 : : * Currently, we only have a stats counter for sequential heap scans (but
407 : : * e.g for bitmap scans the underlying bitmap index scans will be counted,
408 : : * and for sample scans we update stats for tuple fetches).
409 : : */
1792 410 [ + + ]: 818191 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN)
1861 411 [ + + + + : 801548 : pgstat_count_heap_scan(scan->rs_base.rs_rd);
+ + ]
10141 scrappy@hub.org 412 : 818191 : }
413 : :
414 : : /*
415 : : * heap_setscanlimits - restrict range of a heapscan
416 : : *
417 : : * startBlk is the page to start at
418 : : * numBlks is number of pages to scan (InvalidBlockNumber means "all")
419 : : */
420 : : void
1861 andres@anarazel.de 421 : 1878 : heap_setscanlimits(TableScanDesc sscan, BlockNumber startBlk, BlockNumber numBlks)
422 : : {
423 : 1878 : HeapScanDesc scan = (HeapScanDesc) sscan;
424 : :
3190 tgl@sss.pgh.pa.us 425 [ - + ]: 1878 : Assert(!scan->rs_inited); /* else too late to change */
426 : : /* else rs_startblock is significant */
1792 andres@anarazel.de 427 [ - + ]: 1878 : Assert(!(scan->rs_base.rs_flags & SO_ALLOW_SYNC));
428 : :
429 : : /* Check startBlk is valid (but allow case of zero blocks...) */
3190 tgl@sss.pgh.pa.us 430 [ + + - + ]: 1878 : Assert(startBlk == 0 || startBlk < scan->rs_nblocks);
431 : :
3446 alvherre@alvh.no-ip. 432 : 1878 : scan->rs_startblock = startBlk;
433 : 1878 : scan->rs_numblocks = numBlks;
434 : 1878 : }
435 : :
436 : : /*
437 : : * Per-tuple loop for heap_prepare_pagescan(). Pulled out so it can be called
438 : : * multiple times, with constant arguments for all_visible,
439 : : * check_serializable.
440 : : */
441 : : pg_attribute_always_inline
442 : : static int
7 andres@anarazel.de 443 :GNC 1966403 : page_collect_tuples(HeapScanDesc scan, Snapshot snapshot,
444 : : Page page, Buffer buffer,
445 : : BlockNumber block, int lines,
446 : : bool all_visible, bool check_serializable)
447 : : {
8 448 : 1966403 : int ntup = 0;
449 : : OffsetNumber lineoff;
450 : :
451 [ + + ]: 101458717 : for (lineoff = FirstOffsetNumber; lineoff <= lines; lineoff++)
452 : : {
453 : 99492322 : ItemId lpp = PageGetItemId(page, lineoff);
454 : : HeapTupleData loctup;
455 : : bool valid;
456 : :
457 [ + + ]: 99492322 : if (!ItemIdIsNormal(lpp))
458 : 14234086 : continue;
459 : :
460 : 85258236 : loctup.t_data = (HeapTupleHeader) PageGetItem(page, lpp);
461 : 85258236 : loctup.t_len = ItemIdGetLength(lpp);
462 : 85258236 : loctup.t_tableOid = RelationGetRelid(scan->rs_base.rs_rd);
463 : 85258236 : ItemPointerSet(&(loctup.t_self), block, lineoff);
464 : :
465 [ + + ]: 85258236 : if (all_visible)
466 : 33135767 : valid = true;
467 : : else
468 : 52122469 : valid = HeapTupleSatisfiesVisibility(&loctup, snapshot, buffer);
469 : :
470 [ + + ]: 85258236 : if (check_serializable)
471 : 1417 : HeapCheckForSerializableConflictOut(valid, scan->rs_base.rs_rd,
472 : : &loctup, buffer, snapshot);
473 : :
474 [ + + ]: 85258228 : if (valid)
475 : : {
476 : 80135615 : scan->rs_vistuples[ntup] = lineoff;
477 : 80135615 : ntup++;
478 : : }
479 : : }
480 : :
481 [ - + ]: 1966395 : Assert(ntup <= MaxHeapTuplesPerPage);
482 : :
483 : 1966395 : return ntup;
484 : : }
485 : :
486 : : /*
487 : : * heap_prepare_pagescan - Prepare current scan page to be scanned in pagemode
488 : : *
489 : : * Preparation currently consists of 1. prune the scan's rs_cbuf page, and 2.
490 : : * fill the rs_vistuples[] array with the OffsetNumbers of visible tuples.
491 : : */
492 : : void
10 drowley@postgresql.o 493 : 1966403 : heap_prepare_pagescan(TableScanDesc sscan)
494 : : {
1861 andres@anarazel.de 495 :CBC 1966403 : HeapScanDesc scan = (HeapScanDesc) sscan;
10 drowley@postgresql.o 496 :GNC 1966403 : Buffer buffer = scan->rs_cbuf;
497 : 1966403 : BlockNumber block = scan->rs_cblock;
498 : : Snapshot snapshot;
499 : : Page page;
500 : : int lines;
501 : : bool all_visible;
502 : : bool check_serializable;
503 : :
504 [ - + ]: 1966403 : Assert(BufferGetBlockNumber(buffer) == block);
505 : :
506 : : /* ensure we're not accidentally being used when not in pagemode */
507 [ - + ]: 1966403 : Assert(scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE);
1861 andres@anarazel.de 508 :CBC 1966403 : snapshot = scan->rs_base.rs_snapshot;
509 : :
510 : : /*
511 : : * Prune and repair fragmentation for the whole page, if possible.
512 : : */
513 : 1966403 : heap_page_prune_opt(scan->rs_base.rs_rd, buffer);
514 : :
515 : : /*
516 : : * We must hold share lock on the buffer content while examining tuple
517 : : * visibility. Afterwards, however, the tuples we have found to be
518 : : * visible are guaranteed good as long as we hold the buffer pin.
519 : : */
6714 tgl@sss.pgh.pa.us 520 : 1966403 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
521 : :
515 peter@eisentraut.org 522 : 1966403 : page = BufferGetPage(buffer);
523 : 1966403 : lines = PageGetMaxOffsetNumber(page);
524 : :
525 : : /*
526 : : * If the all-visible flag indicates that all tuples on the page are
527 : : * visible to everyone, we can skip the per-tuple visibility tests.
528 : : *
529 : : * Note: In hot standby, a tuple that's already visible to all
530 : : * transactions on the primary might still be invisible to a read-only
531 : : * transaction in the standby. We partly handle this problem by tracking
532 : : * the minimum xmin of visible tuples as the cut-off XID while marking a
533 : : * page all-visible on the primary and WAL log that along with the
534 : : * visibility map SET operation. In hot standby, we wait for (or abort)
535 : : * all transactions that can potentially may not see one or more tuples on
536 : : * the page. That's how index-only scans work fine in hot standby. A
537 : : * crucial difference between index-only scans and heap scans is that the
538 : : * index-only scan completely relies on the visibility map where as heap
539 : : * scan looks at the page-level PD_ALL_VISIBLE flag. We are not sure if
540 : : * the page-level flag can be trusted in the same way, because it might
541 : : * get propagated somehow without being explicitly WAL-logged, e.g. via a
542 : : * full page write. Until we can prove that beyond doubt, let's check each
543 : : * tuple for visibility the hard way.
544 : : */
545 [ + + + + ]: 1966403 : all_visible = PageIsAllVisible(page) && !snapshot->takenDuringRecovery;
546 : : check_serializable =
8 andres@anarazel.de 547 :GNC 1966403 : CheckForSerializableConflictOutNeeded(scan->rs_base.rs_rd, snapshot);
548 : :
549 : : /*
550 : : * We call page_collect_tuples() with constant arguments, to get the
551 : : * compiler to constant fold the constant arguments. Separate calls with
552 : : * constant arguments, rather than variables, are needed on several
553 : : * compilers to actually perform constant folding.
554 : : */
555 [ + + ]: 1966403 : if (likely(all_visible))
556 : : {
557 [ + - ]: 746346 : if (likely(!check_serializable))
7 558 : 746346 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
559 : : block, lines, true, false);
560 : : else
7 andres@anarazel.de 561 :UNC 0 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
562 : : block, lines, true, true);
563 : : }
564 : : else
565 : : {
8 andres@anarazel.de 566 [ + + ]:GNC 1220057 : if (likely(!check_serializable))
7 567 : 1219429 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
568 : : block, lines, false, false);
569 : : else
570 : 628 : scan->rs_ntuples = page_collect_tuples(scan, snapshot, page, buffer,
571 : : block, lines, false, true);
572 : : }
573 : :
6714 tgl@sss.pgh.pa.us 574 :CBC 1966395 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6714 tgl@sss.pgh.pa.us 575 :GNC 1966395 : }
576 : :
577 : : /*
578 : : * heap_fetch_next_buffer - read and pin the next block from MAIN_FORKNUM.
579 : : *
580 : : * Read the next block of the scan relation from the read stream and save it
581 : : * in the scan descriptor. It is already pinned.
582 : : */
583 : : static inline void
10 drowley@postgresql.o 584 : 2722165 : heap_fetch_next_buffer(HeapScanDesc scan, ScanDirection dir)
585 : : {
6 tmunro@postgresql.or 586 [ - + ]: 2722165 : Assert(scan->rs_read_stream);
587 : :
588 : : /* release previous scan buffer, if any */
10 drowley@postgresql.o 589 [ + + ]: 2722165 : if (BufferIsValid(scan->rs_cbuf))
590 : : {
591 : 1921085 : ReleaseBuffer(scan->rs_cbuf);
592 : 1921085 : scan->rs_cbuf = InvalidBuffer;
593 : : }
594 : :
595 : : /*
596 : : * Be sure to check for interrupts at least once per page. Checks at
597 : : * higher code levels won't be able to stop a seqscan that encounters many
598 : : * pages' worth of consecutive dead tuples.
599 : : */
600 [ + + ]: 2722165 : CHECK_FOR_INTERRUPTS();
601 : :
602 : : /*
603 : : * If the scan direction is changing, reset the prefetch block to the
604 : : * current block. Otherwise, we will incorrectly prefetch the blocks
605 : : * between the prefetch block and the current block again before
606 : : * prefetching blocks in the new, correct scan direction.
607 : : */
6 tmunro@postgresql.or 608 [ + + ]: 2722161 : if (unlikely(scan->rs_dir != dir))
609 : : {
610 : 76 : scan->rs_prefetch_block = scan->rs_cblock;
611 : 76 : read_stream_reset(scan->rs_read_stream);
612 : : }
613 : :
614 : 2722161 : scan->rs_dir = dir;
615 : :
616 : 2722161 : scan->rs_cbuf = read_stream_next_buffer(scan->rs_read_stream, NULL);
617 [ + + ]: 2722161 : if (BufferIsValid(scan->rs_cbuf))
618 : 2054330 : scan->rs_cblock = BufferGetBlockNumber(scan->rs_cbuf);
10 drowley@postgresql.o 619 :GIC 2722161 : }
620 : :
621 : : /*
622 : : * heapgettup_initial_block - return the first BlockNumber to scan
623 : : *
624 : : * Returns InvalidBlockNumber when there are no blocks to scan. This can
625 : : * occur with empty tables and in parallel scans when parallel workers get all
626 : : * of the pages before we can get a chance to get our first page.
627 : : */
628 : : static pg_noinline BlockNumber
437 drowley@postgresql.o 629 :CBC 799536 : heapgettup_initial_block(HeapScanDesc scan, ScanDirection dir)
630 : : {
631 [ - + ]: 799536 : Assert(!scan->rs_inited);
6 tmunro@postgresql.or 632 [ - + ]:GNC 799536 : Assert(scan->rs_base.rs_parallel == NULL);
633 : :
634 : : /* When there are no pages to scan, return InvalidBlockNumber */
437 drowley@postgresql.o 635 [ + + + + ]:CBC 799536 : if (scan->rs_nblocks == 0 || scan->rs_numblocks == 0)
636 : 380715 : return InvalidBlockNumber;
637 : :
638 [ + + ]: 418821 : if (ScanDirectionIsForward(dir))
639 : : {
6 tmunro@postgresql.or 640 :GNC 418790 : return scan->rs_startblock;
641 : : }
642 : : else
643 : : {
644 : : /*
645 : : * Disable reporting to syncscan logic in a backwards scan; it's not
646 : : * very likely anyone else is doing the same thing at the same time,
647 : : * and much more likely that we'll just bollix things for forward
648 : : * scanners.
649 : : */
437 drowley@postgresql.o 650 :CBC 31 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
651 : :
652 : : /*
653 : : * Start from last page of the scan. Ensure we take into account
654 : : * rs_numblocks if it's been adjusted by heap_setscanlimits().
655 : : */
656 [ + + ]: 31 : if (scan->rs_numblocks != InvalidBlockNumber)
657 : 3 : return (scan->rs_startblock + scan->rs_numblocks - 1) % scan->rs_nblocks;
658 : :
659 [ - + ]: 28 : if (scan->rs_startblock > 0)
437 drowley@postgresql.o 660 :UBC 0 : return scan->rs_startblock - 1;
661 : :
437 drowley@postgresql.o 662 :CBC 28 : return scan->rs_nblocks - 1;
663 : : }
664 : : }
665 : :
666 : :
667 : : /*
668 : : * heapgettup_start_page - helper function for heapgettup()
669 : : *
670 : : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
671 : : * to the number of tuples on this page. Also set *lineoff to the first
672 : : * offset to scan with forward scans getting the first offset and backward
673 : : * getting the final offset on the page.
674 : : */
675 : : static Page
436 676 : 92202 : heapgettup_start_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
677 : : OffsetNumber *lineoff)
678 : : {
679 : : Page page;
680 : :
681 [ - + ]: 92202 : Assert(scan->rs_inited);
682 [ - + ]: 92202 : Assert(BufferIsValid(scan->rs_cbuf));
683 : :
684 : : /* Caller is responsible for ensuring buffer is locked if needed */
685 : 92202 : page = BufferGetPage(scan->rs_cbuf);
686 : :
432 687 : 92202 : *linesleft = PageGetMaxOffsetNumber(page) - FirstOffsetNumber + 1;
688 : :
436 689 [ + - ]: 92202 : if (ScanDirectionIsForward(dir))
690 : 92202 : *lineoff = FirstOffsetNumber;
691 : : else
436 drowley@postgresql.o 692 :UBC 0 : *lineoff = (OffsetNumber) (*linesleft);
693 : :
694 : : /* lineoff now references the physically previous or next tid */
436 drowley@postgresql.o 695 :CBC 92202 : return page;
696 : : }
697 : :
698 : :
699 : : /*
700 : : * heapgettup_continue_page - helper function for heapgettup()
701 : : *
702 : : * Return the next page to scan based on the scan->rs_cbuf and set *linesleft
703 : : * to the number of tuples left to scan on this page. Also set *lineoff to
704 : : * the next offset to scan according to the ScanDirection in 'dir'.
705 : : */
706 : : static inline Page
707 : 7744809 : heapgettup_continue_page(HeapScanDesc scan, ScanDirection dir, int *linesleft,
708 : : OffsetNumber *lineoff)
709 : : {
710 : : Page page;
711 : :
712 [ - + ]: 7744809 : Assert(scan->rs_inited);
713 [ - + ]: 7744809 : Assert(BufferIsValid(scan->rs_cbuf));
714 : :
715 : : /* Caller is responsible for ensuring buffer is locked if needed */
716 : 7744809 : page = BufferGetPage(scan->rs_cbuf);
717 : :
718 [ + - ]: 7744809 : if (ScanDirectionIsForward(dir))
719 : : {
720 : 7744809 : *lineoff = OffsetNumberNext(scan->rs_coffset);
721 : 7744809 : *linesleft = PageGetMaxOffsetNumber(page) - (*lineoff) + 1;
722 : : }
723 : : else
724 : : {
725 : : /*
726 : : * The previous returned tuple may have been vacuumed since the
727 : : * previous scan when we use a non-MVCC snapshot, so we must
728 : : * re-establish the lineoff <= PageGetMaxOffsetNumber(page) invariant
729 : : */
436 drowley@postgresql.o 730 [ # # ]:UBC 0 : *lineoff = Min(PageGetMaxOffsetNumber(page), OffsetNumberPrev(scan->rs_coffset));
731 : 0 : *linesleft = *lineoff;
732 : : }
733 : :
734 : : /* lineoff now references the physically previous or next tid */
436 drowley@postgresql.o 735 :CBC 7744809 : return page;
736 : : }
737 : :
738 : : /*
739 : : * heapgettup_advance_block - helper for heap_fetch_next_buffer()
740 : : *
741 : : * Given the current block number, the scan direction, and various information
742 : : * contained in the scan descriptor, calculate the BlockNumber to scan next
743 : : * and return it. If there are no further blocks to scan, return
744 : : * InvalidBlockNumber to indicate this fact to the caller.
745 : : *
746 : : * This should not be called to determine the initial block number -- only for
747 : : * subsequent blocks.
748 : : *
749 : : * This also adjusts rs_numblocks when a limit has been imposed by
750 : : * heap_setscanlimits().
751 : : */
752 : : static inline BlockNumber
753 : 2391835 : heapgettup_advance_block(HeapScanDesc scan, BlockNumber block, ScanDirection dir)
754 : : {
6 tmunro@postgresql.or 755 [ - + ]:GNC 2391835 : Assert(scan->rs_base.rs_parallel == NULL);
756 : :
757 [ + + ]: 2391835 : if (likely(ScanDirectionIsForward(dir)))
758 : : {
759 : 2391777 : block++;
760 : :
761 : : /* wrap back to the start of the heap */
762 [ + + ]: 2391777 : if (block >= scan->rs_nblocks)
763 : 367070 : block = 0;
764 : :
765 : : /*
766 : : * Report our new scan position for synchronization purposes. We don't
767 : : * do that when moving backwards, however. That would just mess up any
768 : : * other forward-moving scanners.
769 : : *
770 : : * Note: we do this before checking for end of scan so that the final
771 : : * state of the position hint is back at the start of the rel. That's
772 : : * not strictly necessary, but otherwise when you run the same query
773 : : * multiple times the starting position would shift a little bit
774 : : * backwards on every invocation, which is confusing. We don't
775 : : * guarantee any specific ordering in general, though.
776 : : */
777 [ + + ]: 2391777 : if (scan->rs_base.rs_flags & SO_ALLOW_SYNC)
778 : 8698 : ss_report_location(scan->rs_base.rs_rd, block);
779 : :
780 : : /* we're done if we're back at where we started */
781 [ + + ]: 2391777 : if (block == scan->rs_startblock)
782 : 367029 : return InvalidBlockNumber;
783 : :
784 : : /* check if the limit imposed by heap_setscanlimits() is met */
785 [ + + ]: 2024748 : if (scan->rs_numblocks != InvalidBlockNumber)
786 : : {
787 [ + + ]: 1596 : if (--scan->rs_numblocks == 0)
6 tmunro@postgresql.or 788 :CBC 1532 : return InvalidBlockNumber;
789 : : }
790 : :
6 tmunro@postgresql.or 791 :GNC 2023216 : return block;
792 : : }
793 : : else
794 : : {
795 : : /* we're done if the last block is the start position */
436 drowley@postgresql.o 796 [ + - ]:CBC 58 : if (block == scan->rs_startblock)
797 : 58 : return InvalidBlockNumber;
798 : :
799 : : /* check if the limit imposed by heap_setscanlimits() is met */
436 drowley@postgresql.o 800 [ # # ]:UBC 0 : if (scan->rs_numblocks != InvalidBlockNumber)
801 : : {
802 [ # # ]: 0 : if (--scan->rs_numblocks == 0)
803 : 0 : return InvalidBlockNumber;
804 : : }
805 : :
806 : : /* wrap to the end of the heap when the last page was page 0 */
807 [ # # ]: 0 : if (block == 0)
808 : 0 : block = scan->rs_nblocks;
809 : :
810 : 0 : block--;
811 : :
812 : 0 : return block;
813 : : }
814 : : }
815 : :
816 : : /* ----------------
817 : : * heapgettup - fetch next heap tuple
818 : : *
819 : : * Initialize the scan if not already done; then advance to the next
820 : : * tuple as indicated by "dir"; return the next tuple in scan->rs_ctup,
821 : : * or set scan->rs_ctup.t_data = NULL if no more tuples.
822 : : *
823 : : * Note: the reason nkeys/key are passed separately, even though they are
824 : : * kept in the scan descriptor, is that the caller may not want us to check
825 : : * the scankeys.
826 : : *
827 : : * Note: when we fall off the end of the scan in either direction, we
828 : : * reset rs_inited. This means that a further request with the same
829 : : * scan direction will restart the scan, which is a bit odd, but a
830 : : * request with the opposite scan direction will start a fresh scan
831 : : * in the proper direction. The latter is required behavior for cursors,
832 : : * while the former case is generally undefined behavior in Postgres
833 : : * so we don't care too much.
834 : : * ----------------
835 : : */
836 : : static void
6714 tgl@sss.pgh.pa.us 837 :CBC 7764019 : heapgettup(HeapScanDesc scan,
838 : : ScanDirection dir,
839 : : int nkeys,
840 : : ScanKey key)
841 : : {
842 : 7764019 : HeapTuple tuple = &(scan->rs_ctup);
843 : : Page page;
844 : : OffsetNumber lineoff;
845 : : int linesleft;
846 : :
10 drowley@postgresql.o 847 [ + + ]:GNC 7764019 : if (likely(scan->rs_inited))
848 : : {
849 : : /* continue from previously returned page/tuple */
436 drowley@postgresql.o 850 :CBC 7744809 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
851 : 7744809 : page = heapgettup_continue_page(scan, dir, &linesleft, &lineoff);
432 852 : 7744809 : goto continue_page;
853 : : }
854 : :
855 : : /*
856 : : * advance the scan until we find a qualifying tuple or run out of stuff
857 : : * to scan
858 : : */
859 : : while (true)
860 : : {
10 drowley@postgresql.o 861 :GNC 111263 : heap_fetch_next_buffer(scan, dir);
862 : :
863 : : /* did we run out of blocks to scan? */
864 [ + + ]: 111263 : if (!BufferIsValid(scan->rs_cbuf))
865 : 19061 : break;
866 : :
867 [ - + ]: 92202 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
868 : :
432 drowley@postgresql.o 869 :CBC 92202 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_SHARE);
870 : 92202 : page = heapgettup_start_page(scan, dir, &linesleft, &lineoff);
871 : 7837011 : continue_page:
872 : :
873 : : /*
874 : : * Only continue scanning the page while we have lines left.
875 : : *
876 : : * Note that this protects us from accessing line pointers past
877 : : * PageGetMaxOffsetNumber(); both for forward scans when we resume the
878 : : * table scan, and for when we start scanning a new page.
879 : : */
880 [ + + ]: 7871455 : for (; linesleft > 0; linesleft--, lineoff += dir)
881 : : {
882 : : bool visible;
883 : 7779402 : ItemId lpp = PageGetItemId(page, lineoff);
884 : :
885 [ + + ]: 7779402 : if (!ItemIdIsNormal(lpp))
886 : 29261 : continue;
887 : :
888 : 7750141 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
889 : 7750141 : tuple->t_len = ItemIdGetLength(lpp);
10 drowley@postgresql.o 890 :GNC 7750141 : ItemPointerSet(&(tuple->t_self), scan->rs_cblock, lineoff);
891 : :
432 drowley@postgresql.o 892 :CBC 7750141 : visible = HeapTupleSatisfiesVisibility(tuple,
893 : : scan->rs_base.rs_snapshot,
894 : : scan->rs_cbuf);
895 : :
896 : 7750141 : HeapCheckForSerializableConflictOut(visible, scan->rs_base.rs_rd,
897 : : tuple, scan->rs_cbuf,
898 : : scan->rs_base.rs_snapshot);
899 : :
900 : : /* skip tuples not visible to this snapshot */
901 [ + + ]: 7750141 : if (!visible)
902 : 5183 : continue;
903 : :
904 : : /* skip any tuples that don't match the scan key */
905 [ - + ]: 7744958 : if (key != NULL &&
432 drowley@postgresql.o 906 [ # # ]:UBC 0 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
907 : : nkeys, key))
908 : 0 : continue;
909 : :
432 drowley@postgresql.o 910 :CBC 7744958 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
911 : 7744958 : scan->rs_coffset = lineoff;
912 : 7744958 : return;
913 : : }
914 : :
915 : : /*
916 : : * if we get here, it means we've exhausted the items on this page and
917 : : * it's time to move to the next.
918 : : */
6714 tgl@sss.pgh.pa.us 919 : 92053 : LockBuffer(scan->rs_cbuf, BUFFER_LOCK_UNLOCK);
920 : : }
921 : :
922 : : /* end of scan */
432 drowley@postgresql.o 923 [ - + ]: 19061 : if (BufferIsValid(scan->rs_cbuf))
432 drowley@postgresql.o 924 :LBC (3840) : ReleaseBuffer(scan->rs_cbuf);
925 : :
432 drowley@postgresql.o 926 :CBC 19061 : scan->rs_cbuf = InvalidBuffer;
927 : 19061 : scan->rs_cblock = InvalidBlockNumber;
6 tmunro@postgresql.or 928 :GNC 19061 : scan->rs_prefetch_block = InvalidBlockNumber;
432 drowley@postgresql.o 929 :CBC 19061 : tuple->t_data = NULL;
930 : 19061 : scan->rs_inited = false;
931 : : }
932 : :
933 : : /* ----------------
934 : : * heapgettup_pagemode - fetch next heap tuple in page-at-a-time mode
935 : : *
936 : : * Same API as heapgettup, but used in page-at-a-time mode
937 : : *
938 : : * The internal logic is much the same as heapgettup's too, but there are some
939 : : * differences: we do not take the buffer content lock (that only needs to
940 : : * happen inside heap_prepare_pagescan), and we iterate through just the
941 : : * tuples listed in rs_vistuples[] rather than all tuples on the page. Notice
942 : : * that lineindex is 0-based, where the corresponding loop variable lineoff in
943 : : * heapgettup is 1-based.
944 : : * ----------------
945 : : */
946 : : static void
6714 tgl@sss.pgh.pa.us 947 : 39152766 : heapgettup_pagemode(HeapScanDesc scan,
948 : : ScanDirection dir,
949 : : int nkeys,
950 : : ScanKey key)
951 : : {
952 : 39152766 : HeapTuple tuple = &(scan->rs_ctup);
953 : : Page page;
954 : : int lineindex;
955 : : int linesleft;
956 : :
10 drowley@postgresql.o 957 [ + + ]:GNC 39152766 : if (likely(scan->rs_inited))
958 : : {
959 : : /* continue from previously returned page/tuple */
436 drowley@postgresql.o 960 :CBC 38370896 : page = BufferGetPage(scan->rs_cbuf);
961 : :
962 : 38370896 : lineindex = scan->rs_cindex + dir;
963 [ + + ]: 38370896 : if (ScanDirectionIsForward(dir))
964 : 38370568 : linesleft = scan->rs_ntuples - lineindex;
965 : : else
966 : 328 : linesleft = scan->rs_cindex;
967 : : /* lineindex now references the next or previous visible tid */
968 : :
432 969 : 38370896 : goto continue_page;
970 : : }
971 : :
972 : : /*
973 : : * advance the scan until we find a qualifying tuple or run out of stuff
974 : : * to scan
975 : : */
976 : : while (true)
977 : : {
10 drowley@postgresql.o 978 :GNC 2610902 : heap_fetch_next_buffer(scan, dir);
979 : :
980 : : /* did we run out of blocks to scan? */
981 [ + + ]: 2610898 : if (!BufferIsValid(scan->rs_cbuf))
982 : 648770 : break;
983 : :
984 [ - + ]: 1962128 : Assert(BufferGetBlockNumber(scan->rs_cbuf) == scan->rs_cblock);
985 : :
986 : : /* prune the page and determine visible tuple offsets */
987 : 1962128 : heap_prepare_pagescan((TableScanDesc) scan);
432 drowley@postgresql.o 988 :CBC 1962120 : page = BufferGetPage(scan->rs_cbuf);
989 : 1962120 : linesleft = scan->rs_ntuples;
990 [ + + ]: 1962120 : lineindex = ScanDirectionIsForward(dir) ? 0 : linesleft - 1;
991 : :
992 : : /* lineindex now references the next or previous visible tid */
993 : 40333016 : continue_page:
994 : :
995 [ + + ]: 79325552 : for (; linesleft > 0; linesleft--, lineindex += dir)
996 : : {
997 : : ItemId lpp;
998 : : OffsetNumber lineoff;
999 : :
6714 tgl@sss.pgh.pa.us 1000 : 77496520 : lineoff = scan->rs_vistuples[lineindex];
515 peter@eisentraut.org 1001 : 77496520 : lpp = PageGetItemId(page, lineoff);
6059 tgl@sss.pgh.pa.us 1002 [ - + ]: 77496520 : Assert(ItemIdIsNormal(lpp));
1003 : :
515 peter@eisentraut.org 1004 : 77496520 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lpp);
6714 tgl@sss.pgh.pa.us 1005 : 77496520 : tuple->t_len = ItemIdGetLength(lpp);
10 drowley@postgresql.o 1006 :GNC 77496520 : ItemPointerSet(&(tuple->t_self), scan->rs_cblock, lineoff);
1007 : :
1008 : : /* skip any tuples that don't match the scan key */
432 drowley@postgresql.o 1009 [ + + ]:CBC 77496520 : if (key != NULL &&
1010 [ + + ]: 39293587 : !HeapKeyTest(tuple, RelationGetDescr(scan->rs_base.rs_rd),
1011 : : nkeys, key))
1012 : 38992536 : continue;
1013 : :
1014 : 38503984 : scan->rs_cindex = lineindex;
1015 : 38503984 : return;
1016 : : }
1017 : : }
1018 : :
1019 : : /* end of scan */
1020 [ - + ]: 648770 : if (BufferIsValid(scan->rs_cbuf))
432 drowley@postgresql.o 1021 :LBC (445653) : ReleaseBuffer(scan->rs_cbuf);
432 drowley@postgresql.o 1022 :CBC 648770 : scan->rs_cbuf = InvalidBuffer;
1023 : 648770 : scan->rs_cblock = InvalidBlockNumber;
6 tmunro@postgresql.or 1024 :GNC 648770 : scan->rs_prefetch_block = InvalidBlockNumber;
432 drowley@postgresql.o 1025 :CBC 648770 : tuple->t_data = NULL;
1026 : 648770 : scan->rs_inited = false;
1027 : : }
1028 : :
1029 : :
1030 : : /* ----------------------------------------------------------------
1031 : : * heap access method interface
1032 : : * ----------------------------------------------------------------
1033 : : */
1034 : :
1035 : :
1036 : : TableScanDesc
8000 tgl@sss.pgh.pa.us 1037 : 318697 : heap_beginscan(Relation relation, Snapshot snapshot,
1038 : : int nkeys, ScanKey key,
1039 : : ParallelTableScanDesc parallel_scan,
1040 : : uint32 flags)
1041 : : {
1042 : : HeapScanDesc scan;
1043 : :
1044 : : /*
1045 : : * increment relation ref count while scanning relation
1046 : : *
1047 : : * This is just to make really sure the relcache entry won't go away while
1048 : : * the scan has a pointer to it. Caller should be holding the rel open
1049 : : * anyway, so this is redundant in all normal scenarios...
1050 : : */
8558 1051 : 318697 : RelationIncrementReferenceCount(relation);
1052 : :
1053 : : /*
1054 : : * allocate and initialize scan descriptor
1055 : : */
9370 bruce@momjian.us 1056 : 318697 : scan = (HeapScanDesc) palloc(sizeof(HeapScanDescData));
1057 : :
1861 andres@anarazel.de 1058 : 318697 : scan->rs_base.rs_rd = relation;
1059 : 318697 : scan->rs_base.rs_snapshot = snapshot;
1060 : 318697 : scan->rs_base.rs_nkeys = nkeys;
1792 1061 : 318697 : scan->rs_base.rs_flags = flags;
1861 1062 : 318697 : scan->rs_base.rs_parallel = parallel_scan;
1792 1063 : 318697 : scan->rs_strategy = NULL; /* set in initscan */
7 tomas.vondra@postgre 1064 :GNC 318697 : scan->rs_vmbuffer = InvalidBuffer;
1065 : 318697 : scan->rs_empty_tuples_pending = 0;
1066 : :
1067 : : /*
1068 : : * Disable page-at-a-time mode if it's not a MVCC-safe snapshot.
1069 : : */
1792 andres@anarazel.de 1070 [ + + + + :CBC 318697 : if (!(snapshot && IsMVCCSnapshot(snapshot)))
+ + ]
1071 : 26031 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1072 : :
1073 : : /*
1074 : : * For seqscan and sample scans in a serializable transaction, acquire a
1075 : : * predicate lock on the entire relation. This is required not only to
1076 : : * lock all the matching tuples, but also to conflict with new insertions
1077 : : * into the table. In an indexscan, we take page locks on the index pages
1078 : : * covering the range specified in the scan qual, but in a heap scan there
1079 : : * is nothing more fine-grained to lock. A bitmap scan is a different
1080 : : * story, there we have already scanned the index and locked the index
1081 : : * pages covering the predicate. But in that case we still have to lock
1082 : : * any matching heap tuples. For sample scan we could optimize the locking
1083 : : * to be at least page-level granularity, but we'd need to add per-tuple
1084 : : * locking for that.
1085 : : */
1086 [ + + ]: 318697 : if (scan->rs_base.rs_flags & (SO_TYPE_SEQSCAN | SO_TYPE_SAMPLESCAN))
1087 : : {
1088 : : /*
1089 : : * Ensure a missing snapshot is noticed reliably, even if the
1090 : : * isolation mode means predicate locking isn't performed (and
1091 : : * therefore the snapshot isn't used here).
1092 : : */
1093 [ - + ]: 304168 : Assert(snapshot);
4673 heikki.linnakangas@i 1094 : 304168 : PredicateLockRelation(relation, snapshot);
1095 : : }
1096 : :
1097 : : /* we only need to set this up once */
6714 tgl@sss.pgh.pa.us 1098 : 318697 : scan->rs_ctup.t_tableOid = RelationGetRelid(relation);
1099 : :
1100 : : /*
1101 : : * Allocate memory to keep track of page allocation for parallel workers
1102 : : * when doing a parallel scan.
1103 : : */
1111 drowley@postgresql.o 1104 [ + + ]: 318697 : if (parallel_scan != NULL)
1105 : 1929 : scan->rs_parallelworkerdata = palloc(sizeof(ParallelBlockTableScanWorkerData));
1106 : : else
1107 : 316768 : scan->rs_parallelworkerdata = NULL;
1108 : :
1109 : : /*
1110 : : * we do this here instead of in initscan() because heap_rescan also calls
1111 : : * initscan() and we don't want to allocate memory again
1112 : : */
8000 tgl@sss.pgh.pa.us 1113 [ + + ]: 318697 : if (nkeys > 0)
1861 andres@anarazel.de 1114 : 181972 : scan->rs_base.rs_key = (ScanKey) palloc(sizeof(ScanKeyData) * nkeys);
1115 : : else
1116 : 136725 : scan->rs_base.rs_key = NULL;
1117 : :
5422 tgl@sss.pgh.pa.us 1118 : 318697 : initscan(scan, key, false);
1119 : :
6 tmunro@postgresql.or 1120 :GNC 318695 : scan->rs_read_stream = NULL;
1121 : :
1122 : : /*
1123 : : * Set up a read stream for sequential scans and TID range scans. This
1124 : : * should be done after initscan() because initscan() allocates the
1125 : : * BufferAccessStrategy object passed to the streaming read API.
1126 : : */
1127 [ + + ]: 318695 : if (scan->rs_base.rs_flags & SO_TYPE_SEQSCAN ||
1128 [ + + ]: 14602 : scan->rs_base.rs_flags & SO_TYPE_TIDRANGESCAN)
1129 : : {
1130 : : ReadStreamBlockNumberCB cb;
1131 : :
1132 [ + + ]: 304149 : if (scan->rs_base.rs_parallel)
1133 : 1929 : cb = heap_scan_stream_read_next_parallel;
1134 : : else
1135 : 302220 : cb = heap_scan_stream_read_next_serial;
1136 : :
1137 : 304149 : scan->rs_read_stream = read_stream_begin_relation(READ_STREAM_SEQUENTIAL,
1138 : : scan->rs_strategy,
1139 : : scan->rs_base.rs_rd,
1140 : : MAIN_FORKNUM,
1141 : : cb,
1142 : : scan,
1143 : : 0);
1144 : : }
1145 : :
1146 : :
1861 andres@anarazel.de 1147 :CBC 318695 : return (TableScanDesc) scan;
1148 : : }
1149 : :
1150 : : void
1151 : 499496 : heap_rescan(TableScanDesc sscan, ScanKey key, bool set_params,
1152 : : bool allow_strat, bool allow_sync, bool allow_pagemode)
1153 : : {
1154 : 499496 : HeapScanDesc scan = (HeapScanDesc) sscan;
1155 : :
1156 [ + + ]: 499496 : if (set_params)
1157 : : {
1792 1158 [ + - ]: 15 : if (allow_strat)
1159 : 15 : scan->rs_base.rs_flags |= SO_ALLOW_STRAT;
1160 : : else
1792 andres@anarazel.de 1161 :UBC 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_STRAT;
1162 : :
1792 andres@anarazel.de 1163 [ + + ]:CBC 15 : if (allow_sync)
1164 : 6 : scan->rs_base.rs_flags |= SO_ALLOW_SYNC;
1165 : : else
1166 : 9 : scan->rs_base.rs_flags &= ~SO_ALLOW_SYNC;
1167 : :
1168 [ + - + - ]: 15 : if (allow_pagemode && scan->rs_base.rs_snapshot &&
1169 [ - + - - ]: 15 : IsMVCCSnapshot(scan->rs_base.rs_snapshot))
1170 : 15 : scan->rs_base.rs_flags |= SO_ALLOW_PAGEMODE;
1171 : : else
1792 andres@anarazel.de 1172 :UBC 0 : scan->rs_base.rs_flags &= ~SO_ALLOW_PAGEMODE;
1173 : : }
1174 : :
1175 : : /*
1176 : : * unpin scan buffers
1177 : : */
8345 tgl@sss.pgh.pa.us 1178 [ + + ]:CBC 499496 : if (BufferIsValid(scan->rs_cbuf))
1179 : 2729 : ReleaseBuffer(scan->rs_cbuf);
1180 : :
7 tomas.vondra@postgre 1181 [ + + ]:GNC 499496 : if (BufferIsValid(scan->rs_vmbuffer))
1182 : : {
1183 : 27 : ReleaseBuffer(scan->rs_vmbuffer);
1184 : 27 : scan->rs_vmbuffer = InvalidBuffer;
1185 : : }
1186 : :
1187 [ - + ]: 499496 : Assert(scan->rs_empty_tuples_pending == 0);
1188 : :
1189 : : /*
1190 : : * The read stream is reset on rescan. This must be done before
1191 : : * initscan(), as some state referred to by read_stream_reset() is reset
1192 : : * in initscan().
1193 : : */
6 tmunro@postgresql.or 1194 [ + + ]: 499496 : if (scan->rs_read_stream)
1195 : 497488 : read_stream_reset(scan->rs_read_stream);
1196 : :
1197 : : /*
1198 : : * reinitialize scan descriptor
1199 : : */
5422 tgl@sss.pgh.pa.us 1200 :CBC 499496 : initscan(scan, key, true);
10141 scrappy@hub.org 1201 : 499496 : }
1202 : :
1203 : : void
1861 andres@anarazel.de 1204 : 317441 : heap_endscan(TableScanDesc sscan)
1205 : : {
1206 : 317441 : HeapScanDesc scan = (HeapScanDesc) sscan;
1207 : :
1208 : : /* Note: no locking manipulations needed */
1209 : :
1210 : : /*
1211 : : * unpin scan buffers
1212 : : */
8345 tgl@sss.pgh.pa.us 1213 [ + + ]: 317441 : if (BufferIsValid(scan->rs_cbuf))
1214 : 136934 : ReleaseBuffer(scan->rs_cbuf);
1215 : :
7 tomas.vondra@postgre 1216 [ + + ]:GNC 317441 : if (BufferIsValid(scan->rs_vmbuffer))
1217 : 133 : ReleaseBuffer(scan->rs_vmbuffer);
1218 : :
1219 [ - + ]: 317441 : Assert(scan->rs_empty_tuples_pending == 0);
1220 : :
1221 : : /*
1222 : : * Must free the read stream before freeing the BufferAccessStrategy.
1223 : : */
6 tmunro@postgresql.or 1224 [ + + ]: 317441 : if (scan->rs_read_stream)
1225 : 302984 : read_stream_end(scan->rs_read_stream);
1226 : :
1227 : : /*
1228 : : * decrement relation reference count and free scan descriptor storage
1229 : : */
1861 andres@anarazel.de 1230 :CBC 317441 : RelationDecrementReferenceCount(scan->rs_base.rs_rd);
1231 : :
1232 [ + + ]: 317441 : if (scan->rs_base.rs_key)
1233 : 181941 : pfree(scan->rs_base.rs_key);
1234 : :
6164 tgl@sss.pgh.pa.us 1235 [ + + ]: 317441 : if (scan->rs_strategy != NULL)
1236 : 9501 : FreeAccessStrategy(scan->rs_strategy);
1237 : :
1111 drowley@postgresql.o 1238 [ + + ]: 317441 : if (scan->rs_parallelworkerdata != NULL)
1239 : 1929 : pfree(scan->rs_parallelworkerdata);
1240 : :
1792 andres@anarazel.de 1241 [ + + ]: 317441 : if (scan->rs_base.rs_flags & SO_TEMP_SNAPSHOT)
1861 1242 : 44927 : UnregisterSnapshot(scan->rs_base.rs_snapshot);
1243 : :
8975 tgl@sss.pgh.pa.us 1244 : 317441 : pfree(scan);
10141 scrappy@hub.org 1245 : 317441 : }
1246 : :
1247 : : HeapTuple
1861 andres@anarazel.de 1248 : 8815957 : heap_getnext(TableScanDesc sscan, ScanDirection direction)
1249 : : {
1250 : 8815957 : HeapScanDesc scan = (HeapScanDesc) sscan;
1251 : :
1252 : : /*
1253 : : * This is still widely used directly, without going through table AM, so
1254 : : * add a safety check. It's possible we should, at a later point,
1255 : : * downgrade this to an assert. The reason for checking the AM routine,
1256 : : * rather than the AM oid, is that this allows to write regression tests
1257 : : * that create another AM reusing the heap handler.
1258 : : */
1259 [ - + ]: 8815957 : if (unlikely(sscan->rs_rd->rd_tableam != GetHeapamTableAmRoutine()))
1861 andres@anarazel.de 1260 [ # # ]:UBC 0 : ereport(ERROR,
1261 : : (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
1262 : : errmsg_internal("only heap AM is supported")));
1263 : :
1264 : : /*
1265 : : * We don't expect direct calls to heap_getnext with valid CheckXidAlive
1266 : : * for catalog or regular tables. See detailed comments in xact.c where
1267 : : * these variables are declared. Normally we have such a check at tableam
1268 : : * level API but this is called from many places so we need to ensure it
1269 : : * here.
1270 : : */
1345 akapila@postgresql.o 1271 [ - + - - :CBC 8815957 : if (unlikely(TransactionIdIsValid(CheckXidAlive) && !bsysscan))
- + ]
1345 akapila@postgresql.o 1272 [ # # ]:UBC 0 : elog(ERROR, "unexpected heap_getnext call during logical decoding");
1273 : :
1274 : : /* Note: no locking manipulations needed */
1275 : :
1792 andres@anarazel.de 1276 [ + + ]:CBC 8815957 : if (scan->rs_base.rs_flags & SO_ALLOW_PAGEMODE)
6627 neilc@samurai.com 1277 : 1547545 : heapgettup_pagemode(scan, direction,
1861 andres@anarazel.de 1278 : 1547545 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1279 : : else
1280 : 7268412 : heapgettup(scan, direction,
1281 : 7268412 : scan->rs_base.rs_nkeys, scan->rs_base.rs_key);
1282 : :
6714 tgl@sss.pgh.pa.us 1283 [ + + ]: 8815957 : if (scan->rs_ctup.t_data == NULL)
8000 1284 : 55682 : return NULL;
1285 : :
1286 : : /*
1287 : : * if we get here it means we have a new current scan tuple, so point to
1288 : : * the proper return buffer and return the tuple.
1289 : : */
1290 : :
1861 andres@anarazel.de 1291 [ - + - - : 8760275 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
+ - ]
1292 : :
1293 : 8760275 : return &scan->rs_ctup;
1294 : : }
1295 : :
1296 : : bool
1297 : 38097765 : heap_getnextslot(TableScanDesc sscan, ScanDirection direction, TupleTableSlot *slot)
1298 : : {
1299 : 38097765 : HeapScanDesc scan = (HeapScanDesc) sscan;
1300 : :
1301 : : /* Note: no locking manipulations needed */
1302 : :
1792 1303 [ + + ]: 38097765 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1304 : 37602158 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1305 : : else
1306 : 495607 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1307 : :
1861 1308 [ + + ]: 38097753 : if (scan->rs_ctup.t_data == NULL)
1309 : : {
1310 : 612102 : ExecClearTuple(slot);
1311 : 612102 : return false;
1312 : : }
1313 : :
1314 : : /*
1315 : : * if we get here it means we have a new current scan tuple, so point to
1316 : : * the proper return buffer and return the tuple.
1317 : : */
1318 : :
1319 [ + + - + : 37485651 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
+ + ]
1320 : :
1321 : 37485651 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot,
1322 : : scan->rs_cbuf);
1323 : 37485651 : return true;
1324 : : }
1325 : :
1326 : : void
1142 drowley@postgresql.o 1327 : 89 : heap_set_tidrange(TableScanDesc sscan, ItemPointer mintid,
1328 : : ItemPointer maxtid)
1329 : : {
1330 : 89 : HeapScanDesc scan = (HeapScanDesc) sscan;
1331 : : BlockNumber startBlk;
1332 : : BlockNumber numBlks;
1333 : : ItemPointerData highestItem;
1334 : : ItemPointerData lowestItem;
1335 : :
1336 : : /*
1337 : : * For relations without any pages, we can simply leave the TID range
1338 : : * unset. There will be no tuples to scan, therefore no tuples outside
1339 : : * the given TID range.
1340 : : */
1341 [ + + ]: 89 : if (scan->rs_nblocks == 0)
1342 : 24 : return;
1343 : :
1344 : : /*
1345 : : * Set up some ItemPointers which point to the first and last possible
1346 : : * tuples in the heap.
1347 : : */
1348 : 83 : ItemPointerSet(&highestItem, scan->rs_nblocks - 1, MaxOffsetNumber);
1349 : 83 : ItemPointerSet(&lowestItem, 0, FirstOffsetNumber);
1350 : :
1351 : : /*
1352 : : * If the given maximum TID is below the highest possible TID in the
1353 : : * relation, then restrict the range to that, otherwise we scan to the end
1354 : : * of the relation.
1355 : : */
1356 [ + + ]: 83 : if (ItemPointerCompare(maxtid, &highestItem) < 0)
1357 : 66 : ItemPointerCopy(maxtid, &highestItem);
1358 : :
1359 : : /*
1360 : : * If the given minimum TID is above the lowest possible TID in the
1361 : : * relation, then restrict the range to only scan for TIDs above that.
1362 : : */
1363 [ + + ]: 83 : if (ItemPointerCompare(mintid, &lowestItem) > 0)
1364 : 26 : ItemPointerCopy(mintid, &lowestItem);
1365 : :
1366 : : /*
1367 : : * Check for an empty range and protect from would be negative results
1368 : : * from the numBlks calculation below.
1369 : : */
1370 [ + + ]: 83 : if (ItemPointerCompare(&highestItem, &lowestItem) < 0)
1371 : : {
1372 : : /* Set an empty range of blocks to scan */
1373 : 18 : heap_setscanlimits(sscan, 0, 0);
1374 : 18 : return;
1375 : : }
1376 : :
1377 : : /*
1378 : : * Calculate the first block and the number of blocks we must scan. We
1379 : : * could be more aggressive here and perform some more validation to try
1380 : : * and further narrow the scope of blocks to scan by checking if the
1381 : : * lowestItem has an offset above MaxOffsetNumber. In this case, we could
1382 : : * advance startBlk by one. Likewise, if highestItem has an offset of 0
1383 : : * we could scan one fewer blocks. However, such an optimization does not
1384 : : * seem worth troubling over, currently.
1385 : : */
1386 : 65 : startBlk = ItemPointerGetBlockNumberNoCheck(&lowestItem);
1387 : :
1388 : 65 : numBlks = ItemPointerGetBlockNumberNoCheck(&highestItem) -
1389 : 65 : ItemPointerGetBlockNumberNoCheck(&lowestItem) + 1;
1390 : :
1391 : : /* Set the start block and number of blocks to scan */
1392 : 65 : heap_setscanlimits(sscan, startBlk, numBlks);
1393 : :
1394 : : /* Finally, set the TID range in sscan */
1395 : 65 : ItemPointerCopy(&lowestItem, &sscan->rs_mintid);
1396 : 65 : ItemPointerCopy(&highestItem, &sscan->rs_maxtid);
1397 : : }
1398 : :
1399 : : bool
1400 : 2970 : heap_getnextslot_tidrange(TableScanDesc sscan, ScanDirection direction,
1401 : : TupleTableSlot *slot)
1402 : : {
1403 : 2970 : HeapScanDesc scan = (HeapScanDesc) sscan;
1404 : 2970 : ItemPointer mintid = &sscan->rs_mintid;
1405 : 2970 : ItemPointer maxtid = &sscan->rs_maxtid;
1406 : :
1407 : : /* Note: no locking manipulations needed */
1408 : : for (;;)
1409 : : {
1410 [ + - ]: 3063 : if (sscan->rs_flags & SO_ALLOW_PAGEMODE)
1411 : 3063 : heapgettup_pagemode(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1412 : : else
1142 drowley@postgresql.o 1413 :UBC 0 : heapgettup(scan, direction, sscan->rs_nkeys, sscan->rs_key);
1414 : :
1142 drowley@postgresql.o 1415 [ + + ]:CBC 3063 : if (scan->rs_ctup.t_data == NULL)
1416 : : {
1417 : 47 : ExecClearTuple(slot);
1418 : 47 : return false;
1419 : : }
1420 : :
1421 : : /*
1422 : : * heap_set_tidrange will have used heap_setscanlimits to limit the
1423 : : * range of pages we scan to only ones that can contain the TID range
1424 : : * we're scanning for. Here we must filter out any tuples from these
1425 : : * pages that are outside of that range.
1426 : : */
1427 [ + + ]: 3016 : if (ItemPointerCompare(&scan->rs_ctup.t_self, mintid) < 0)
1428 : : {
1429 : 93 : ExecClearTuple(slot);
1430 : :
1431 : : /*
1432 : : * When scanning backwards, the TIDs will be in descending order.
1433 : : * Future tuples in this direction will be lower still, so we can
1434 : : * just return false to indicate there will be no more tuples.
1435 : : */
1436 [ - + ]: 93 : if (ScanDirectionIsBackward(direction))
1142 drowley@postgresql.o 1437 :UBC 0 : return false;
1438 : :
1142 drowley@postgresql.o 1439 :CBC 93 : continue;
1440 : : }
1441 : :
1442 : : /*
1443 : : * Likewise for the final page, we must filter out TIDs greater than
1444 : : * maxtid.
1445 : : */
1446 [ + + ]: 2923 : if (ItemPointerCompare(&scan->rs_ctup.t_self, maxtid) > 0)
1447 : : {
1448 : 36 : ExecClearTuple(slot);
1449 : :
1450 : : /*
1451 : : * When scanning forward, the TIDs will be in ascending order.
1452 : : * Future tuples in this direction will be higher still, so we can
1453 : : * just return false to indicate there will be no more tuples.
1454 : : */
1455 [ + - ]: 36 : if (ScanDirectionIsForward(direction))
1456 : 36 : return false;
1142 drowley@postgresql.o 1457 :UBC 0 : continue;
1458 : : }
1459 : :
1142 drowley@postgresql.o 1460 :CBC 2887 : break;
1461 : : }
1462 : :
1463 : : /*
1464 : : * if we get here it means we have a new current scan tuple, so point to
1465 : : * the proper return buffer and return the tuple.
1466 : : */
1467 [ - + - - : 2887 : pgstat_count_heap_getnext(scan->rs_base.rs_rd);
+ - ]
1468 : :
1469 : 2887 : ExecStoreBufferHeapTuple(&scan->rs_ctup, slot, scan->rs_cbuf);
1470 : 2887 : return true;
1471 : : }
1472 : :
1473 : : /*
1474 : : * heap_fetch - retrieve tuple with given tid
1475 : : *
1476 : : * On entry, tuple->t_self is the TID to fetch. We pin the buffer holding
1477 : : * the tuple, fill in the remaining fields of *tuple, and check the tuple
1478 : : * against the specified snapshot.
1479 : : *
1480 : : * If successful (tuple found and passes snapshot time qual), then *userbuf
1481 : : * is set to the buffer holding the tuple and true is returned. The caller
1482 : : * must unpin the buffer when done with the tuple.
1483 : : *
1484 : : * If the tuple is not found (ie, item number references a deleted slot),
1485 : : * then tuple->t_data is set to NULL, *userbuf is set to InvalidBuffer,
1486 : : * and false is returned.
1487 : : *
1488 : : * If the tuple is found but fails the time qual check, then the behavior
1489 : : * depends on the keep_buf parameter. If keep_buf is false, the results
1490 : : * are the same as for the tuple-not-found case. If keep_buf is true,
1491 : : * then tuple->t_data and *userbuf are returned as for the success case,
1492 : : * and again the caller must unpin the buffer; but false is returned.
1493 : : *
1494 : : * heap_fetch does not follow HOT chains: only the exact TID requested will
1495 : : * be fetched.
1496 : : *
1497 : : * It is somewhat inconsistent that we ereport() on invalid block number but
1498 : : * return false on invalid item number. There are a couple of reasons though.
1499 : : * One is that the caller can relatively easily check the block number for
1500 : : * validity, but cannot check the item number without reading the page
1501 : : * himself. Another is that when we are following a t_ctid link, we can be
1502 : : * reasonably confident that the page number is valid (since VACUUM shouldn't
1503 : : * truncate off the destination page without having killed the referencing
1504 : : * tuple first), but the item number might well not be good.
1505 : : */
1506 : : bool
10141 scrappy@hub.org 1507 : 312668 : heap_fetch(Relation relation,
1508 : : Snapshot snapshot,
1509 : : HeapTuple tuple,
1510 : : Buffer *userbuf,
1511 : : bool keep_buf)
1512 : : {
7996 tgl@sss.pgh.pa.us 1513 : 312668 : ItemPointer tid = &(tuple->t_self);
1514 : : ItemId lp;
1515 : : Buffer buffer;
1516 : : Page page;
1517 : : OffsetNumber offnum;
1518 : : bool valid;
1519 : :
1520 : : /*
1521 : : * Fetch and pin the appropriate page of the relation.
1522 : : */
5855 1523 : 312668 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
1524 : :
1525 : : /*
1526 : : * Need share lock on buffer to examine tuple commit status.
1527 : : */
9252 vadim4o@yahoo.com 1528 : 312668 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
2916 kgrittn@postgresql.o 1529 : 312668 : page = BufferGetPage(buffer);
1530 : :
1531 : : /*
1532 : : * We'd better check for out-of-range offnum in case of VACUUM since the
1533 : : * TID was obtained.
1534 : : */
9716 bruce@momjian.us 1535 : 312668 : offnum = ItemPointerGetOffsetNumber(tid);
5754 tgl@sss.pgh.pa.us 1536 [ + - - + ]: 312668 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1537 : : {
6958 tgl@sss.pgh.pa.us 1538 :UBC 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1849 andres@anarazel.de 1539 : 0 : ReleaseBuffer(buffer);
1540 : 0 : *userbuf = InvalidBuffer;
6958 tgl@sss.pgh.pa.us 1541 : 0 : tuple->t_data = NULL;
1542 : 0 : return false;
1543 : : }
1544 : :
1545 : : /*
1546 : : * get the item line pointer corresponding to the requested tid
1547 : : */
5754 tgl@sss.pgh.pa.us 1548 :CBC 312668 : lp = PageGetItemId(page, offnum);
1549 : :
1550 : : /*
1551 : : * Must check for deleted tuple.
1552 : : */
6059 1553 [ + + ]: 312668 : if (!ItemIdIsNormal(lp))
1554 : : {
8685 vadim4o@yahoo.com 1555 : 1857 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1849 andres@anarazel.de 1556 : 1857 : ReleaseBuffer(buffer);
1557 : 1857 : *userbuf = InvalidBuffer;
7996 tgl@sss.pgh.pa.us 1558 : 1857 : tuple->t_data = NULL;
1559 : 1857 : return false;
1560 : : }
1561 : :
1562 : : /*
1563 : : * fill in *tuple fields
1564 : : */
5754 1565 : 310811 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
9270 vadim4o@yahoo.com 1566 : 310811 : tuple->t_len = ItemIdGetLength(lp);
6812 tgl@sss.pgh.pa.us 1567 : 310811 : tuple->t_tableOid = RelationGetRelid(relation);
1568 : :
1569 : : /*
1570 : : * check tuple visibility, then release lock
1571 : : */
6714 1572 : 310811 : valid = HeapTupleSatisfiesVisibility(tuple, snapshot, buffer);
1573 : :
4815 heikki.linnakangas@i 1574 [ + + ]: 310811 : if (valid)
1538 tmunro@postgresql.or 1575 : 310753 : PredicateLockTID(relation, &(tuple->t_self), snapshot,
1576 [ + + ]: 310753 : HeapTupleHeaderGetXmin(tuple->t_data));
1577 : :
1578 : 310811 : HeapCheckForSerializableConflictOut(valid, relation, tuple, buffer, snapshot);
1579 : :
4790 heikki.linnakangas@i 1580 : 310811 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
1581 : :
7996 tgl@sss.pgh.pa.us 1582 [ + + ]: 310811 : if (valid)
1583 : : {
1584 : : /*
1585 : : * All checks passed, so return the tuple as valid. Caller is now
1586 : : * responsible for releasing the buffer.
1587 : : */
8969 1588 : 310753 : *userbuf = buffer;
1589 : :
7996 1590 : 310753 : return true;
1591 : : }
1592 : :
1593 : : /* Tuple failed time qual, but maybe caller wants to see it anyway. */
732 1594 [ + + ]: 58 : if (keep_buf)
1595 : 29 : *userbuf = buffer;
1596 : : else
1597 : : {
1598 : 29 : ReleaseBuffer(buffer);
1599 : 29 : *userbuf = InvalidBuffer;
1600 : 29 : tuple->t_data = NULL;
1601 : : }
1602 : :
7996 1603 : 58 : return false;
1604 : : }
1605 : :
1606 : : /*
1607 : : * heap_hot_search_buffer - search HOT chain for tuple satisfying snapshot
1608 : : *
1609 : : * On entry, *tid is the TID of a tuple (either a simple tuple, or the root
1610 : : * of a HOT chain), and buffer is the buffer holding this tuple. We search
1611 : : * for the first chain member satisfying the given snapshot. If one is
1612 : : * found, we update *tid to reference that tuple's offset number, and
1613 : : * return true. If no match, return false without modifying *tid.
1614 : : *
1615 : : * heapTuple is a caller-supplied buffer. When a match is found, we return
1616 : : * the tuple here, in addition to updating *tid. If no match is found, the
1617 : : * contents of this buffer on return are undefined.
1618 : : *
1619 : : * If all_dead is not NULL, we check non-visible tuples to see if they are
1620 : : * globally dead; *all_dead is set true if all members of the HOT chain
1621 : : * are vacuumable, false if not.
1622 : : *
1623 : : * Unlike heap_fetch, the caller must already have pin and (at least) share
1624 : : * lock on the buffer; it is still pinned/locked at exit.
1625 : : */
1626 : : bool
4815 heikki.linnakangas@i 1627 : 19475875 : heap_hot_search_buffer(ItemPointer tid, Relation relation, Buffer buffer,
1628 : : Snapshot snapshot, HeapTuple heapTuple,
1629 : : bool *all_dead, bool first_call)
1630 : : {
515 peter@eisentraut.org 1631 : 19475875 : Page page = BufferGetPage(buffer);
6051 tgl@sss.pgh.pa.us 1632 : 19475875 : TransactionId prev_xmax = InvalidTransactionId;
1633 : : BlockNumber blkno;
1634 : : OffsetNumber offnum;
1635 : : bool at_chain_start;
1636 : : bool valid;
1637 : : bool skip;
1341 andres@anarazel.de 1638 : 19475875 : GlobalVisState *vistest = NULL;
1639 : :
1640 : : /* If this is not the first call, previous call returned a (live!) tuple */
6051 tgl@sss.pgh.pa.us 1641 [ + + ]: 19475875 : if (all_dead)
4675 rhaas@postgresql.org 1642 : 16627156 : *all_dead = first_call;
1643 : :
1712 heikki.linnakangas@i 1644 : 19475875 : blkno = ItemPointerGetBlockNumber(tid);
6051 tgl@sss.pgh.pa.us 1645 : 19475875 : offnum = ItemPointerGetOffsetNumber(tid);
4675 rhaas@postgresql.org 1646 : 19475875 : at_chain_start = first_call;
1647 : 19475875 : skip = !first_call;
1648 : :
1649 : : /* XXX: we should assert that a snapshot is pushed or registered */
1341 andres@anarazel.de 1650 [ - + ]: 19475875 : Assert(TransactionIdIsValid(RecentXmin));
1712 heikki.linnakangas@i 1651 [ + - ]: 19475875 : Assert(BufferGetBlockNumber(buffer) == blkno);
1652 : :
1653 : : /* Scan through possible multiple members of HOT-chain */
1654 : : for (;;)
6051 tgl@sss.pgh.pa.us 1655 : 873252 : {
1656 : : ItemId lp;
1657 : :
1658 : : /* check for bogus TID */
515 peter@eisentraut.org 1659 [ + - + - ]: 20349127 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1660 : : break;
1661 : :
1662 : 20349127 : lp = PageGetItemId(page, offnum);
1663 : :
1664 : : /* check for unused, dead, or redirected items */
6051 tgl@sss.pgh.pa.us 1665 [ + + ]: 20349127 : if (!ItemIdIsNormal(lp))
1666 : : {
1667 : : /* We should only see a redirect at start of chain */
1668 [ + + + - ]: 713826 : if (ItemIdIsRedirected(lp) && at_chain_start)
1669 : : {
1670 : : /* Follow the redirect */
1671 : 397984 : offnum = ItemIdGetRedirect(lp);
1672 : 397984 : at_chain_start = false;
1673 : 397984 : continue;
1674 : : }
1675 : : /* else must be end of chain */
1676 : 315842 : break;
1677 : : }
1678 : :
1679 : : /*
1680 : : * Update heapTuple to point to the element of the HOT chain we're
1681 : : * currently investigating. Having t_self set correctly is important
1682 : : * because the SSI checks and the *Satisfies routine for historical
1683 : : * MVCC snapshots need the correct tid to decide about the visibility.
1684 : : */
515 peter@eisentraut.org 1685 : 19635301 : heapTuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
4675 rhaas@postgresql.org 1686 : 19635301 : heapTuple->t_len = ItemIdGetLength(lp);
3919 1687 : 19635301 : heapTuple->t_tableOid = RelationGetRelid(relation);
1712 heikki.linnakangas@i 1688 : 19635301 : ItemPointerSet(&heapTuple->t_self, blkno, offnum);
1689 : :
1690 : : /*
1691 : : * Shouldn't see a HEAP_ONLY tuple at chain start.
1692 : : */
4675 rhaas@postgresql.org 1693 [ + + - + ]: 19635301 : if (at_chain_start && HeapTupleIsHeapOnly(heapTuple))
6051 tgl@sss.pgh.pa.us 1694 :UBC 0 : break;
1695 : :
1696 : : /*
1697 : : * The xmin should match the previous xmax value, else chain is
1698 : : * broken.
1699 : : */
6051 tgl@sss.pgh.pa.us 1700 [ + + - + ]:CBC 20110569 : if (TransactionIdIsValid(prev_xmax) &&
2355 alvherre@alvh.no-ip. 1701 [ + - ]: 475268 : !TransactionIdEquals(prev_xmax,
1702 : : HeapTupleHeaderGetXmin(heapTuple->t_data)))
6051 tgl@sss.pgh.pa.us 1703 :UBC 0 : break;
1704 : :
1705 : : /*
1706 : : * When first_call is true (and thus, skip is initially false) we'll
1707 : : * return the first tuple we find. But on later passes, heapTuple
1708 : : * will initially be pointing to the tuple we returned last time.
1709 : : * Returning it again would be incorrect (and would loop forever), so
1710 : : * we skip it and return the next match we find.
1711 : : */
4675 rhaas@postgresql.org 1712 [ + + ]:CBC 19635301 : if (!skip)
1713 : : {
1714 : : /* If it's visible per the snapshot, we must return it */
1715 : 19563421 : valid = HeapTupleSatisfiesVisibility(heapTuple, snapshot, buffer);
1538 tmunro@postgresql.or 1716 : 19563421 : HeapCheckForSerializableConflictOut(valid, relation, heapTuple,
1717 : : buffer, snapshot);
1718 : :
4675 rhaas@postgresql.org 1719 [ + + ]: 19563416 : if (valid)
1720 : : {
1721 : 13048746 : ItemPointerSetOffsetNumber(tid, offnum);
1538 tmunro@postgresql.or 1722 : 13048746 : PredicateLockTID(relation, &heapTuple->t_self, snapshot,
1723 [ + + ]: 13048746 : HeapTupleHeaderGetXmin(heapTuple->t_data));
4675 rhaas@postgresql.org 1724 [ + + ]: 13048746 : if (all_dead)
1725 : 10451460 : *all_dead = false;
1726 : 13048746 : return true;
1727 : : }
1728 : : }
1729 : 6586550 : skip = false;
1730 : :
1731 : : /*
1732 : : * If we can't see it, maybe no one else can either. At caller
1733 : : * request, check whether all chain members are dead to all
1734 : : * transactions.
1735 : : *
1736 : : * Note: if you change the criterion here for what is "dead", fix the
1737 : : * planner's get_actual_variable_range() function to match.
1738 : : */
1341 andres@anarazel.de 1739 [ + + + + ]: 6586550 : if (all_dead && *all_dead)
1740 : : {
1741 [ + + ]: 6285392 : if (!vistest)
1742 : 6197614 : vistest = GlobalVisTestFor(relation);
1743 : :
1744 [ + + ]: 6285392 : if (!HeapTupleIsSurelyDead(heapTuple, vistest))
1745 : 5998055 : *all_dead = false;
1746 : : }
1747 : :
1748 : : /*
1749 : : * Check to see if HOT chain continues past this tuple; if so fetch
1750 : : * the next offnum and loop around.
1751 : : */
4675 rhaas@postgresql.org 1752 [ + + + - : 6586550 : if (HeapTupleIsHotUpdated(heapTuple))
+ + ]
1753 : : {
1754 [ - + ]: 475268 : Assert(ItemPointerGetBlockNumber(&heapTuple->t_data->t_ctid) ==
1755 : : blkno);
1756 : 475268 : offnum = ItemPointerGetOffsetNumber(&heapTuple->t_data->t_ctid);
6051 tgl@sss.pgh.pa.us 1757 : 475268 : at_chain_start = false;
4099 alvherre@alvh.no-ip. 1758 [ + - + + : 475268 : prev_xmax = HeapTupleHeaderGetUpdateXid(heapTuple->t_data);
+ - ]
1759 : : }
1760 : : else
1761 : : break; /* end of chain */
1762 : : }
1763 : :
4703 heikki.linnakangas@i 1764 : 6427124 : return false;
1765 : : }
1766 : :
1767 : : /*
1768 : : * heap_get_latest_tid - get the latest tid of a specified tuple
1769 : : *
1770 : : * Actually, this gets the latest version that is visible according to the
1771 : : * scan's snapshot. Create a scan using SnapshotDirty to get the very latest,
1772 : : * possibly uncommitted version.
1773 : : *
1774 : : * *tid is both an input and an output parameter: it is updated to
1775 : : * show the latest version of the row. Note that it will not be changed
1776 : : * if no version of the row passes the snapshot test.
1777 : : */
1778 : : void
1794 andres@anarazel.de 1779 : 147 : heap_get_latest_tid(TableScanDesc sscan,
1780 : : ItemPointer tid)
1781 : : {
1789 tgl@sss.pgh.pa.us 1782 : 147 : Relation relation = sscan->rs_rd;
1783 : 147 : Snapshot snapshot = sscan->rs_snapshot;
1784 : : ItemPointerData ctid;
1785 : : TransactionId priorXmax;
1786 : :
1787 : : /*
1788 : : * table_tuple_get_latest_tid() verified that the passed in tid is valid.
1789 : : * Assume that t_ctid links are valid however - there shouldn't be invalid
1790 : : * ones in the table.
1791 : : */
1794 andres@anarazel.de 1792 [ - + ]: 147 : Assert(ItemPointerIsValid(tid));
1793 : :
1794 : : /*
1795 : : * Loop to chase down t_ctid links. At top of loop, ctid is the tuple we
1796 : : * need to examine, and *tid is the TID we will return if ctid turns out
1797 : : * to be bogus.
1798 : : *
1799 : : * Note that we will loop until we reach the end of the t_ctid chain.
1800 : : * Depending on the snapshot passed, there might be at most one visible
1801 : : * version of the row, but we don't try to optimize for that.
1802 : : */
6812 tgl@sss.pgh.pa.us 1803 : 147 : ctid = *tid;
1804 : 147 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
1805 : : for (;;)
1806 : 45 : {
1807 : : Buffer buffer;
1808 : : Page page;
1809 : : OffsetNumber offnum;
1810 : : ItemId lp;
1811 : : HeapTupleData tp;
1812 : : bool valid;
1813 : :
1814 : : /*
1815 : : * Read, pin, and lock the page.
1816 : : */
1817 : 192 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&ctid));
1818 : 192 : LockBuffer(buffer, BUFFER_LOCK_SHARE);
2916 kgrittn@postgresql.o 1819 : 192 : page = BufferGetPage(buffer);
1820 : :
1821 : : /*
1822 : : * Check for bogus item number. This is not treated as an error
1823 : : * condition because it can happen while following a t_ctid link. We
1824 : : * just assume that the prior tid is OK and return it unchanged.
1825 : : */
6812 tgl@sss.pgh.pa.us 1826 : 192 : offnum = ItemPointerGetOffsetNumber(&ctid);
5754 1827 [ + - - + ]: 192 : if (offnum < FirstOffsetNumber || offnum > PageGetMaxOffsetNumber(page))
1828 : : {
6589 tgl@sss.pgh.pa.us 1829 :UBC 0 : UnlockReleaseBuffer(buffer);
6812 1830 : 0 : break;
1831 : : }
5754 tgl@sss.pgh.pa.us 1832 :CBC 192 : lp = PageGetItemId(page, offnum);
6059 1833 [ - + ]: 192 : if (!ItemIdIsNormal(lp))
1834 : : {
6589 tgl@sss.pgh.pa.us 1835 :UBC 0 : UnlockReleaseBuffer(buffer);
6812 1836 : 0 : break;
1837 : : }
1838 : :
1839 : : /* OK to access the tuple */
6812 tgl@sss.pgh.pa.us 1840 :CBC 192 : tp.t_self = ctid;
5754 1841 : 192 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
6812 1842 : 192 : tp.t_len = ItemIdGetLength(lp);
3919 rhaas@postgresql.org 1843 : 192 : tp.t_tableOid = RelationGetRelid(relation);
1844 : :
1845 : : /*
1846 : : * After following a t_ctid link, we might arrive at an unrelated
1847 : : * tuple. Check for XMIN match.
1848 : : */
6812 tgl@sss.pgh.pa.us 1849 [ + + - + ]: 237 : if (TransactionIdIsValid(priorXmax) &&
2355 alvherre@alvh.no-ip. 1850 [ + - ]: 45 : !TransactionIdEquals(priorXmax, HeapTupleHeaderGetXmin(tp.t_data)))
1851 : : {
6589 tgl@sss.pgh.pa.us 1852 :UBC 0 : UnlockReleaseBuffer(buffer);
6812 1853 : 0 : break;
1854 : : }
1855 : :
1856 : : /*
1857 : : * Check tuple visibility; if visible, set it as the new result
1858 : : * candidate.
1859 : : */
6714 tgl@sss.pgh.pa.us 1860 :CBC 192 : valid = HeapTupleSatisfiesVisibility(&tp, snapshot, buffer);
1538 tmunro@postgresql.or 1861 : 192 : HeapCheckForSerializableConflictOut(valid, relation, &tp, buffer, snapshot);
6812 tgl@sss.pgh.pa.us 1862 [ + + ]: 192 : if (valid)
1863 : 135 : *tid = ctid;
1864 : :
1865 : : /*
1866 : : * If there's a valid t_ctid link, follow it, else we're done.
1867 : : */
4099 alvherre@alvh.no-ip. 1868 [ + + + + ]: 273 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
1869 [ + - ]: 138 : HeapTupleHeaderIsOnlyLocked(tp.t_data) ||
2199 andres@anarazel.de 1870 [ + + ]: 114 : HeapTupleHeaderIndicatesMovedPartitions(tp.t_data) ||
6812 tgl@sss.pgh.pa.us 1871 : 57 : ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
1872 : : {
6589 1873 : 147 : UnlockReleaseBuffer(buffer);
6812 1874 : 147 : break;
1875 : : }
1876 : :
1877 : 45 : ctid = tp.t_data->t_ctid;
4099 alvherre@alvh.no-ip. 1878 [ + - - + : 45 : priorXmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
- - ]
6589 tgl@sss.pgh.pa.us 1879 : 45 : UnlockReleaseBuffer(buffer);
1880 : : } /* end of loop */
8952 inoue@tpf.co.jp 1881 : 147 : }
1882 : :
1883 : :
1884 : : /*
1885 : : * UpdateXmaxHintBits - update tuple hint bits after xmax transaction ends
1886 : : *
1887 : : * This is called after we have waited for the XMAX transaction to terminate.
1888 : : * If the transaction aborted, we guarantee the XMAX_INVALID hint bit will
1889 : : * be set on exit. If the transaction committed, we set the XMAX_COMMITTED
1890 : : * hint bit if possible --- but beware that that may not yet be possible,
1891 : : * if the transaction committed asynchronously.
1892 : : *
1893 : : * Note that if the transaction was a locker only, we set HEAP_XMAX_INVALID
1894 : : * even if it commits.
1895 : : *
1896 : : * Hence callers should look only at XMAX_INVALID.
1897 : : *
1898 : : * Note this is not allowed for tuples whose xmax is a multixact.
1899 : : */
1900 : : static void
6088 tgl@sss.pgh.pa.us 1901 : 172 : UpdateXmaxHintBits(HeapTupleHeader tuple, Buffer buffer, TransactionId xid)
1902 : : {
4099 alvherre@alvh.no-ip. 1903 [ - + ]: 172 : Assert(TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple), xid));
1904 [ - + ]: 172 : Assert(!(tuple->t_infomask & HEAP_XMAX_IS_MULTI));
1905 : :
6088 tgl@sss.pgh.pa.us 1906 [ + - ]: 172 : if (!(tuple->t_infomask & (HEAP_XMAX_COMMITTED | HEAP_XMAX_INVALID)))
1907 : : {
4099 alvherre@alvh.no-ip. 1908 [ + + + - : 314 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask) &&
+ + ]
1909 : 142 : TransactionIdDidCommit(xid))
6088 tgl@sss.pgh.pa.us 1910 : 116 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_COMMITTED,
1911 : : xid);
1912 : : else
1913 : 56 : HeapTupleSetHintBits(tuple, buffer, HEAP_XMAX_INVALID,
1914 : : InvalidTransactionId);
1915 : : }
1916 : 172 : }
1917 : :
1918 : :
1919 : : /*
1920 : : * GetBulkInsertState - prepare status object for a bulk insert
1921 : : */
1922 : : BulkInsertState
5638 1923 : 2309 : GetBulkInsertState(void)
1924 : : {
1925 : : BulkInsertState bistate;
1926 : :
1927 : 2309 : bistate = (BulkInsertState) palloc(sizeof(BulkInsertStateData));
1928 : 2309 : bistate->strategy = GetAccessStrategy(BAS_BULKWRITE);
1929 : 2309 : bistate->current_buf = InvalidBuffer;
374 andres@anarazel.de 1930 : 2309 : bistate->next_free = InvalidBlockNumber;
1931 : 2309 : bistate->last_free = InvalidBlockNumber;
244 1932 : 2309 : bistate->already_extended_by = 0;
5638 tgl@sss.pgh.pa.us 1933 : 2309 : return bistate;
1934 : : }
1935 : :
1936 : : /*
1937 : : * FreeBulkInsertState - clean up after finishing a bulk insert
1938 : : */
1939 : : void
1940 : 2194 : FreeBulkInsertState(BulkInsertState bistate)
1941 : : {
1942 [ + + ]: 2194 : if (bistate->current_buf != InvalidBuffer)
5421 bruce@momjian.us 1943 : 1742 : ReleaseBuffer(bistate->current_buf);
5638 tgl@sss.pgh.pa.us 1944 : 2194 : FreeAccessStrategy(bistate->strategy);
1945 : 2194 : pfree(bistate);
1946 : 2194 : }
1947 : :
1948 : : /*
1949 : : * ReleaseBulkInsertStatePin - release a buffer currently held in bistate
1950 : : */
1951 : : void
2637 rhaas@postgresql.org 1952 : 80756 : ReleaseBulkInsertStatePin(BulkInsertState bistate)
1953 : : {
1954 [ + + ]: 80756 : if (bistate->current_buf != InvalidBuffer)
1955 : 30021 : ReleaseBuffer(bistate->current_buf);
1956 : 80756 : bistate->current_buf = InvalidBuffer;
1957 : :
1958 : : /*
1959 : : * Despite the name, we also reset bulk relation extension state.
1960 : : * Otherwise we can end up erroring out due to looking for free space in
1961 : : * ->next_free of one partition, even though ->next_free was set when
1962 : : * extending another partition. It could obviously also be bad for
1963 : : * efficiency to look at existing blocks at offsets from another
1964 : : * partition, even if we don't error out.
1965 : : */
184 andres@anarazel.de 1966 : 80756 : bistate->next_free = InvalidBlockNumber;
1967 : 80756 : bistate->last_free = InvalidBlockNumber;
2637 rhaas@postgresql.org 1968 : 80756 : }
1969 : :
1970 : :
1971 : : /*
1972 : : * heap_insert - insert tuple into a heap
1973 : : *
1974 : : * The new tuple is stamped with current transaction ID and the specified
1975 : : * command ID.
1976 : : *
1977 : : * See table_tuple_insert for comments about most of the input flags, except
1978 : : * that this routine directly takes a tuple rather than a slot.
1979 : : *
1980 : : * There's corresponding HEAP_INSERT_ options to all the TABLE_INSERT_
1981 : : * options, and there additionally is HEAP_INSERT_SPECULATIVE which is used to
1982 : : * implement table_tuple_insert_speculative().
1983 : : *
1984 : : * On return the header fields of *tup are updated to match the stored tuple;
1985 : : * in particular tup->t_self receives the actual TID where the tuple was
1986 : : * stored. But note that any toasting of fields within the tuple data is NOT
1987 : : * reflected into *tup.
1988 : : */
1989 : : void
6873 tgl@sss.pgh.pa.us 1990 : 7654548 : heap_insert(Relation relation, HeapTuple tup, CommandId cid,
1991 : : int options, BulkInsertState bistate)
1992 : : {
7150 1993 : 7654548 : TransactionId xid = GetCurrentTransactionId();
1994 : : HeapTuple heaptup;
1995 : : Buffer buffer;
4681 rhaas@postgresql.org 1996 : 7654545 : Buffer vmbuffer = InvalidBuffer;
5611 heikki.linnakangas@i 1997 : 7654545 : bool all_visible_cleared = false;
1998 : :
1999 : : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
1070 tgl@sss.pgh.pa.us 2000 [ - + ]: 7654545 : Assert(HeapTupleHeaderGetNatts(tup->t_data) <=
2001 : : RelationGetNumberOfAttributes(relation));
2002 : :
2003 : : /*
2004 : : * Fill in tuple header fields and toast the tuple if necessary.
2005 : : *
2006 : : * Note: below this point, heaptup is the data we actually intend to store
2007 : : * into the relation; tup is the caller's original untoasted data.
2008 : : */
4540 heikki.linnakangas@i 2009 : 7654545 : heaptup = heap_prepare_insert(relation, tup, xid, cid, options);
2010 : :
2011 : : /*
2012 : : * Find buffer to insert this tuple into. If the page is all visible,
2013 : : * this will also pin the requisite visibility map page.
2014 : : */
3088 kgrittn@postgresql.o 2015 : 7654545 : buffer = RelationGetBufferForTuple(relation, heaptup->t_len,
2016 : : InvalidBuffer, options, bistate,
2017 : : &vmbuffer, NULL,
2018 : : 0);
2019 : :
2020 : : /*
2021 : : * We're about to do the actual insert -- but check for conflict first, to
2022 : : * avoid possibly having to roll back work we've just done.
2023 : : *
2024 : : * This is safe without a recheck as long as there is no possibility of
2025 : : * another process scanning the page between this check and the insert
2026 : : * being visible to the scan (i.e., an exclusive buffer content lock is
2027 : : * continuously held from this point until the tuple insert is visible).
2028 : : *
2029 : : * For a heap insert, we only need to check for table-level SSI locks. Our
2030 : : * new tuple can't possibly conflict with existing tuple locks, and heap
2031 : : * page locks are only consolidated versions of tuple locks; they do not
2032 : : * lock "gaps" as index page locks do. So we don't need to specify a
2033 : : * buffer when making the call, which makes for a faster check.
2034 : : */
1538 tmunro@postgresql.or 2035 : 7654545 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2036 : :
2037 : : /* NO EREPORT(ERROR) from here till changes are logged */
8493 tgl@sss.pgh.pa.us 2038 : 7654533 : START_CRIT_SECTION();
2039 : :
3264 andres@anarazel.de 2040 : 7654533 : RelationPutHeapTuple(relation, buffer, heaptup,
2041 : 7654533 : (options & HEAP_INSERT_SPECULATIVE) != 0);
2042 : :
1046 tomas.vondra@postgre 2043 [ + + ]: 7654533 : if (PageIsAllVisible(BufferGetPage(buffer)))
2044 : : {
5611 heikki.linnakangas@i 2045 : 6499 : all_visible_cleared = true;
2916 kgrittn@postgresql.o 2046 : 6499 : PageClearAllVisible(BufferGetPage(buffer));
4681 rhaas@postgresql.org 2047 : 6499 : visibilitymap_clear(relation,
2048 : 6499 : ItemPointerGetBlockNumber(&(heaptup->t_self)),
2049 : : vmbuffer, VISIBILITYMAP_VALID_BITS);
2050 : : }
2051 : :
2052 : : /*
2053 : : * XXX Should we set PageSetPrunable on this page ?
2054 : : *
2055 : : * The inserting transaction may eventually abort thus making this tuple
2056 : : * DEAD and hence available for pruning. Though we don't want to optimize
2057 : : * for aborts, if no other tuple in this page is UPDATEd/DELETEd, the
2058 : : * aborted tuple will never be pruned until next vacuum is triggered.
2059 : : *
2060 : : * If you do add PageSetPrunable here, add it in heap_xlog_insert too.
2061 : : */
2062 : :
6589 tgl@sss.pgh.pa.us 2063 : 7654533 : MarkBufferDirty(buffer);
2064 : :
2065 : : /* XLOG stuff */
1471 noah@leadboat.com 2066 [ + + + + : 7654533 : if (RelationNeedsWAL(relation))
+ + + + ]
2067 : : {
2068 : : xl_heap_insert xlrec;
2069 : : xl_heap_header xlhdr;
2070 : : XLogRecPtr recptr;
2916 kgrittn@postgresql.o 2071 : 6659531 : Page page = BufferGetPage(buffer);
8424 bruce@momjian.us 2072 : 6659531 : uint8 info = XLOG_HEAP_INSERT;
3433 heikki.linnakangas@i 2073 : 6659531 : int bufflags = 0;
2074 : :
2075 : : /*
2076 : : * If this is a catalog, we need to transmit combo CIDs to properly
2077 : : * decode, so log that as well.
2078 : : */
3778 rhaas@postgresql.org 2079 [ + + + - : 6659531 : if (RelationIsAccessibleInLogicalDecoding(relation))
- + - - -
- + + + +
- + - - +
+ ]
2080 : 2833 : log_heap_new_cid(relation, heaptup);
2081 : :
2082 : : /*
2083 : : * If this is the single and first tuple on page, we can reinit the
2084 : : * page instead of restoring the whole thing. Set flag, and hide
2085 : : * buffer references from XLogInsert.
2086 : : */
3433 heikki.linnakangas@i 2087 [ + + + + ]: 6740277 : if (ItemPointerGetOffsetNumber(&(heaptup->t_self)) == FirstOffsetNumber &&
2088 : 80746 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
2089 : : {
2090 : 79950 : info |= XLOG_HEAP_INIT_PAGE;
2091 : 79950 : bufflags |= REGBUF_WILL_INIT;
2092 : : }
2093 : :
2094 : 6659531 : xlrec.offnum = ItemPointerGetOffsetNumber(&heaptup->t_self);
3264 andres@anarazel.de 2095 : 6659531 : xlrec.flags = 0;
2096 [ + + ]: 6659531 : if (all_visible_cleared)
2097 : 6496 : xlrec.flags |= XLH_INSERT_ALL_VISIBLE_CLEARED;
2098 [ + + ]: 6659531 : if (options & HEAP_INSERT_SPECULATIVE)
2099 : 2007 : xlrec.flags |= XLH_INSERT_IS_SPECULATIVE;
3433 heikki.linnakangas@i 2100 [ - + ]: 6659531 : Assert(ItemPointerGetBlockNumber(&heaptup->t_self) == BufferGetBlockNumber(buffer));
2101 : :
2102 : : /*
2103 : : * For logical decoding, we need the tuple even if we're doing a full
2104 : : * page write, so make sure it's included even if we take a full-page
2105 : : * image. (XXX We could alternatively store a pointer into the FPW).
2106 : : */
2013 andres@anarazel.de 2107 [ + + + - : 6659531 : if (RelationIsLogicallyLogged(relation) &&
- + - - -
- + - +
+ ]
2108 [ + + ]: 244691 : !(options & HEAP_INSERT_NO_LOGICAL))
2109 : : {
3264 2110 : 244664 : xlrec.flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
3433 heikki.linnakangas@i 2111 : 244664 : bufflags |= REGBUF_KEEP_DATA;
2112 : :
1345 akapila@postgresql.o 2113 [ + + ]: 244664 : if (IsToastRelation(relation))
2114 : 1684 : xlrec.flags |= XLH_INSERT_ON_TOAST_RELATION;
2115 : : }
2116 : :
3433 heikki.linnakangas@i 2117 : 6659531 : XLogBeginInsert();
2118 : 6659531 : XLogRegisterData((char *) &xlrec, SizeOfHeapInsert);
2119 : :
2120 : 6659531 : xlhdr.t_infomask2 = heaptup->t_data->t_infomask2;
2121 : 6659531 : xlhdr.t_infomask = heaptup->t_data->t_infomask;
2122 : 6659531 : xlhdr.t_hoff = heaptup->t_data->t_hoff;
2123 : :
2124 : : /*
2125 : : * note we mark xlhdr as belonging to buffer; if XLogInsert decides to
2126 : : * write the whole page to the xlog, we don't need to store
2127 : : * xl_heap_header in the xlog.
2128 : : */
2129 : 6659531 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2130 : 6659531 : XLogRegisterBufData(0, (char *) &xlhdr, SizeOfHeapHeader);
2131 : : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
2132 : 6659531 : XLogRegisterBufData(0,
3340 tgl@sss.pgh.pa.us 2133 : 6659531 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2134 : 6659531 : heaptup->t_len - SizeofHeapTupleHeader);
2135 : :
2136 : : /* filtering by origin on a row level is much more efficient */
2670 andres@anarazel.de 2137 : 6659531 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2138 : :
3433 heikki.linnakangas@i 2139 : 6659531 : recptr = XLogInsert(RM_HEAP_ID, info);
2140 : :
8508 vadim4o@yahoo.com 2141 : 6659531 : PageSetLSN(page, recptr);
2142 : : }
2143 : :
8493 tgl@sss.pgh.pa.us 2144 [ - + ]: 7654533 : END_CRIT_SECTION();
2145 : :
6589 2146 : 7654533 : UnlockReleaseBuffer(buffer);
4681 rhaas@postgresql.org 2147 [ + + ]: 7654533 : if (vmbuffer != InvalidBuffer)
2148 : 6637 : ReleaseBuffer(vmbuffer);
2149 : :
2150 : : /*
2151 : : * If tuple is cachable, mark it for invalidation from the caches in case
2152 : : * we abort. Note it is OK to do this after releasing the buffer, because
2153 : : * the heaptup data structure is all in local memory, not in the shared
2154 : : * buffer.
2155 : : */
4625 tgl@sss.pgh.pa.us 2156 : 7654533 : CacheInvalidateHeapTuple(relation, heaptup, NULL);
2157 : :
2158 : : /* Note: speculative insertions are counted too, even if aborted later */
4540 heikki.linnakangas@i 2159 : 7654533 : pgstat_count_heap_insert(relation, 1);
2160 : :
2161 : : /*
2162 : : * If heaptup is a private copy, release it. Don't forget to copy t_self
2163 : : * back to the caller's image, too.
2164 : : */
6720 tgl@sss.pgh.pa.us 2165 [ + + ]: 7654533 : if (heaptup != tup)
2166 : : {
2167 : 15774 : tup->t_self = heaptup->t_self;
2168 : 15774 : heap_freetuple(heaptup);
2169 : : }
10141 scrappy@hub.org 2170 : 7654533 : }
2171 : :
2172 : : /*
2173 : : * Subroutine for heap_insert(). Prepares a tuple for insertion. This sets the
2174 : : * tuple header fields and toasts the tuple if necessary. Returns a toasted
2175 : : * version of the tuple if it was toasted, or the original tuple if not. Note
2176 : : * that in any case, the header fields are also set in the original tuple.
2177 : : */
2178 : : static HeapTuple
4540 heikki.linnakangas@i 2179 : 9107803 : heap_prepare_insert(Relation relation, HeapTuple tup, TransactionId xid,
2180 : : CommandId cid, int options)
2181 : : {
2182 : : /*
2183 : : * To allow parallel inserts, we need to ensure that they are safe to be
2184 : : * performed in workers. We have the infrastructure to allow parallel
2185 : : * inserts in general except for the cases where inserts generate a new
2186 : : * CommandId (eg. inserts into a table having a foreign key column).
2187 : : */
2383 rhaas@postgresql.org 2188 [ - + ]: 9107803 : if (IsParallelWorker())
3272 rhaas@postgresql.org 2189 [ # # ]:UBC 0 : ereport(ERROR,
2190 : : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2191 : : errmsg("cannot insert tuples in a parallel worker")));
2192 : :
4540 heikki.linnakangas@i 2193 :CBC 9107803 : tup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
2194 : 9107803 : tup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
2195 : 9107803 : tup->t_data->t_infomask |= HEAP_XMAX_INVALID;
3766 rhaas@postgresql.org 2196 : 9107803 : HeapTupleHeaderSetXmin(tup->t_data, xid);
4151 simon@2ndQuadrant.co 2197 [ + + ]: 9107803 : if (options & HEAP_INSERT_FROZEN)
3766 rhaas@postgresql.org 2198 [ - + ]: 100362 : HeapTupleHeaderSetXminFrozen(tup->t_data);
2199 : :
4540 heikki.linnakangas@i 2200 [ - + ]: 9107803 : HeapTupleHeaderSetCmin(tup->t_data, cid);
2489 tgl@sss.pgh.pa.us 2201 : 9107803 : HeapTupleHeaderSetXmax(tup->t_data, 0); /* for cleanliness */
4540 heikki.linnakangas@i 2202 : 9107803 : tup->t_tableOid = RelationGetRelid(relation);
2203 : :
2204 : : /*
2205 : : * If the new tuple is too big for storage or contains already toasted
2206 : : * out-of-line attributes from some other relation, invoke the toaster.
2207 : : */
4060 kgrittn@postgresql.o 2208 [ + + ]: 9107803 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
2209 [ + + ]: 27379 : relation->rd_rel->relkind != RELKIND_MATVIEW)
2210 : : {
2211 : : /* toast table entries should never be recursively toasted */
4540 heikki.linnakangas@i 2212 [ - + ]: 25717 : Assert(!HeapTupleHasExternal(tup));
2213 : 25717 : return tup;
2214 : : }
2215 [ + + + + ]: 9082086 : else if (HeapTupleHasExternal(tup) || tup->t_len > TOAST_TUPLE_THRESHOLD)
1654 rhaas@postgresql.org 2216 : 15815 : return heap_toast_insert_or_update(relation, tup, NULL, options);
2217 : : else
4540 heikki.linnakangas@i 2218 : 9066271 : return tup;
2219 : : }
2220 : :
2221 : : /*
2222 : : * Helper for heap_multi_insert() that computes the number of entire pages
2223 : : * that inserting the remaining heaptuples requires. Used to determine how
2224 : : * much the relation needs to be extended by.
2225 : : */
2226 : : static int
374 andres@anarazel.de 2227 : 308974 : heap_multi_insert_pages(HeapTuple *heaptuples, int done, int ntuples, Size saveFreeSpace)
2228 : : {
2229 : 308974 : size_t page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2230 : 308974 : int npages = 1;
2231 : :
2232 [ + + ]: 2437449 : for (int i = done; i < ntuples; i++)
2233 : : {
2234 : 2128475 : size_t tup_sz = sizeof(ItemIdData) + MAXALIGN(heaptuples[i]->t_len);
2235 : :
2236 [ + + ]: 2128475 : if (page_avail < tup_sz)
2237 : : {
2238 : 16235 : npages++;
2239 : 16235 : page_avail = BLCKSZ - SizeOfPageHeaderData - saveFreeSpace;
2240 : : }
2241 : 2128475 : page_avail -= tup_sz;
2242 : : }
2243 : :
2244 : 308974 : return npages;
2245 : : }
2246 : :
2247 : : /*
2248 : : * heap_multi_insert - insert multiple tuples into a heap
2249 : : *
2250 : : * This is like heap_insert(), but inserts multiple tuples in one operation.
2251 : : * That's faster than calling heap_insert() in a loop, because when multiple
2252 : : * tuples can be inserted on a single page, we can write just a single WAL
2253 : : * record covering all of them, and only need to lock/unlock the page once.
2254 : : *
2255 : : * Note: this leaks memory into the current memory context. You can create a
2256 : : * temporary context before calling this, if that's a problem.
2257 : : */
2258 : : void
1837 2259 : 302969 : heap_multi_insert(Relation relation, TupleTableSlot **slots, int ntuples,
2260 : : CommandId cid, int options, BulkInsertState bistate)
2261 : : {
4540 heikki.linnakangas@i 2262 : 302969 : TransactionId xid = GetCurrentTransactionId();
2263 : : HeapTuple *heaptuples;
2264 : : int i;
2265 : : int ndone;
2266 : : PGAlignedBlock scratch;
2267 : : Page page;
1183 tomas.vondra@postgre 2268 : 302969 : Buffer vmbuffer = InvalidBuffer;
2269 : : bool needwal;
2270 : : Size saveFreeSpace;
3778 rhaas@postgresql.org 2271 [ + + + - : 302969 : bool need_tuple_data = RelationIsLogicallyLogged(relation);
- + - - -
- + - +
+ ]
2272 [ + + + - : 302969 : bool need_cids = RelationIsAccessibleInLogicalDecoding(relation);
- + - - -
- + + - +
- - - - -
- ]
374 andres@anarazel.de 2273 : 302969 : bool starting_with_empty_page = false;
2274 : 302969 : int npages = 0;
2275 : 302969 : int npages_used = 0;
2276 : :
2277 : : /* currently not needed (thus unsupported) for heap_multi_insert() */
534 peter@eisentraut.org 2278 [ - + ]: 302969 : Assert(!(options & HEAP_INSERT_NO_LOGICAL));
2279 : :
1471 noah@leadboat.com 2280 [ + + + + : 302969 : needwal = RelationNeedsWAL(relation);
+ + + + ]
3 akorotkov@postgresql 2281 [ + + ]: 302969 : saveFreeSpace = RelationGetTargetPageFreeSpace(relation,
2282 : : HEAP_DEFAULT_FILLFACTOR);
2283 : :
2284 : : /* Toast and set header data in all the slots */
4540 heikki.linnakangas@i 2285 : 302969 : heaptuples = palloc(ntuples * sizeof(HeapTuple));
2286 [ + + ]: 1756227 : for (i = 0; i < ntuples; i++)
2287 : : {
2288 : : HeapTuple tuple;
2289 : :
1837 andres@anarazel.de 2290 : 1453258 : tuple = ExecFetchSlotHeapTuple(slots[i], true, NULL);
2291 : 1453258 : slots[i]->tts_tableOid = RelationGetRelid(relation);
2292 : 1453258 : tuple->t_tableOid = slots[i]->tts_tableOid;
2293 : 1453258 : heaptuples[i] = heap_prepare_insert(relation, tuple, xid, cid,
2294 : : options);
2295 : : }
2296 : :
2297 : : /*
2298 : : * We're about to do the actual inserts -- but check for conflict first,
2299 : : * to minimize the possibility of having to roll back work we've just
2300 : : * done.
2301 : : *
2302 : : * A check here does not definitively prevent a serialization anomaly;
2303 : : * that check MUST be done at least past the point of acquiring an
2304 : : * exclusive buffer content lock on every buffer that will be affected,
2305 : : * and MAY be done after all inserts are reflected in the buffers and
2306 : : * those locks are released; otherwise there is a race condition. Since
2307 : : * multiple buffers can be locked and unlocked in the loop below, and it
2308 : : * would not be feasible to identify and lock all of those buffers before
2309 : : * the loop, we must do a final check at the end.
2310 : : *
2311 : : * The check here could be omitted with no loss of correctness; it is
2312 : : * present strictly as an optimization.
2313 : : *
2314 : : * For heap inserts, we only need to check for table-level SSI locks. Our
2315 : : * new tuples can't possibly conflict with existing tuple locks, and heap
2316 : : * page locks are only consolidated versions of tuple locks; they do not
2317 : : * lock "gaps" as index page locks do. So we don't need to specify a
2318 : : * buffer when making the call, which makes for a faster check.
2319 : : */
1538 tmunro@postgresql.or 2320 : 302969 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2321 : :
4540 heikki.linnakangas@i 2322 : 302969 : ndone = 0;
2323 [ + + ]: 620358 : while (ndone < ntuples)
2324 : : {
2325 : : Buffer buffer;
2326 : 317389 : bool all_visible_cleared = false;
1183 tomas.vondra@postgre 2327 : 317389 : bool all_frozen_set = false;
2328 : : int nthispage;
2329 : :
3583 rhaas@postgresql.org 2330 [ + + ]: 317389 : CHECK_FOR_INTERRUPTS();
2331 : :
2332 : : /*
2333 : : * Compute number of pages needed to fit the to-be-inserted tuples in
2334 : : * the worst case. This will be used to determine how much to extend
2335 : : * the relation by in RelationGetBufferForTuple(), if needed. If we
2336 : : * filled a prior page from scratch, we can just update our last
2337 : : * computation, but if we started with a partially filled page,
2338 : : * recompute from scratch, the number of potentially required pages
2339 : : * can vary due to tuples needing to fit onto the page, page headers
2340 : : * etc.
2341 : : */
374 andres@anarazel.de 2342 [ + + + + ]: 317389 : if (ndone == 0 || !starting_with_empty_page)
2343 : : {
2344 : 308974 : npages = heap_multi_insert_pages(heaptuples, ndone, ntuples,
2345 : : saveFreeSpace);
2346 : 308974 : npages_used = 0;
2347 : : }
2348 : : else
2349 : 8415 : npages_used++;
2350 : :
2351 : : /*
2352 : : * Find buffer where at least the next tuple will fit. If the page is
2353 : : * all-visible, this will also pin the requisite visibility map page.
2354 : : *
2355 : : * Also pin visibility map page if COPY FREEZE inserts tuples into an
2356 : : * empty page. See all_frozen_set below.
2357 : : */
4540 heikki.linnakangas@i 2358 : 317389 : buffer = RelationGetBufferForTuple(relation, heaptuples[ndone]->t_len,
2359 : : InvalidBuffer, options, bistate,
2360 : : &vmbuffer, NULL,
2361 : : npages - npages_used);
2916 kgrittn@postgresql.o 2362 : 317389 : page = BufferGetPage(buffer);
2363 : :
1183 tomas.vondra@postgre 2364 : 317389 : starting_with_empty_page = PageGetMaxOffsetNumber(page) == 0;
2365 : :
2366 [ + + + + ]: 317389 : if (starting_with_empty_page && (options & HEAP_INSERT_FROZEN))
2367 : 1661 : all_frozen_set = true;
2368 : :
2369 : : /* NO EREPORT(ERROR) from here till changes are logged */
4540 heikki.linnakangas@i 2370 : 317389 : START_CRIT_SECTION();
2371 : :
2372 : : /*
2373 : : * RelationGetBufferForTuple has ensured that the first tuple fits.
2374 : : * Put that on the page, and then as many other tuples as fit.
2375 : : */
3264 andres@anarazel.de 2376 : 317389 : RelationPutHeapTuple(relation, buffer, heaptuples[ndone], false);
2377 : :
2378 : : /*
2379 : : * For logical decoding we need combo CIDs to properly decode the
2380 : : * catalog.
2381 : : */
1510 michael@paquier.xyz 2382 [ + + + + ]: 317389 : if (needwal && need_cids)
2383 : 4263 : log_heap_new_cid(relation, heaptuples[ndone]);
2384 : :
4141 heikki.linnakangas@i 2385 [ + + ]: 1453258 : for (nthispage = 1; ndone + nthispage < ntuples; nthispage++)
2386 : : {
4540 2387 : 1150289 : HeapTuple heaptup = heaptuples[ndone + nthispage];
2388 : :
4351 2389 [ + + ]: 1150289 : if (PageGetHeapFreeSpace(page) < MAXALIGN(heaptup->t_len) + saveFreeSpace)
4540 2390 : 14420 : break;
2391 : :
3264 andres@anarazel.de 2392 : 1135869 : RelationPutHeapTuple(relation, buffer, heaptup, false);
2393 : :
2394 : : /*
2395 : : * For logical decoding we need combo CIDs to properly decode the
2396 : : * catalog.
2397 : : */
3433 heikki.linnakangas@i 2398 [ + + + + ]: 1135869 : if (needwal && need_cids)
2399 : 4229 : log_heap_new_cid(relation, heaptup);
2400 : : }
2401 : :
2402 : : /*
2403 : : * If the page is all visible, need to clear that, unless we're only
2404 : : * going to add further frozen rows to it.
2405 : : *
2406 : : * If we're only adding already frozen rows to a previously empty
2407 : : * page, mark it as all-visible.
2408 : : */
1183 tomas.vondra@postgre 2409 [ + + + + ]: 317389 : if (PageIsAllVisible(page) && !(options & HEAP_INSERT_FROZEN))
2410 : : {
4329 rhaas@postgresql.org 2411 : 3514 : all_visible_cleared = true;
2412 : 3514 : PageClearAllVisible(page);
2413 : 3514 : visibilitymap_clear(relation,
2414 : : BufferGetBlockNumber(buffer),
2415 : : vmbuffer, VISIBILITYMAP_VALID_BITS);
2416 : : }
1183 tomas.vondra@postgre 2417 [ + + ]: 313875 : else if (all_frozen_set)
2418 : 1661 : PageSetAllVisible(page);
2419 : :
2420 : : /*
2421 : : * XXX Should we set PageSetPrunable on this page ? See heap_insert()
2422 : : */
2423 : :
4540 heikki.linnakangas@i 2424 : 317389 : MarkBufferDirty(buffer);
2425 : :
2426 : : /* XLOG stuff */
2427 [ + + ]: 317389 : if (needwal)
2428 : : {
2429 : : XLogRecPtr recptr;
2430 : : xl_heap_multi_insert *xlrec;
2431 : 313029 : uint8 info = XLOG_HEAP2_MULTI_INSERT;
2432 : : char *tupledata;
2433 : : int totaldatalen;
2052 tgl@sss.pgh.pa.us 2434 : 313029 : char *scratchptr = scratch.data;
2435 : : bool init;
3433 heikki.linnakangas@i 2436 : 313029 : int bufflags = 0;
2437 : :
2438 : : /*
2439 : : * If the page was previously empty, we can reinit the page
2440 : : * instead of restoring the whole thing.
2441 : : */
1183 tomas.vondra@postgre 2442 : 313029 : init = starting_with_empty_page;
2443 : :
2444 : : /* allocate xl_heap_multi_insert struct from the scratch area */
4540 heikki.linnakangas@i 2445 : 313029 : xlrec = (xl_heap_multi_insert *) scratchptr;
2446 : 313029 : scratchptr += SizeOfHeapMultiInsert;
2447 : :
2448 : : /*
2449 : : * Allocate offsets array. Unless we're reinitializing the page,
2450 : : * in that case the tuples are stored in order starting at
2451 : : * FirstOffsetNumber and we don't need to store the offsets
2452 : : * explicitly.
2453 : : */
2454 [ + + ]: 313029 : if (!init)
2455 : 301732 : scratchptr += nthispage * sizeof(OffsetNumber);
2456 : :
2457 : : /* the rest of the scratch space is used for tuple data */
2458 : 313029 : tupledata = scratchptr;
2459 : :
2460 : : /* check that the mutually exclusive flags are not both set */
1068 tgl@sss.pgh.pa.us 2461 [ + + - + ]: 313029 : Assert(!(all_visible_cleared && all_frozen_set));
2462 : :
1183 tomas.vondra@postgre 2463 : 313029 : xlrec->flags = 0;
2464 [ + + ]: 313029 : if (all_visible_cleared)
2465 : 3514 : xlrec->flags = XLH_INSERT_ALL_VISIBLE_CLEARED;
2466 [ + + ]: 313029 : if (all_frozen_set)
2467 : 13 : xlrec->flags = XLH_INSERT_ALL_FROZEN_SET;
2468 : :
4540 heikki.linnakangas@i 2469 : 313029 : xlrec->ntuples = nthispage;
2470 : :
2471 : : /*
2472 : : * Write out an xl_multi_insert_tuple and the tuple data itself
2473 : : * for each tuple.
2474 : : */
2475 [ + + ]: 1460895 : for (i = 0; i < nthispage; i++)
2476 : : {
2477 : 1147866 : HeapTuple heaptup = heaptuples[ndone + i];
2478 : : xl_multi_insert_tuple *tuphdr;
2479 : : int datalen;
2480 : :
2481 [ + + ]: 1147866 : if (!init)
2482 : 643078 : xlrec->offsets[i] = ItemPointerGetOffsetNumber(&heaptup->t_self);
2483 : : /* xl_multi_insert_tuple needs two-byte alignment. */
2484 : 1147866 : tuphdr = (xl_multi_insert_tuple *) SHORTALIGN(scratchptr);
2485 : 1147866 : scratchptr = ((char *) tuphdr) + SizeOfMultiInsertTuple;
2486 : :
2487 : 1147866 : tuphdr->t_infomask2 = heaptup->t_data->t_infomask2;
2488 : 1147866 : tuphdr->t_infomask = heaptup->t_data->t_infomask;
2489 : 1147866 : tuphdr->t_hoff = heaptup->t_data->t_hoff;
2490 : :
2491 : : /* write bitmap [+ padding] [+ oid] + data */
3340 tgl@sss.pgh.pa.us 2492 : 1147866 : datalen = heaptup->t_len - SizeofHeapTupleHeader;
4540 heikki.linnakangas@i 2493 : 1147866 : memcpy(scratchptr,
3340 tgl@sss.pgh.pa.us 2494 : 1147866 : (char *) heaptup->t_data + SizeofHeapTupleHeader,
2495 : : datalen);
4540 heikki.linnakangas@i 2496 : 1147866 : tuphdr->datalen = datalen;
2497 : 1147866 : scratchptr += datalen;
2498 : : }
2499 : 313029 : totaldatalen = scratchptr - tupledata;
2052 tgl@sss.pgh.pa.us 2500 [ - + ]: 313029 : Assert((scratchptr - scratch.data) < BLCKSZ);
2501 : :
3778 rhaas@postgresql.org 2502 [ + + ]: 313029 : if (need_tuple_data)
3264 andres@anarazel.de 2503 : 73 : xlrec->flags |= XLH_INSERT_CONTAINS_NEW_TUPLE;
2504 : :
2505 : : /*
2506 : : * Signal that this is the last xl_heap_multi_insert record
2507 : : * emitted by this call to heap_multi_insert(). Needed for logical
2508 : : * decoding so it knows when to cleanup temporary data.
2509 : : */
3433 heikki.linnakangas@i 2510 [ + + ]: 313029 : if (ndone + nthispage == ntuples)
3264 andres@anarazel.de 2511 : 302460 : xlrec->flags |= XLH_INSERT_LAST_IN_MULTI;
2512 : :
4540 heikki.linnakangas@i 2513 [ + + ]: 313029 : if (init)
2514 : : {
2515 : 11297 : info |= XLOG_HEAP_INIT_PAGE;
3433 2516 : 11297 : bufflags |= REGBUF_WILL_INIT;
2517 : : }
2518 : :
2519 : : /*
2520 : : * If we're doing logical decoding, include the new tuple data
2521 : : * even if we take a full-page image of the page.
2522 : : */
2523 [ + + ]: 313029 : if (need_tuple_data)
2524 : 73 : bufflags |= REGBUF_KEEP_DATA;
2525 : :
2526 : 313029 : XLogBeginInsert();
2052 tgl@sss.pgh.pa.us 2527 : 313029 : XLogRegisterData((char *) xlrec, tupledata - scratch.data);
3433 heikki.linnakangas@i 2528 : 313029 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD | bufflags);
2529 : :
2530 : 313029 : XLogRegisterBufData(0, tupledata, totaldatalen);
2531 : :
2532 : : /* filtering by origin on a row level is much more efficient */
2670 andres@anarazel.de 2533 : 313029 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
2534 : :
3433 heikki.linnakangas@i 2535 : 313029 : recptr = XLogInsert(RM_HEAP2_ID, info);
2536 : :
4540 2537 : 313029 : PageSetLSN(page, recptr);
2538 : : }
2539 : :
2540 [ - + ]: 317389 : END_CRIT_SECTION();
2541 : :
2542 : : /*
2543 : : * If we've frozen everything on the page, update the visibilitymap.
2544 : : * We're already holding pin on the vmbuffer.
2545 : : */
1183 tomas.vondra@postgre 2546 [ + + ]: 317389 : if (all_frozen_set)
2547 : : {
2548 [ - + ]: 1661 : Assert(PageIsAllVisible(page));
2549 [ - + ]: 1661 : Assert(visibilitymap_pin_ok(BufferGetBlockNumber(buffer), vmbuffer));
2550 : :
2551 : : /*
2552 : : * It's fine to use InvalidTransactionId here - this is only used
2553 : : * when HEAP_INSERT_FROZEN is specified, which intentionally
2554 : : * violates visibility rules.
2555 : : */
2556 : 1661 : visibilitymap_set(relation, BufferGetBlockNumber(buffer), buffer,
2557 : : InvalidXLogRecPtr, vmbuffer,
2558 : : InvalidTransactionId,
2559 : : VISIBILITYMAP_ALL_VISIBLE | VISIBILITYMAP_ALL_FROZEN);
2560 : : }
2561 : :
2562 : 317389 : UnlockReleaseBuffer(buffer);
4540 heikki.linnakangas@i 2563 : 317389 : ndone += nthispage;
2564 : :
2565 : : /*
2566 : : * NB: Only release vmbuffer after inserting all tuples - it's fairly
2567 : : * likely that we'll insert into subsequent heap pages that are likely
2568 : : * to use the same vm page.
2569 : : */
2570 : : }
2571 : :
2572 : : /* We're done with inserting all tuples, so release the last vmbuffer. */
1183 tomas.vondra@postgre 2573 [ + + ]: 302969 : if (vmbuffer != InvalidBuffer)
2574 : 3601 : ReleaseBuffer(vmbuffer);
2575 : :
2576 : : /*
2577 : : * We're done with the actual inserts. Check for conflicts again, to
2578 : : * ensure that all rw-conflicts in to these inserts are detected. Without
2579 : : * this final check, a sequential scan of the heap may have locked the
2580 : : * table after the "before" check, missing one opportunity to detect the
2581 : : * conflict, and then scanned the table before the new tuples were there,
2582 : : * missing the other chance to detect the conflict.
2583 : : *
2584 : : * For heap inserts, we only need to check for table-level SSI locks. Our
2585 : : * new tuples can't possibly conflict with existing tuple locks, and heap
2586 : : * page locks are only consolidated versions of tuple locks; they do not
2587 : : * lock "gaps" as index page locks do. So we don't need to specify a
2588 : : * buffer when making the call.
2589 : : */
1538 tmunro@postgresql.or 2590 : 302969 : CheckForSerializableConflictIn(relation, NULL, InvalidBlockNumber);
2591 : :
2592 : : /*
2593 : : * If tuples are cachable, mark them for invalidation from the caches in
2594 : : * case we abort. Note it is OK to do this after releasing the buffer,
2595 : : * because the heaptuples data structure is all in local memory, not in
2596 : : * the shared buffer.
2597 : : */
3790 rhaas@postgresql.org 2598 [ + + ]: 302969 : if (IsCatalogRelation(relation))
2599 : : {
4540 heikki.linnakangas@i 2600 [ + + ]: 1061783 : for (i = 0; i < ntuples; i++)
2601 : 760073 : CacheInvalidateHeapTuple(relation, heaptuples[i], NULL);
2602 : : }
2603 : :
2604 : : /* copy t_self fields back to the caller's slots */
4444 2605 [ + + ]: 1756227 : for (i = 0; i < ntuples; i++)
1837 andres@anarazel.de 2606 : 1453258 : slots[i]->tts_tid = heaptuples[i]->t_self;
2607 : :
4540 heikki.linnakangas@i 2608 : 302969 : pgstat_count_heap_insert(relation, ntuples);
2609 : 302969 : }
2610 : :
2611 : : /*
2612 : : * simple_heap_insert - insert a tuple
2613 : : *
2614 : : * Currently, this routine differs from heap_insert only in supplying
2615 : : * a default command ID and not allowing access to the speedup options.
2616 : : *
2617 : : * This should be used rather than using heap_insert directly in most places
2618 : : * where we are modifying system catalogs.
2619 : : */
2620 : : void
7999 tgl@sss.pgh.pa.us 2621 : 728122 : simple_heap_insert(Relation relation, HeapTuple tup)
2622 : : {
1972 andres@anarazel.de 2623 : 728122 : heap_insert(relation, tup, GetCurrentCommandId(true), 0, NULL);
7999 tgl@sss.pgh.pa.us 2624 : 728122 : }
2625 : :
2626 : : /*
2627 : : * Given infomask/infomask2, compute the bits that must be saved in the
2628 : : * "infobits" field of xl_heap_delete, xl_heap_update, xl_heap_lock,
2629 : : * xl_heap_lock_updated WAL records.
2630 : : *
2631 : : * See fix_infomask_from_infobits.
2632 : : */
2633 : : static uint8
4099 alvherre@alvh.no-ip. 2634 : 1849505 : compute_infobits(uint16 infomask, uint16 infomask2)
2635 : : {
2636 : : return
2637 : 1849505 : ((infomask & HEAP_XMAX_IS_MULTI) != 0 ? XLHL_XMAX_IS_MULTI : 0) |
2638 : 1849505 : ((infomask & HEAP_XMAX_LOCK_ONLY) != 0 ? XLHL_XMAX_LOCK_ONLY : 0) |
2639 : 1849505 : ((infomask & HEAP_XMAX_EXCL_LOCK) != 0 ? XLHL_XMAX_EXCL_LOCK : 0) |
2640 : : /* note we ignore HEAP_XMAX_SHR_LOCK here */
2641 : 3699010 : ((infomask & HEAP_XMAX_KEYSHR_LOCK) != 0 ? XLHL_XMAX_KEYSHR_LOCK : 0) |
2642 : : ((infomask2 & HEAP_KEYS_UPDATED) != 0 ?
2643 : 1849505 : XLHL_KEYS_UPDATED : 0);
2644 : : }
2645 : :
2646 : : /*
2647 : : * Given two versions of the same t_infomask for a tuple, compare them and
2648 : : * return whether the relevant status for a tuple Xmax has changed. This is
2649 : : * used after a buffer lock has been released and reacquired: we want to ensure
2650 : : * that the tuple state continues to be the same it was when we previously
2651 : : * examined it.
2652 : : *
2653 : : * Note the Xmax field itself must be compared separately.
2654 : : */
2655 : : static inline bool
3643 2656 : 5323 : xmax_infomask_changed(uint16 new_infomask, uint16 old_infomask)
2657 : : {
3631 bruce@momjian.us 2658 : 5323 : const uint16 interesting =
2659 : : HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY | HEAP_LOCK_MASK;
2660 : :
3643 alvherre@alvh.no-ip. 2661 [ + + ]: 5323 : if ((new_infomask & interesting) != (old_infomask & interesting))
2662 : 13 : return true;
2663 : :
2664 : 5310 : return false;
2665 : : }
2666 : :
2667 : : /*
2668 : : * heap_delete - delete a tuple
2669 : : *
2670 : : * See table_tuple_delete() for an explanation of the parameters, except that
2671 : : * this routine directly takes a tuple rather than a slot.
2672 : : *
2673 : : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
2674 : : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
2675 : : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
2676 : : * generated by another transaction).
2677 : : */
2678 : : TM_Result
7999 tgl@sss.pgh.pa.us 2679 : 1432659 : heap_delete(Relation relation, ItemPointer tid,
2680 : : CommandId cid, Snapshot crosscheck, bool wait,
2681 : : TM_FailureData *tmfd, bool changingPart)
2682 : : {
2683 : : TM_Result result;
7150 2684 : 1432659 : TransactionId xid = GetCurrentTransactionId();
2685 : : ItemId lp;
2686 : : HeapTupleData tp;
2687 : : Page page;
2688 : : BlockNumber block;
2689 : : Buffer buffer;
4681 rhaas@postgresql.org 2690 : 1432659 : Buffer vmbuffer = InvalidBuffer;
2691 : : TransactionId new_xmax;
2692 : : uint16 new_infomask,
2693 : : new_infomask2;
6924 tgl@sss.pgh.pa.us 2694 : 1432659 : bool have_tuple_lock = false;
2695 : : bool iscombo;
5611 heikki.linnakangas@i 2696 : 1432659 : bool all_visible_cleared = false;
3631 bruce@momjian.us 2697 : 1432659 : HeapTuple old_key_tuple = NULL; /* replica identity of the tuple */
3778 rhaas@postgresql.org 2698 : 1432659 : bool old_key_copied = false;
2699 : :
9716 bruce@momjian.us 2700 [ - + ]: 1432659 : Assert(ItemPointerIsValid(tid));
2701 : :
2702 : : /*
2703 : : * Forbid this during a parallel operation, lest it allocate a combo CID.
2704 : : * Other workers might need that combo CID for visibility checks, and we
2705 : : * have no provision for broadcasting it to them.
2706 : : */
3272 rhaas@postgresql.org 2707 [ - + ]: 1432659 : if (IsInParallelMode())
3272 rhaas@postgresql.org 2708 [ # # ]:UBC 0 : ereport(ERROR,
2709 : : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
2710 : : errmsg("cannot delete tuples during a parallel operation")));
2711 : :
4681 rhaas@postgresql.org 2712 :CBC 1432659 : block = ItemPointerGetBlockNumber(tid);
2713 : 1432659 : buffer = ReadBuffer(relation, block);
2916 kgrittn@postgresql.o 2714 : 1432659 : page = BufferGetPage(buffer);
2715 : :
2716 : : /*
2717 : : * Before locking the buffer, pin the visibility map page if it appears to
2718 : : * be necessary. Since we haven't got the lock yet, someone else might be
2719 : : * in the middle of changing this, so we'll need to recheck after we have
2720 : : * the lock.
2721 : : */
4681 rhaas@postgresql.org 2722 [ + + ]: 1432659 : if (PageIsAllVisible(page))
2723 : 130 : visibilitymap_pin(relation, block, &vmbuffer);
2724 : :
9252 vadim4o@yahoo.com 2725 : 1432659 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2726 : :
570 jdavis@postgresql.or 2727 : 1432659 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
2728 [ - + ]: 1432659 : Assert(ItemIdIsNormal(lp));
2729 : :
2730 : 1432659 : tp.t_tableOid = RelationGetRelid(relation);
2731 : 1432659 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
2732 : 1432659 : tp.t_len = ItemIdGetLength(lp);
2733 : 1432659 : tp.t_self = *tid;
2734 : :
2735 : 1 : l1:
2736 : :
2737 : : /*
2738 : : * If we didn't pin the visibility map page and the page has become all
2739 : : * visible while we were busy locking the buffer, we'll have to unlock and
2740 : : * re-lock, to avoid holding the buffer lock across an I/O. That's a bit
2741 : : * unfortunate, but hopefully shouldn't happen often.
2742 : : */
4681 rhaas@postgresql.org 2743 [ + + - + ]: 1432660 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
2744 : : {
4681 rhaas@postgresql.org 2745 :UBC 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2746 : 0 : visibilitymap_pin(relation, block, &vmbuffer);
2747 : 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2748 : : }
2749 : :
3919 rhaas@postgresql.org 2750 :CBC 1432660 : result = HeapTupleSatisfiesUpdate(&tp, cid, buffer);
2751 : :
1849 andres@anarazel.de 2752 [ - + ]: 1432660 : if (result == TM_Invisible)
2753 : : {
6589 tgl@sss.pgh.pa.us 2754 :UBC 0 : UnlockReleaseBuffer(buffer);
3178 2755 [ # # ]: 0 : ereport(ERROR,
2756 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
2757 : : errmsg("attempted to delete invisible tuple")));
2758 : : }
3 akorotkov@postgresql 2759 [ + + + - ]:CBC 1432660 : else if (result == TM_BeingModified && wait)
2760 : : {
2761 : : TransactionId xwait;
2762 : : uint16 infomask;
2763 : :
2764 : : /* must copy state data before unlocking buffer */
4099 alvherre@alvh.no-ip. 2765 : 40549 : xwait = HeapTupleHeaderGetRawXmax(tp.t_data);
6924 tgl@sss.pgh.pa.us 2766 : 40549 : infomask = tp.t_data->t_infomask;
2767 : :
2768 : : /*
2769 : : * Sleep until concurrent transaction ends -- except when there's a
2770 : : * single locker and it's our own transaction. Note we don't care
2771 : : * which lock mode the locker has, because we need the strongest one.
2772 : : *
2773 : : * Before sleeping, we need to acquire tuple lock to establish our
2774 : : * priority for the tuple (see heap_lock_tuple). LockTuple will
2775 : : * release us when we are next-in-line for the tuple.
2776 : : *
2777 : : * If we are forced to "start over" below, we keep the tuple lock;
2778 : : * this arranges that we stay at the head of the line while rechecking
2779 : : * tuple state.
2780 : : */
6926 2781 [ + + ]: 40549 : if (infomask & HEAP_XMAX_IS_MULTI)
2782 : : {
1762 alvherre@alvh.no-ip. 2783 : 8 : bool current_is_member = false;
2784 : :
3292 2785 [ + - ]: 8 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
2786 : : LockTupleExclusive, ¤t_is_member))
2787 : : {
2788 : 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2789 : :
2790 : : /*
2791 : : * Acquire the lock, if necessary (but skip it when we're
2792 : : * requesting a lock and already have one; avoids deadlock).
2793 : : */
1762 2794 [ + + ]: 8 : if (!current_is_member)
2795 : 6 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2796 : : LockWaitBlock, &have_tuple_lock);
2797 : :
2798 : : /* wait for multixact */
3292 2799 : 8 : MultiXactIdWait((MultiXactId) xwait, MultiXactStatusUpdate, infomask,
2800 : : relation, &(tp.t_self), XLTW_Delete,
2801 : : NULL);
2802 : 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2803 : :
2804 : : /*
2805 : : * If xwait had just locked the tuple then some other xact
2806 : : * could update this tuple before we get to this point. Check
2807 : : * for xmax change, and start over if so.
2808 : : *
2809 : : * We also must start over if we didn't pin the VM page, and
2810 : : * the page has become all visible.
2811 : : */
570 jdavis@postgresql.or 2812 [ + - + - : 16 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
+ - ]
2813 : 8 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
3292 alvherre@alvh.no-ip. 2814 [ - + ]: 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2815 : : xwait))
3292 alvherre@alvh.no-ip. 2816 :UBC 0 : goto l1;
2817 : : }
2818 : :
2819 : : /*
2820 : : * You might think the multixact is necessarily done here, but not
2821 : : * so: it could have surviving members, namely our own xact or
2822 : : * other subxacts of this backend. It is legal for us to delete
2823 : : * the tuple in either case, however (the latter case is
2824 : : * essentially a situation of upgrading our former shared lock to
2825 : : * exclusive). We don't bother changing the on-disk hint bits
2826 : : * since we are about to overwrite the xmax altogether.
2827 : : */
2828 : : }
3292 alvherre@alvh.no-ip. 2829 [ + + ]:CBC 40541 : else if (!TransactionIdIsCurrentTransactionId(xwait))
2830 : : {
2831 : : /*
2832 : : * Wait for regular transaction to end; but first, acquire tuple
2833 : : * lock.
2834 : : */
2835 : 40 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2836 : 40 : heap_acquire_tuplock(relation, &(tp.t_self), LockTupleExclusive,
2837 : : LockWaitBlock, &have_tuple_lock);
3357 heikki.linnakangas@i 2838 : 40 : XactLockTableWait(xwait, relation, &(tp.t_self), XLTW_Delete);
6926 tgl@sss.pgh.pa.us 2839 : 36 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2840 : :
2841 : : /*
2842 : : * xwait is done, but if xwait had just locked the tuple then some
2843 : : * other xact could update this tuple before we get to this point.
2844 : : * Check for xmax change, and start over if so.
2845 : : *
2846 : : * We also must start over if we didn't pin the VM page, and the
2847 : : * page has become all visible.
2848 : : */
570 jdavis@postgresql.or 2849 [ + - + - : 72 : if ((vmbuffer == InvalidBuffer && PageIsAllVisible(page)) ||
+ + ]
2850 : 36 : xmax_infomask_changed(tp.t_data->t_infomask, infomask) ||
4099 alvherre@alvh.no-ip. 2851 [ - + ]: 35 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tp.t_data),
2852 : : xwait))
6926 tgl@sss.pgh.pa.us 2853 : 1 : goto l1;
2854 : :
2855 : : /* Otherwise check if it committed or aborted */
6088 2856 : 35 : UpdateXmaxHintBits(tp.t_data, buffer, xwait);
2857 : : }
2858 : :
2859 : : /*
2860 : : * We may overwrite if previous xmax aborted, or if it committed but
2861 : : * only locked the tuple without updating it.
2862 : : */
4099 alvherre@alvh.no-ip. 2863 [ + + ]: 40544 : if ((tp.t_data->t_infomask & HEAP_XMAX_INVALID) ||
2864 [ + + + - : 40555 : HEAP_XMAX_IS_LOCKED_ONLY(tp.t_data->t_infomask) ||
+ + ]
2865 : 25 : HeapTupleHeaderIsOnlyLocked(tp.t_data))
1849 andres@anarazel.de 2866 : 40523 : result = TM_Ok;
1147 alvherre@alvh.no-ip. 2867 [ + + ]: 21 : else if (!ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid))
1849 andres@anarazel.de 2868 : 17 : result = TM_Updated;
2869 : : else
2870 : 4 : result = TM_Deleted;
2871 : : }
2872 : :
2873 : : /* sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
2874 [ + + ]: 1432655 : if (result != TM_Ok)
2875 : : {
2876 [ + + + + : 55 : Assert(result == TM_SelfModified ||
- + - - ]
2877 : : result == TM_Updated ||
2878 : : result == TM_Deleted ||
2879 : : result == TM_BeingModified);
6812 tgl@sss.pgh.pa.us 2880 [ - + ]: 55 : Assert(!(tp.t_data->t_infomask & HEAP_XMAX_INVALID));
1849 andres@anarazel.de 2881 [ + + - + ]: 55 : Assert(result != TM_Updated ||
2882 : : !ItemPointerEquals(&tp.t_self, &tp.t_data->t_ctid));
2883 : : }
2884 : :
138 heikki.linnakangas@i 2885 [ + + + - ]: 1432655 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
2886 : : {
2887 : : /* Perform additional check for transaction-snapshot mode RI updates */
2888 [ + - ]: 1 : if (!HeapTupleSatisfiesVisibility(&tp, crosscheck, buffer))
2889 : 1 : result = TM_Updated;
2890 : : }
2891 : :
2892 [ + + ]: 1432655 : if (result != TM_Ok)
2893 : : {
1849 andres@anarazel.de 2894 : 56 : tmfd->ctid = tp.t_data->t_ctid;
2895 [ + + - + : 56 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tp.t_data);
- - ]
2896 [ + + ]: 56 : if (result == TM_SelfModified)
2897 : 21 : tmfd->cmax = HeapTupleHeaderGetCmax(tp.t_data);
2898 : : else
2899 : 35 : tmfd->cmax = InvalidCommandId;
3 akorotkov@postgresql 2900 : 56 : UnlockReleaseBuffer(buffer);
6924 tgl@sss.pgh.pa.us 2901 [ + + ]: 56 : if (have_tuple_lock)
4099 alvherre@alvh.no-ip. 2902 : 21 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
4681 rhaas@postgresql.org 2903 [ - + ]: 56 : if (vmbuffer != InvalidBuffer)
4681 rhaas@postgresql.org 2904 :UBC 0 : ReleaseBuffer(vmbuffer);
9252 vadim4o@yahoo.com 2905 :CBC 56 : return result;
2906 : : }
2907 : :
2908 : : /*
2909 : : * We're about to do the actual delete -- check for conflict first, to
2910 : : * avoid possibly having to roll back work we've just done.
2911 : : *
2912 : : * This is safe without a recheck as long as there is no possibility of
2913 : : * another process scanning the page between this check and the delete
2914 : : * being visible to the scan (i.e., an exclusive buffer content lock is
2915 : : * continuously held from this point until the tuple delete is visible).
2916 : : */
1538 tmunro@postgresql.or 2917 : 1432599 : CheckForSerializableConflictIn(relation, tid, BufferGetBlockNumber(buffer));
2918 : :
2919 : : /* replace cid with a combo CID if necessary */
6274 tgl@sss.pgh.pa.us 2920 : 1432585 : HeapTupleHeaderAdjustCmax(tp.t_data, &cid, &iscombo);
2921 : :
2922 : : /*
2923 : : * Compute replica identity tuple before entering the critical section so
2924 : : * we don't PANIC upon a memory allocation failure.
2925 : : */
3778 rhaas@postgresql.org 2926 : 1432585 : old_key_tuple = ExtractReplicaIdentity(relation, &tp, true, &old_key_copied);
2927 : :
2928 : : /*
2929 : : * If this is the first possibly-multixact-able operation in the current
2930 : : * transaction, set my per-backend OldestMemberMXactId setting. We can be
2931 : : * certain that the transaction will never become a member of any older
2932 : : * MultiXactIds than that. (We have to do this even if we end up just
2933 : : * using our own TransactionId below, since some other backend could
2934 : : * incorporate our XID into a MultiXact immediately afterwards.)
2935 : : */
3663 heikki.linnakangas@i 2936 : 1432585 : MultiXactIdSetOldestMember();
2937 : :
2938 : 1432585 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(tp.t_data),
2939 : 1432585 : tp.t_data->t_infomask, tp.t_data->t_infomask2,
2940 : : xid, LockTupleExclusive, true,
2941 : : &new_xmax, &new_infomask, &new_infomask2);
2942 : :
8493 tgl@sss.pgh.pa.us 2943 : 1432585 : START_CRIT_SECTION();
2944 : :
2945 : : /*
2946 : : * If this transaction commits, the tuple will become DEAD sooner or
2947 : : * later. Set flag that this page is a candidate for pruning once our xid
2948 : : * falls below the OldestXmin horizon. If the transaction finally aborts,
2949 : : * the subsequent page pruning will be a no-op and the hint will be
2950 : : * cleared.
2951 : : */
5754 2952 [ - + + + : 1432585 : PageSetPrunable(page, xid);
+ + ]
2953 : :
5611 heikki.linnakangas@i 2954 [ + + ]: 1432585 : if (PageIsAllVisible(page))
2955 : : {
2956 : 130 : all_visible_cleared = true;
2957 : 130 : PageClearAllVisible(page);
4681 rhaas@postgresql.org 2958 : 130 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
2959 : : vmbuffer, VISIBILITYMAP_VALID_BITS);
2960 : : }
2961 : :
2962 : : /* store transaction information of xact deleting the tuple */
4099 alvherre@alvh.no-ip. 2963 : 1432585 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
2964 : 1432585 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
2965 : 1432585 : tp.t_data->t_infomask |= new_infomask;
2966 : 1432585 : tp.t_data->t_infomask2 |= new_infomask2;
6051 tgl@sss.pgh.pa.us 2967 : 1432585 : HeapTupleHeaderClearHotUpdated(tp.t_data);
4099 alvherre@alvh.no-ip. 2968 : 1432585 : HeapTupleHeaderSetXmax(tp.t_data, new_xmax);
6274 tgl@sss.pgh.pa.us 2969 [ - + + + ]: 1432585 : HeapTupleHeaderSetCmax(tp.t_data, cid, iscombo);
2970 : : /* Make sure there is no forward chain link in t_ctid */
7915 2971 : 1432585 : tp.t_data->t_ctid = tp.t_self;
2972 : :
2973 : : /* Signal that this is actually a move into another partition */
2199 andres@anarazel.de 2974 [ + + ]: 1432585 : if (changingPart)
2975 : 446 : HeapTupleHeaderSetMovedPartitions(tp.t_data);
2976 : :
6589 tgl@sss.pgh.pa.us 2977 : 1432585 : MarkBufferDirty(buffer);
2978 : :
2979 : : /*
2980 : : * XLOG stuff
2981 : : *
2982 : : * NB: heap_abort_speculative() uses the same xlog record and replay
2983 : : * routines.
2984 : : */
4871 rhaas@postgresql.org 2985 [ + + + + : 1432585 : if (RelationNeedsWAL(relation))
+ + + + ]
2986 : : {
2987 : : xl_heap_delete xlrec;
2988 : : xl_heap_header xlhdr;
2989 : : XLogRecPtr recptr;
2990 : :
2991 : : /*
2992 : : * For logical decode we need combo CIDs to properly decode the
2993 : : * catalog
2994 : : */
3778 2995 [ + + + - : 1372003 : if (RelationIsAccessibleInLogicalDecoding(relation))
- + - - -
- + + - +
- - - - -
- ]
2996 : 5286 : log_heap_new_cid(relation, &tp);
2997 : :
2199 andres@anarazel.de 2998 : 1372003 : xlrec.flags = 0;
2999 [ + + ]: 1372003 : if (all_visible_cleared)
3000 : 130 : xlrec.flags |= XLH_DELETE_ALL_VISIBLE_CLEARED;
3001 [ + + ]: 1372003 : if (changingPart)
3002 : 446 : xlrec.flags |= XLH_DELETE_IS_PARTITION_MOVE;
4099 alvherre@alvh.no-ip. 3003 : 2744006 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
3004 : 1372003 : tp.t_data->t_infomask2);
3433 heikki.linnakangas@i 3005 : 1372003 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
4099 alvherre@alvh.no-ip. 3006 : 1372003 : xlrec.xmax = new_xmax;
3007 : :
3433 heikki.linnakangas@i 3008 [ + + ]: 1372003 : if (old_key_tuple != NULL)
3009 : : {
3010 [ + + ]: 47000 : if (relation->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
3264 andres@anarazel.de 3011 : 121 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_TUPLE;
3012 : : else
3013 : 46879 : xlrec.flags |= XLH_DELETE_CONTAINS_OLD_KEY;
3014 : : }
3015 : :
3433 heikki.linnakangas@i 3016 : 1372003 : XLogBeginInsert();
3017 : 1372003 : XLogRegisterData((char *) &xlrec, SizeOfHeapDelete);
3018 : :
3019 : 1372003 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3020 : :
3021 : : /*
3022 : : * Log replica identity of the deleted tuple if there is one
3023 : : */
3778 rhaas@postgresql.org 3024 [ + + ]: 1372003 : if (old_key_tuple != NULL)
3025 : : {
3026 : 47000 : xlhdr.t_infomask2 = old_key_tuple->t_data->t_infomask2;
3027 : 47000 : xlhdr.t_infomask = old_key_tuple->t_data->t_infomask;
3028 : 47000 : xlhdr.t_hoff = old_key_tuple->t_data->t_hoff;
3029 : :
3433 heikki.linnakangas@i 3030 : 47000 : XLogRegisterData((char *) &xlhdr, SizeOfHeapHeader);
3031 : 47000 : XLogRegisterData((char *) old_key_tuple->t_data
3032 : : + SizeofHeapTupleHeader,
3033 : 47000 : old_key_tuple->t_len
3034 : : - SizeofHeapTupleHeader);
3035 : : }
3036 : :
3037 : : /* filtering by origin on a row level is much more efficient */
2670 andres@anarazel.de 3038 : 1372003 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
3039 : :
3433 heikki.linnakangas@i 3040 : 1372003 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
3041 : :
5754 tgl@sss.pgh.pa.us 3042 : 1372003 : PageSetLSN(page, recptr);
3043 : : }
3044 : :
8493 3045 [ - + ]: 1432585 : END_CRIT_SECTION();
3046 : :
8490 3047 : 1432585 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3048 : :
4681 rhaas@postgresql.org 3049 [ + + ]: 1432585 : if (vmbuffer != InvalidBuffer)
3050 : 130 : ReleaseBuffer(vmbuffer);
3051 : :
3052 : : /*
3053 : : * If the tuple has toasted out-of-line attributes, we need to delete
3054 : : * those items too. We have to do this before releasing the buffer
3055 : : * because we need to look at the contents of the tuple, but it's OK to
3056 : : * release the content lock on the buffer first.
3057 : : */
4060 kgrittn@postgresql.o 3058 [ + + ]: 1432585 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
3059 [ + + ]: 1706 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3060 : : {
3061 : : /* toast table entries should never be recursively toasted */
6221 tgl@sss.pgh.pa.us 3062 [ - + ]: 1696 : Assert(!HeapTupleHasExternal(&tp));
3063 : : }
3064 [ + + ]: 1430889 : else if (HeapTupleHasExternal(&tp))
1654 rhaas@postgresql.org 3065 : 256 : heap_toast_delete(relation, &tp, false);
3066 : :
3067 : : /*
3068 : : * Mark tuple for invalidation from system caches at next command
3069 : : * boundary. We have to do this before releasing the buffer because we
3070 : : * need to look at the contents of the tuple.
3071 : : */
4625 tgl@sss.pgh.pa.us 3072 : 1432585 : CacheInvalidateHeapTuple(relation, &tp, NULL);
3073 : :
3074 : : /* Now we can release the buffer */
3 akorotkov@postgresql 3075 : 1432585 : ReleaseBuffer(buffer);
3076 : :
3077 : : /*
3078 : : * Release the lmgr tuple lock, if we had it.
3079 : : */
6924 tgl@sss.pgh.pa.us 3080 [ + + ]: 1432585 : if (have_tuple_lock)
4099 alvherre@alvh.no-ip. 3081 : 20 : UnlockTupleTuplock(relation, &(tp.t_self), LockTupleExclusive);
3082 : :
6167 tgl@sss.pgh.pa.us 3083 : 1432585 : pgstat_count_heap_delete(relation);
3084 : :
3778 rhaas@postgresql.org 3085 [ + + + + ]: 1432585 : if (old_key_tuple != NULL && old_key_copied)
3086 : 46880 : heap_freetuple(old_key_tuple);
3087 : :
1849 andres@anarazel.de 3088 : 1432585 : return TM_Ok;
3089 : : }
3090 : :
3091 : : /*
3092 : : * simple_heap_delete - delete a tuple
3093 : : *
3094 : : * This routine may be used to delete a tuple when concurrent updates of
3095 : : * the target tuple are not expected (for example, because we have a lock
3096 : : * on the relation associated with the tuple). Any failure is reported
3097 : : * via ereport().
3098 : : */
3099 : : void
8482 tgl@sss.pgh.pa.us 3100 : 571865 : simple_heap_delete(Relation relation, ItemPointer tid)
3101 : : {
3102 : : TM_Result result;
3103 : : TM_FailureData tmfd;
3104 : :
7517 3105 : 571865 : result = heap_delete(relation, tid,
3106 : : GetCurrentCommandId(true), InvalidSnapshot,
3107 : : true /* wait for commit */ ,
3108 : : &tmfd, false /* changingPart */ );
8482 3109 [ - + - - : 571865 : switch (result)
- ]
3110 : : {
1849 andres@anarazel.de 3111 :UBC 0 : case TM_SelfModified:
3112 : : /* Tuple was already updated in current command? */
7573 tgl@sss.pgh.pa.us 3113 [ # # ]: 0 : elog(ERROR, "tuple already updated by self");
3114 : : break;
3115 : :
1849 andres@anarazel.de 3116 :CBC 571865 : case TM_Ok:
3117 : : /* done successfully */
8482 tgl@sss.pgh.pa.us 3118 : 571865 : break;
3119 : :
1849 andres@anarazel.de 3120 :UBC 0 : case TM_Updated:
7573 tgl@sss.pgh.pa.us 3121 [ # # ]: 0 : elog(ERROR, "tuple concurrently updated");
3122 : : break;
3123 : :
1849 andres@anarazel.de 3124 : 0 : case TM_Deleted:
3125 [ # # ]: 0 : elog(ERROR, "tuple concurrently deleted");
3126 : : break;
3127 : :
8482 tgl@sss.pgh.pa.us 3128 : 0 : default:
7573 3129 [ # # ]: 0 : elog(ERROR, "unrecognized heap_delete status: %u", result);
3130 : : break;
3131 : : }
8482 tgl@sss.pgh.pa.us 3132 :CBC 571865 : }
3133 : :
3134 : : /*
3135 : : * heap_update - replace a tuple
3136 : : *
3137 : : * See table_tuple_update() for an explanation of the parameters, except that
3138 : : * this routine directly takes a tuple rather than a slot.
3139 : : *
3140 : : * In the failure cases, the routine fills *tmfd with the tuple's t_ctid,
3141 : : * t_xmax (resolving a possible MultiXact, if necessary), and t_cmax (the last
3142 : : * only for TM_SelfModified, since we cannot obtain cmax from a combo CID
3143 : : * generated by another transaction).
3144 : : */
3145 : : TM_Result
8908 bruce@momjian.us 3146 : 277144 : heap_update(Relation relation, ItemPointer otid, HeapTuple newtup,
3147 : : CommandId cid, Snapshot crosscheck, bool wait,
3148 : : TM_FailureData *tmfd, LockTupleMode *lockmode,
3149 : : TU_UpdateIndexes *update_indexes)
3150 : : {
3151 : : TM_Result result;
7150 tgl@sss.pgh.pa.us 3152 : 277144 : TransactionId xid = GetCurrentTransactionId();
3153 : : Bitmapset *hot_attrs;
3154 : : Bitmapset *sum_attrs;
3155 : : Bitmapset *key_attrs;
3156 : : Bitmapset *id_attrs;
3157 : : Bitmapset *interesting_attrs;
3158 : : Bitmapset *modified_attrs;
3159 : : ItemId lp;
3160 : : HeapTupleData oldtup;
3161 : : HeapTuple heaptup;
3778 rhaas@postgresql.org 3162 : 277144 : HeapTuple old_key_tuple = NULL;
3163 : 277144 : bool old_key_copied = false;
3164 : : Page page;
3165 : : BlockNumber block;
3166 : : MultiXactStatus mxact_status;
3167 : : Buffer buffer,
3168 : : newbuf,
4681 3169 : 277144 : vmbuffer = InvalidBuffer,
3170 : 277144 : vmbuffer_new = InvalidBuffer;
3171 : : bool need_toast;
3172 : : Size newtupsize,
3173 : : pagefree;
6924 tgl@sss.pgh.pa.us 3174 : 277144 : bool have_tuple_lock = false;
3175 : : bool iscombo;
6051 3176 : 277144 : bool use_hot_update = false;
391 tomas.vondra@postgre 3177 : 277144 : bool summarized_update = false;
3178 : : bool key_intact;
5611 heikki.linnakangas@i 3179 : 277144 : bool all_visible_cleared = false;
3180 : 277144 : bool all_visible_cleared_new = false;
3181 : : bool checked_lockers;
3182 : : bool locker_remains;
790 akapila@postgresql.o 3183 : 277144 : bool id_has_external = false;
3184 : : TransactionId xmax_new_tuple,
3185 : : xmax_old_tuple;
3186 : : uint16 infomask_old_tuple,
3187 : : infomask2_old_tuple,
3188 : : infomask_new_tuple,
3189 : : infomask2_new_tuple;
3190 : :
9716 bruce@momjian.us 3191 [ - + ]: 277144 : Assert(ItemPointerIsValid(otid));
3192 : :
3193 : : /* Cheap, simplistic check that the tuple matches the rel's rowtype. */
1070 tgl@sss.pgh.pa.us 3194 [ - + ]: 277144 : Assert(HeapTupleHeaderGetNatts(newtup->t_data) <=
3195 : : RelationGetNumberOfAttributes(relation));
3196 : :
3197 : : /*
3198 : : * Forbid this during a parallel operation, lest it allocate a combo CID.
3199 : : * Other workers might need that combo CID for visibility checks, and we
3200 : : * have no provision for broadcasting it to them.
3201 : : */
3272 rhaas@postgresql.org 3202 [ - + ]: 277144 : if (IsInParallelMode())
3272 rhaas@postgresql.org 3203 [ # # ]:UBC 0 : ereport(ERROR,
3204 : : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
3205 : : errmsg("cannot update tuples during a parallel operation")));
3206 : :
3207 : : /*
3208 : : * Fetch the list of attributes to be checked for various operations.
3209 : : *
3210 : : * For HOT considerations, this is wasted effort if we fail to update or
3211 : : * have to put the new tuple on a different page. But we must compute the
3212 : : * list before obtaining buffer lock --- in the worst case, if we are
3213 : : * doing an update on one of the relevant system catalogs, we could
3214 : : * deadlock if we try to fetch the list later. In any case, the relcache
3215 : : * caches the data so this is usually pretty cheap.
3216 : : *
3217 : : * We also need columns used by the replica identity and columns that are
3218 : : * considered the "key" of rows in the table.
3219 : : *
3220 : : * Note that we get copies of each bitmap, so we need not worry about
3221 : : * relcache flush happening midway through.
3222 : : */
391 tomas.vondra@postgre 3223 :CBC 277144 : hot_attrs = RelationGetIndexAttrBitmap(relation,
3224 : : INDEX_ATTR_BITMAP_HOT_BLOCKING);
3225 : 277144 : sum_attrs = RelationGetIndexAttrBitmap(relation,
3226 : : INDEX_ATTR_BITMAP_SUMMARIZED);
3778 rhaas@postgresql.org 3227 : 277144 : key_attrs = RelationGetIndexAttrBitmap(relation, INDEX_ATTR_BITMAP_KEY);
3228 : 277144 : id_attrs = RelationGetIndexAttrBitmap(relation,
3229 : : INDEX_ATTR_BITMAP_IDENTITY_KEY);
870 pg@bowt.ie 3230 : 277144 : interesting_attrs = NULL;
3231 : 277144 : interesting_attrs = bms_add_members(interesting_attrs, hot_attrs);
391 tomas.vondra@postgre 3232 : 277144 : interesting_attrs = bms_add_members(interesting_attrs, sum_attrs);
870 pg@bowt.ie 3233 : 277144 : interesting_attrs = bms_add_members(interesting_attrs, key_attrs);
3234 : 277144 : interesting_attrs = bms_add_members(interesting_attrs, id_attrs);
3235 : :
4681 rhaas@postgresql.org 3236 : 277144 : block = ItemPointerGetBlockNumber(otid);
3237 : 277144 : buffer = ReadBuffer(relation, block);
2916 kgrittn@postgresql.o 3238 : 277144 : page = BufferGetPage(buffer);
3239 : :
3240 : : /*
3241 : : * Before locking the buffer, pin the visibility map page if it appears to
3242 : : * be necessary. Since we haven't got the lock yet, someone else might be
3243 : : * in the middle of changing this, so we'll need to recheck after we have
3244 : : * the lock.
3245 : : */
4681 rhaas@postgresql.org 3246 [ + + ]: 277144 : if (PageIsAllVisible(page))
3247 : 1176 : visibilitymap_pin(relation, block, &vmbuffer);
3248 : :
9252 vadim4o@yahoo.com 3249 : 277144 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3250 : :
5754 tgl@sss.pgh.pa.us 3251 : 277144 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(otid));
6059 3252 [ - + ]: 277144 : Assert(ItemIdIsNormal(lp));
3253 : :
3254 : : /*
3255 : : * Fill in enough data in oldtup for HeapDetermineColumnsInfo to work
3256 : : * properly.
3257 : : */
4090 alvherre@alvh.no-ip. 3258 : 277144 : oldtup.t_tableOid = RelationGetRelid(relation);
5754 tgl@sss.pgh.pa.us 3259 : 277144 : oldtup.t_data = (HeapTupleHeader) PageGetItem(page, lp);
9270 vadim4o@yahoo.com 3260 : 277144 : oldtup.t_len = ItemIdGetLength(lp);
3261 : 277144 : oldtup.t_self = *otid;
3262 : :
3263 : : /* the new tuple is ready, except for this: */
4090 alvherre@alvh.no-ip. 3264 : 277144 : newtup->t_tableOid = RelationGetRelid(relation);
3265 : :
3266 : : /*
3267 : : * Determine columns modified by the update. Additionally, identify
3268 : : * whether any of the unmodified replica identity key attributes in the
3269 : : * old tuple is externally stored or not. This is required because for
3270 : : * such attributes the flattened value won't be WAL logged as part of the
3271 : : * new tuple so we must include it as part of the old_key_tuple. See
3272 : : * ExtractReplicaIdentity.
3273 : : */
790 akapila@postgresql.o 3274 : 277144 : modified_attrs = HeapDetermineColumnsInfo(relation, interesting_attrs,
3275 : : id_attrs, &oldtup,
3276 : : newtup, &id_has_external);
3277 : :
3278 : : /*
3279 : : * If we're not updating any "key" column, we can grab a weaker lock type.
3280 : : * This allows for more concurrency when we are running simultaneously
3281 : : * with foreign key checks.
3282 : : *
3283 : : * Note that if a column gets detoasted while executing the update, but
3284 : : * the value ends up being the same, this test will fail and we will use
3285 : : * the stronger lock. This is acceptable; the important case to optimize
3286 : : * is updates that don't manipulate key columns, not those that
3287 : : * serendipitously arrive at the same key values.
3288 : : */
2573 alvherre@alvh.no-ip. 3289 [ + + ]: 277144 : if (!bms_overlap(modified_attrs, key_attrs))
3290 : : {
2194 simon@2ndQuadrant.co 3291 : 273387 : *lockmode = LockTupleNoKeyExclusive;
4099 alvherre@alvh.no-ip. 3292 : 273387 : mxact_status = MultiXactStatusNoKeyUpdate;
3293 : 273387 : key_intact = true;
3294 : :
3295 : : /*
3296 : : * If this is the first possibly-multixact-able operation in the
3297 : : * current transaction, set my per-backend OldestMemberMXactId
3298 : : * setting. We can be certain that the transaction will never become a
3299 : : * member of any older MultiXactIds than that. (We have to do this
3300 : : * even if we end up just using our own TransactionId below, since
3301 : : * some other backend could incorporate our XID into a MultiXact
3302 : : * immediately afterwards.)
3303 : : */
3304 : 273387 : MultiXactIdSetOldestMember();
3305 : : }
3306 : : else
3307 : : {
2194 simon@2ndQuadrant.co 3308 : 3757 : *lockmode = LockTupleExclusive;
4099 alvherre@alvh.no-ip. 3309 : 3757 : mxact_status = MultiXactStatusUpdate;
3310 : 3757 : key_intact = false;
3311 : : }
3312 : :
3313 : : /*
3314 : : * Note: beyond this point, use oldtup not otid to refer to old tuple.
3315 : : * otid may very well point at newtup->t_self, which we will overwrite
3316 : : * with the new tuple's location, so there's great risk of confusion if we
3317 : : * use otid anymore.
3318 : : */
3319 : :
9252 vadim4o@yahoo.com 3320 : 1 : l2:
4099 alvherre@alvh.no-ip. 3321 : 277145 : checked_lockers = false;
3322 : 277145 : locker_remains = false;
3919 rhaas@postgresql.org 3323 : 277145 : result = HeapTupleSatisfiesUpdate(&oldtup, cid, buffer);
3324 : :
3325 : : /* see below about the "no wait" case */
3 akorotkov@postgresql 3326 [ + + - + ]: 277145 : Assert(result != TM_BeingModified || wait);
3327 : :
1849 andres@anarazel.de 3328 [ - + ]: 277145 : if (result == TM_Invisible)
3329 : : {
6589 tgl@sss.pgh.pa.us 3330 :UBC 0 : UnlockReleaseBuffer(buffer);
3178 3331 [ # # ]: 0 : ereport(ERROR,
3332 : : (errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
3333 : : errmsg("attempted to update invisible tuple")));
3334 : : }
3 akorotkov@postgresql 3335 [ + + + - ]:CBC 277145 : else if (result == TM_BeingModified && wait)
3336 : : {
3337 : : TransactionId xwait;
3338 : : uint16 infomask;
4099 alvherre@alvh.no-ip. 3339 : 35904 : bool can_continue = false;
3340 : :
3341 : : /*
3342 : : * XXX note that we don't consider the "no wait" case here. This
3343 : : * isn't a problem currently because no caller uses that case, but it
3344 : : * should be fixed if such a caller is introduced. It wasn't a
3345 : : * problem previously because this code would always wait, but now
3346 : : * that some tuple locks do not conflict with one of the lock modes we
3347 : : * use, it is possible that this case is interesting to handle
3348 : : * specially.
3349 : : *
3350 : : * This may cause failures with third-party code that calls
3351 : : * heap_update directly.
3352 : : */
3353 : :
3354 : : /* must copy state data before unlocking buffer */
3355 : 35904 : xwait = HeapTupleHeaderGetRawXmax(oldtup.t_data);
6924 tgl@sss.pgh.pa.us 3356 : 35904 : infomask = oldtup.t_data->t_infomask;
3357 : :
3358 : : /*
3359 : : * Now we have to do something about the existing locker. If it's a
3360 : : * multi, sleep on it; we might be awakened before it is completely
3361 : : * gone (or even not sleep at all in some cases); we need to preserve
3362 : : * it as locker, unless it is gone completely.
3363 : : *
3364 : : * If it's not a multi, we need to check for sleeping conditions
3365 : : * before actually going to sleep. If the update doesn't conflict
3366 : : * with the locks, we just continue without sleeping (but making sure
3367 : : * it is preserved).
3368 : : *
3369 : : * Before sleeping, we need to acquire tuple lock to establish our
3370 : : * priority for the tuple (see heap_lock_tuple). LockTuple will
3371 : : * release us when we are next-in-line for the tuple. Note we must
3372 : : * not acquire the tuple lock until we're sure we're going to sleep;
3373 : : * otherwise we're open for race conditions with other transactions
3374 : : * holding the tuple lock which sleep on us.
3375 : : *
3376 : : * If we are forced to "start over" below, we keep the tuple lock;
3377 : : * this arranges that we stay at the head of the line while rechecking
3378 : : * tuple state.
3379 : : */
6926 3380 [ + + ]: 35904 : if (infomask & HEAP_XMAX_IS_MULTI)
3381 : : {
3382 : : TransactionId update_xact;
3383 : : int remain;
1762 alvherre@alvh.no-ip. 3384 : 60 : bool current_is_member = false;
3385 : :
3292 3386 [ + + ]: 60 : if (DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
3387 : : *lockmode, ¤t_is_member))
3388 : : {
3389 : 8 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3390 : :
3391 : : /*
3392 : : * Acquire the lock, if necessary (but skip it when we're
3393 : : * requesting a lock and already have one; avoids deadlock).
3394 : : */
1762 3395 [ - + ]: 8 : if (!current_is_member)
1762 alvherre@alvh.no-ip. 3396 :UBC 0 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3397 : : LockWaitBlock, &have_tuple_lock);
3398 : :
3399 : : /* wait for multixact */
3292 alvherre@alvh.no-ip. 3400 :CBC 8 : MultiXactIdWait((MultiXactId) xwait, mxact_status, infomask,
3401 : : relation, &oldtup.t_self, XLTW_Update,
3402 : : &remain);
3403 : 8 : checked_lockers = true;
3404 : 8 : locker_remains = remain != 0;
3405 : 8 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3406 : :
3407 : : /*
3408 : : * If xwait had just locked the tuple then some other xact
3409 : : * could update this tuple before we get to this point. Check
3410 : : * for xmax change, and start over if so.
3411 : : */
3412 [ + - ]: 8 : if (xmax_infomask_changed(oldtup.t_data->t_infomask,
3413 : 8 : infomask) ||
2489 tgl@sss.pgh.pa.us 3414 [ - + ]: 8 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3415 : : xwait))
3292 alvherre@alvh.no-ip. 3416 :UBC 0 : goto l2;
3417 : : }
3418 : :
3419 : : /*
3420 : : * Note that the multixact may not be done by now. It could have
3421 : : * surviving members; our own xact or other subxacts of this
3422 : : * backend, and also any other concurrent transaction that locked
3423 : : * the tuple with LockTupleKeyShare if we only got
3424 : : * LockTupleNoKeyExclusive. If this is the case, we have to be
3425 : : * careful to mark the updated tuple with the surviving members in
3426 : : * Xmax.
3427 : : *
3428 : : * Note that there could have been another update in the
3429 : : * MultiXact. In that case, we need to check whether it committed
3430 : : * or aborted. If it aborted we are safe to update it again;
3431 : : * otherwise there is an update conflict, and we have to return
3432 : : * TableTuple{Deleted, Updated} below.
3433 : : *
3434 : : * In the LockTupleExclusive case, we still need to preserve the
3435 : : * surviving members: those would include the tuple locks we had
3436 : : * before this one, which are important to keep in case this
3437 : : * subxact aborts.
3438 : : */
4099 alvherre@alvh.no-ip. 3439 [ + + + - ]:CBC 60 : if (!HEAP_XMAX_IS_LOCKED_ONLY(oldtup.t_data->t_infomask))
3440 : 8 : update_xact = HeapTupleGetUpdateXid(oldtup.t_data);
3441 : : else
3292 3442 : 52 : update_xact = InvalidTransactionId;
3443 : :
3444 : : /*
3445 : : * There was no UPDATE in the MultiXact; or it aborted. No
3446 : : * TransactionIdIsInProgress() call needed here, since we called
3447 : : * MultiXactIdWait() above.
3448 : : */
4099 3449 [ + + + + ]: 68 : if (!TransactionIdIsValid(update_xact) ||
3450 : 8 : TransactionIdDidAbort(update_xact))
3451 : 53 : can_continue = true;
3452 : : }
3292 3453 [ + + ]: 35844 : else if (TransactionIdIsCurrentTransactionId(xwait))
3454 : : {
3455 : : /*
3456 : : * The only locker is ourselves; we can avoid grabbing the tuple
3457 : : * lock here, but must preserve our locking information.
3458 : : */
3459 : 35753 : checked_lockers = true;
3460 : 35753 : locker_remains = true;
3461 : 35753 : can_continue = true;
3462 : : }
3463 [ + + + + ]: 91 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) && key_intact)
3464 : : {
3465 : : /*
3466 : : * If it's just a key-share locker, and we're not changing the key
3467 : : * columns, we don't need to wait for it to end; but we need to
3468 : : * preserve it as locker.
3469 : : */
3470 : 29 : checked_lockers = true;
3471 : 29 : locker_remains = true;
3472 : 29 : can_continue = true;
3473 : : }
3474 : : else
3475 : : {
3476 : : /*
3477 : : * Wait for regular transaction to end; but first, acquire tuple
3478 : : * lock.
3479 : : */
3480 : 62 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
2194 simon@2ndQuadrant.co 3481 : 62 : heap_acquire_tuplock(relation, &(oldtup.t_self), *lockmode,
3482 : : LockWaitBlock, &have_tuple_lock);
3292 alvherre@alvh.no-ip. 3483 : 62 : XactLockTableWait(xwait, relation, &oldtup.t_self,
3484 : : XLTW_Update);
3485 : 62 : checked_lockers = true;
3486 : 62 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3487 : :
3488 : : /*
3489 : : * xwait is done, but if xwait had just locked the tuple then some
3490 : : * other xact could update this tuple before we get to this point.
3491 : : * Check for xmax change, and start over if so.
3492 : : */
3493 [ + + ]: 62 : if (xmax_infomask_changed(oldtup.t_data->t_infomask, infomask) ||
3494 [ - + ]: 61 : !TransactionIdEquals(xwait,
3495 : : HeapTupleHeaderGetRawXmax(oldtup.t_data)))
3496 : 1 : goto l2;
3497 : :
3498 : : /* Otherwise check if it committed or aborted */
3499 : 61 : UpdateXmaxHintBits(oldtup.t_data, buffer, xwait);
3500 [ + + ]: 61 : if (oldtup.t_data->t_infomask & HEAP_XMAX_INVALID)
4099 3501 : 12 : can_continue = true;
3502 : : }
3503 : :
1849 andres@anarazel.de 3504 [ + + ]: 35903 : if (can_continue)
3505 : 35847 : result = TM_Ok;
1147 alvherre@alvh.no-ip. 3506 [ + + ]: 56 : else if (!ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid))
1849 andres@anarazel.de 3507 : 51 : result = TM_Updated;
3508 : : else
3509 : 5 : result = TM_Deleted;
3510 : : }
3511 : :
3512 : : /* Sanity check the result HeapTupleSatisfiesUpdate() and the logic above */
3513 [ + + ]: 277144 : if (result != TM_Ok)
3514 : : {
3515 [ + + + + : 153 : Assert(result == TM_SelfModified ||
- + - - ]
3516 : : result == TM_Updated ||
3517 : : result == TM_Deleted ||
3518 : : result == TM_BeingModified);
6812 tgl@sss.pgh.pa.us 3519 [ - + ]: 153 : Assert(!(oldtup.t_data->t_infomask & HEAP_XMAX_INVALID));
1849 andres@anarazel.de 3520 [ + + - + ]: 153 : Assert(result != TM_Updated ||
3521 : : !ItemPointerEquals(&oldtup.t_self, &oldtup.t_data->t_ctid));
3522 : : }
3523 : :
138 heikki.linnakangas@i 3524 [ + + + - ]: 277144 : if (crosscheck != InvalidSnapshot && result == TM_Ok)
3525 : : {
3526 : : /* Perform additional check for transaction-snapshot mode RI updates */
3527 [ + - ]: 1 : if (!HeapTupleSatisfiesVisibility(&oldtup, crosscheck, buffer))
3528 : 1 : result = TM_Updated;
3529 : : }
3530 : :
3531 [ + + ]: 277144 : if (result != TM_Ok)
3532 : : {
1849 andres@anarazel.de 3533 : 154 : tmfd->ctid = oldtup.t_data->t_ctid;
3534 [ + + + + : 154 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(oldtup.t_data);
+ - ]
3535 [ + + ]: 154 : if (result == TM_SelfModified)
3536 : 52 : tmfd->cmax = HeapTupleHeaderGetCmax(oldtup.t_data);
3537 : : else
3538 : 102 : tmfd->cmax = InvalidCommandId;
3 akorotkov@postgresql 3539 : 154 : UnlockReleaseBuffer(buffer);
6924 tgl@sss.pgh.pa.us 3540 [ + + ]: 154 : if (have_tuple_lock)
2194 simon@2ndQuadrant.co 3541 : 49 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
4681 rhaas@postgresql.org 3542 [ - + ]: 154 : if (vmbuffer != InvalidBuffer)
4681 rhaas@postgresql.org 3543 :UBC 0 : ReleaseBuffer(vmbuffer);
391 tomas.vondra@postgre 3544 :CBC 154 : *update_indexes = TU_None;
3545 : :
6051 tgl@sss.pgh.pa.us 3546 : 154 : bms_free(hot_attrs);
391 tomas.vondra@postgre 3547 : 154 : bms_free(sum_attrs);
4099 alvherre@alvh.no-ip. 3548 : 154 : bms_free(key_attrs);
2790 tgl@sss.pgh.pa.us 3549 : 154 : bms_free(id_attrs);
2573 alvherre@alvh.no-ip. 3550 : 154 : bms_free(modified_attrs);
3551 : 154 : bms_free(interesting_attrs);
9252 vadim4o@yahoo.com 3552 : 154 : return result;
3553 : : }
3554 : :
3555 : : /*
3556 : : * If we didn't pin the visibility map page and the page has become all
3557 : : * visible while we were busy locking the buffer, or during some
3558 : : * subsequent window during which we had it unlocked, we'll have to unlock
3559 : : * and re-lock, to avoid holding the buffer lock across an I/O. That's a
3560 : : * bit unfortunate, especially since we'll now have to recheck whether the
3561 : : * tuple has been locked or updated under us, but hopefully it won't
3562 : : * happen very often.
3563 : : */
4675 rhaas@postgresql.org 3564 [ + + - + ]: 276990 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
3565 : : {
4675 rhaas@postgresql.org 3566 :UBC 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3567 : 0 : visibilitymap_pin(relation, block, &vmbuffer);
3568 : 0 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
4583 3569 : 0 : goto l2;
3570 : : }
3571 : :
3572 : : /* Fill in transaction status data */
3573 : :
3574 : : /*
3575 : : * If the tuple we're updating is locked, we need to preserve the locking
3576 : : * info in the old tuple's Xmax. Prepare a new Xmax value for this.
3577 : : */
4099 alvherre@alvh.no-ip. 3578 :CBC 276990 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3579 : 276990 : oldtup.t_data->t_infomask,
3580 : 276990 : oldtup.t_data->t_infomask2,
3581 : : xid, *lockmode, true,
3582 : : &xmax_old_tuple, &infomask_old_tuple,
3583 : : &infomask2_old_tuple);
3584 : :
3585 : : /*
3586 : : * And also prepare an Xmax value for the new copy of the tuple. If there
3587 : : * was no xmax previously, or there was one but all lockers are now gone,
3588 : : * then use InvalidTransactionId; otherwise, get the xmax from the old
3589 : : * tuple. (In rare cases that might also be InvalidTransactionId and yet
3590 : : * not have the HEAP_XMAX_INVALID bit set; that's fine.)
3591 : : */
3592 [ + + ]: 276990 : if ((oldtup.t_data->t_infomask & HEAP_XMAX_INVALID) ||
2851 3593 [ + + + + : 35835 : HEAP_LOCKED_UPGRADED(oldtup.t_data->t_infomask) ||
+ - + + ]
4099 3594 [ - + ]: 35783 : (checked_lockers && !locker_remains))
3595 : 241155 : xmax_new_tuple = InvalidTransactionId;
3596 : : else
3597 : 35835 : xmax_new_tuple = HeapTupleHeaderGetRawXmax(oldtup.t_data);
3598 : :
3599 [ + + ]: 276990 : if (!TransactionIdIsValid(xmax_new_tuple))
3600 : : {
3601 : 241155 : infomask_new_tuple = HEAP_XMAX_INVALID;
3602 : 241155 : infomask2_new_tuple = 0;
3603 : : }
3604 : : else
3605 : : {
3606 : : /*
3607 : : * If we found a valid Xmax for the new tuple, then the infomask bits
3608 : : * to use on the new tuple depend on what was there on the old one.
3609 : : * Note that since we're doing an update, the only possibility is that
3610 : : * the lockers had FOR KEY SHARE lock.
3611 : : */
3612 [ + + ]: 35835 : if (oldtup.t_data->t_infomask & HEAP_XMAX_IS_MULTI)
3613 : : {
3614 : 53 : GetMultiXactIdHintBits(xmax_new_tuple, &infomask_new_tuple,
3615 : : &infomask2_new_tuple);
3616 : : }
3617 : : else
3618 : : {
3619 : 35782 : infomask_new_tuple = HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_LOCK_ONLY;
3620 : 35782 : infomask2_new_tuple = 0;
3621 : : }
3622 : : }
3623 : :
3624 : : /*
3625 : : * Prepare the new tuple with the appropriate initial values of Xmin and
3626 : : * Xmax, as well as initial infomask bits as computed above.
3627 : : */
9270 vadim4o@yahoo.com 3628 : 276990 : newtup->t_data->t_infomask &= ~(HEAP_XACT_MASK);
6051 tgl@sss.pgh.pa.us 3629 : 276990 : newtup->t_data->t_infomask2 &= ~(HEAP2_XACT_MASK);
7150 3630 : 276990 : HeapTupleHeaderSetXmin(newtup->t_data, xid);
7974 bruce@momjian.us 3631 [ - + ]: 276990 : HeapTupleHeaderSetCmin(newtup->t_data, cid);
4099 alvherre@alvh.no-ip. 3632 : 276990 : newtup->t_data->t_infomask |= HEAP_UPDATED | infomask_new_tuple;
3633 : 276990 : newtup->t_data->t_infomask2 |= infomask2_new_tuple;
3634 : 276990 : HeapTupleHeaderSetXmax(newtup->t_data, xmax_new_tuple);
3635 : :
3636 : : /*
3637 : : * Replace cid with a combo CID if necessary. Note that we already put
3638 : : * the plain cid into the new tuple.
3639 : : */
6274 tgl@sss.pgh.pa.us 3640 : 276990 : HeapTupleHeaderAdjustCmax(oldtup.t_data, &cid, &iscombo);
3641 : :
3642 : : /*
3643 : : * If the toaster needs to be activated, OR if the new tuple will not fit
3644 : : * on the same page as the old, then we need to release the content lock
3645 : : * (but not the pin!) on the old tuple's buffer while we are off doing
3646 : : * TOAST and/or table-file-extension work. We must mark the old tuple to
3647 : : * show that it's locked, else other processes may try to update it
3648 : : * themselves.
3649 : : *
3650 : : * We need to invoke the toaster if there are already any out-of-line
3651 : : * toasted values present, or if the new tuple is over-threshold.
3652 : : */
4060 kgrittn@postgresql.o 3653 [ - + ]: 276990 : if (relation->rd_rel->relkind != RELKIND_RELATION &&
4060 kgrittn@postgresql.o 3654 [ # # ]:UBC 0 : relation->rd_rel->relkind != RELKIND_MATVIEW)
3655 : : {
3656 : : /* toast table entries should never be recursively toasted */
6221 tgl@sss.pgh.pa.us 3657 [ # # ]: 0 : Assert(!HeapTupleHasExternal(&oldtup));
3658 [ # # ]: 0 : Assert(!HeapTupleHasExternal(newtup));
3659 : 0 : need_toast = false;
3660 : : }
3661 : : else
6221 tgl@sss.pgh.pa.us 3662 :CBC 276990 : need_toast = (HeapTupleHasExternal(&oldtup) ||
3663 [ + + + + ]: 553695 : HeapTupleHasExternal(newtup) ||
3664 [ + + ]: 276705 : newtup->t_len > TOAST_TUPLE_THRESHOLD);
3665 : :
5754 3666 : 276990 : pagefree = PageGetHeapFreeSpace(page);
3667 : :
6279 3668 : 276990 : newtupsize = MAXALIGN(newtup->t_len);
3669 : :
8369 3670 [ + + + + ]: 276990 : if (need_toast || newtupsize > pagefree)
8620 vadim4o@yahoo.com 3671 : 140457 : {
3672 : : TransactionId xmax_lock_old_tuple;
3673 : : uint16 infomask_lock_old_tuple,
3674 : : infomask2_lock_old_tuple;
2827 andres@anarazel.de 3675 : 140457 : bool cleared_all_frozen = false;
3676 : :
3677 : : /*
3678 : : * To prevent concurrent sessions from updating the tuple, we have to
3679 : : * temporarily mark it locked, while we release the page-level lock.
3680 : : *
3681 : : * To satisfy the rule that any xid potentially appearing in a buffer
3682 : : * written out to disk, we unfortunately have to WAL log this
3683 : : * temporary modification. We can reuse xl_heap_lock for this
3684 : : * purpose. If we crash/error before following through with the
3685 : : * actual update, xmax will be of an aborted transaction, allowing
3686 : : * other sessions to proceed.
3687 : : */
3688 : :
3689 : : /*
3690 : : * Compute xmax / infomask appropriate for locking the tuple. This has
3691 : : * to be done separately from the combo that's going to be used for
3692 : : * updating, because the potentially created multixact would otherwise
3693 : : * be wrong.
3694 : : */
2830 3695 : 140457 : compute_new_xmax_infomask(HeapTupleHeaderGetRawXmax(oldtup.t_data),
3696 : 140457 : oldtup.t_data->t_infomask,
3697 : 140457 : oldtup.t_data->t_infomask2,
3698 : : xid, *lockmode, false,
3699 : : &xmax_lock_old_tuple, &infomask_lock_old_tuple,
3700 : : &infomask2_lock_old_tuple);
3701 : :
3702 [ - + - - ]: 140457 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(infomask_lock_old_tuple));
3703 : :
3704 : 140457 : START_CRIT_SECTION();
3705 : :
3706 : : /* Clear obsolete visibility flags ... */
4099 alvherre@alvh.no-ip. 3707 : 140457 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3708 : 140457 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6051 tgl@sss.pgh.pa.us 3709 : 140457 : HeapTupleClearHotUpdated(&oldtup);
3710 : : /* ... and store info about transaction updating this tuple */
2830 andres@anarazel.de 3711 [ - + ]: 140457 : Assert(TransactionIdIsValid(xmax_lock_old_tuple));
3712 : 140457 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_lock_old_tuple);
3713 : 140457 : oldtup.t_data->t_infomask |= infomask_lock_old_tuple;
3714 : 140457 : oldtup.t_data->t_infomask2 |= infomask2_lock_old_tuple;
6274 tgl@sss.pgh.pa.us 3715 [ - + + + ]: 140457 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
3716 : :
3717 : : /* temporarily make it look not-updated, but locked */
6720 3718 : 140457 : oldtup.t_data->t_ctid = oldtup.t_self;
3719 : :
3720 : : /*
3721 : : * Clear all-frozen bit on visibility map if needed. We could
3722 : : * immediately reset ALL_VISIBLE, but given that the WAL logging
3723 : : * overhead would be unchanged, that doesn't seem necessarily
3724 : : * worthwhile.
3725 : : */
1097 3726 [ + + + + ]: 141097 : if (PageIsAllVisible(page) &&
2827 andres@anarazel.de 3727 : 640 : visibilitymap_clear(relation, block, vmbuffer,
3728 : : VISIBILITYMAP_ALL_FROZEN))
3729 : 465 : cleared_all_frozen = true;
3730 : :
2830 3731 : 140457 : MarkBufferDirty(buffer);
3732 : :
3733 [ + + + + : 140457 : if (RelationNeedsWAL(relation))
+ - + + ]
3734 : : {
3735 : : xl_heap_lock xlrec;
3736 : : XLogRecPtr recptr;
3737 : :
3738 : 130325 : XLogBeginInsert();
3739 : 130325 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
3740 : :
3741 : 130325 : xlrec.offnum = ItemPointerGetOffsetNumber(&oldtup.t_self);
369 pg@bowt.ie 3742 : 130325 : xlrec.xmax = xmax_lock_old_tuple;
2830 andres@anarazel.de 3743 : 260650 : xlrec.infobits_set = compute_infobits(oldtup.t_data->t_infomask,
3744 : 130325 : oldtup.t_data->t_infomask2);
2827 3745 : 130325 : xlrec.flags =
3746 : 130325 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
2830 3747 : 130325 : XLogRegisterData((char *) &xlrec, SizeOfHeapLock);
3748 : 130325 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
3749 : 130325 : PageSetLSN(page, recptr);
3750 : : }
3751 : :
3752 [ - + ]: 140457 : END_CRIT_SECTION();
3753 : :
8620 vadim4o@yahoo.com 3754 : 140457 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3755 : :
3756 : : /*
3757 : : * Let the toaster do its thing, if needed.
3758 : : *
3759 : : * Note: below this point, heaptup is the data we actually intend to
3760 : : * store into the relation; newtup is the caller's original untoasted
3761 : : * data.
3762 : : */
8490 tgl@sss.pgh.pa.us 3763 [ + + ]: 140457 : if (need_toast)
3764 : : {
3765 : : /* Note we always use WAL and FSM during updates */
1654 rhaas@postgresql.org 3766 : 1096 : heaptup = heap_toast_insert_or_update(relation, newtup, &oldtup, 0);
6720 tgl@sss.pgh.pa.us 3767 : 1096 : newtupsize = MAXALIGN(heaptup->t_len);
3768 : : }
3769 : : else
3770 : 139361 : heaptup = newtup;
3771 : :
3772 : : /*
3773 : : * Now, do we need a new page for the tuple, or not? This is a bit
3774 : : * tricky since someone else could have added tuples to the page while
3775 : : * we weren't looking. We have to recheck the available space after
3776 : : * reacquiring the buffer lock. But don't bother to do that if the
3777 : : * former amount of free space is still not enough; it's unlikely
3778 : : * there's more free now than before.
3779 : : *
3780 : : * What's more, if we need to get a new page, we will need to acquire
3781 : : * buffer locks on both old and new pages. To avoid deadlock against
3782 : : * some other backend trying to get the same two locks in the other
3783 : : * order, we must be consistent about the order we get the locks in.
3784 : : * We use the rule "lock the lower-numbered page of the relation
3785 : : * first". To implement this, we must do RelationGetBufferForTuple
3786 : : * while not holding the lock on the old page, and we must rely on it
3787 : : * to get the locks on both pages in the correct order.
3788 : : *
3789 : : * Another consideration is that we need visibility map page pin(s) if
3790 : : * we will have to clear the all-visible flag on either page. If we
3791 : : * call RelationGetBufferForTuple, we rely on it to acquire any such
3792 : : * pins; but if we don't, we have to handle that here. Hence we need
3793 : : * a loop.
3794 : : */
3795 : : for (;;)
3796 : : {
1097 3797 [ + + ]: 140457 : if (newtupsize > pagefree)
3798 : : {
3799 : : /* It doesn't fit, must use RelationGetBufferForTuple. */
3800 : 140150 : newbuf = RelationGetBufferForTuple(relation, heaptup->t_len,
3801 : : buffer, 0, NULL,
3802 : : &vmbuffer_new, &vmbuffer,
3803 : : 0);
3804 : : /* We're all done. */
3805 : 140150 : break;
3806 : : }
3807 : : /* Acquire VM page pin if needed and we don't have it. */
3808 [ + + - + ]: 307 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
1097 tgl@sss.pgh.pa.us 3809 :UBC 0 : visibilitymap_pin(relation, block, &vmbuffer);
3810 : : /* Re-acquire the lock on the old tuple's page. */
8369 tgl@sss.pgh.pa.us 3811 :CBC 307 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
3812 : : /* Re-check using the up-to-date free space */
5754 3813 : 307 : pagefree = PageGetHeapFreeSpace(page);
1097 3814 [ + - ]: 307 : if (newtupsize > pagefree ||
3815 [ + + - + ]: 307 : (vmbuffer == InvalidBuffer && PageIsAllVisible(page)))
3816 : : {
3817 : : /*
3818 : : * Rats, it doesn't fit anymore, or somebody just now set the
3819 : : * all-visible flag. We must now unlock and loop to avoid
3820 : : * deadlock. Fortunately, this path should seldom be taken.
3821 : : */
8369 tgl@sss.pgh.pa.us 3822 :UBC 0 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
3823 : : }
3824 : : else
3825 : : {
3826 : : /* We're all done. */
8369 tgl@sss.pgh.pa.us 3827 :CBC 307 : newbuf = buffer;
1097 3828 : 307 : break;
3829 : : }
3830 : : }
3831 : : }
3832 : : else
3833 : : {
3834 : : /* No TOAST work needed, and it'll fit on same page */
8490 3835 : 136533 : newbuf = buffer;
6720 3836 : 136533 : heaptup = newtup;
3837 : : }
3838 : :
3839 : : /*
3840 : : * We're about to do the actual update -- check for conflict first, to
3841 : : * avoid possibly having to roll back work we've just done.
3842 : : *
3843 : : * This is safe without a recheck as long as there is no possibility of
3844 : : * another process scanning the pages between this check and the update
3845 : : * being visible to the scan (i.e., exclusive buffer content lock(s) are
3846 : : * continuously held from this point until the tuple update is visible).
3847 : : *
3848 : : * For the new tuple the only check needed is at the relation level, but
3849 : : * since both tuples are in the same relation and the check for oldtup
3850 : : * will include checking the relation level, there is no benefit to a
3851 : : * separate check for the new tuple.
3852 : : */
1097 tmunro@postgresql.or 3853 : 276990 : CheckForSerializableConflictIn(relation, &oldtup.t_self,
3854 : : BufferGetBlockNumber(buffer));
3855 : :
3856 : : /*
3857 : : * At this point newbuf and buffer are both pinned and locked, and newbuf
3858 : : * has enough space for the new tuple. If they are the same buffer, only
3859 : : * one pin is held.
3860 : : */
3861 : :
6051 tgl@sss.pgh.pa.us 3862 [ + + ]: 276978 : if (newbuf == buffer)
3863 : : {
3864 : : /*
3865 : : * Since the new tuple is going into the same page, we might be able
3866 : : * to do a HOT update. Check if any of the index columns have been
3867 : : * changed.
3868 : : */
870 pg@bowt.ie 3869 [ + + ]: 136828 : if (!bms_overlap(modified_attrs, hot_attrs))
3870 : : {
6051 tgl@sss.pgh.pa.us 3871 : 126107 : use_hot_update = true;
3872 : :
3873 : : /*
3874 : : * If none of the columns that are used in hot-blocking indexes
3875 : : * were updated, we can apply HOT, but we do still need to check
3876 : : * if we need to update the summarizing indexes, and update those
3877 : : * indexes if the columns were updated, or we may fail to detect
3878 : : * e.g. value bound changes in BRIN minmax indexes.
3879 : : */
391 tomas.vondra@postgre 3880 [ + + ]: 126107 : if (bms_overlap(modified_attrs, sum_attrs))
3881 : 1641 : summarized_update = true;
3882 : : }
3883 : : }
3884 : : else
3885 : : {
3886 : : /* Set a hint that the old page could use prune/defrag */
5754 tgl@sss.pgh.pa.us 3887 : 140150 : PageSetFull(page);
3888 : : }
3889 : :
3890 : : /*
3891 : : * Compute replica identity tuple before entering the critical section so
3892 : : * we don't PANIC upon a memory allocation failure.
3893 : : * ExtractReplicaIdentity() will return NULL if nothing needs to be
3894 : : * logged. Pass old key required as true only if the replica identity key
3895 : : * columns are modified or it has external data.
3896 : : */
2573 alvherre@alvh.no-ip. 3897 : 276978 : old_key_tuple = ExtractReplicaIdentity(relation, &oldtup,
790 akapila@postgresql.o 3898 [ + + + + ]: 276978 : bms_overlap(modified_attrs, id_attrs) ||
3899 : : id_has_external,
2573 alvherre@alvh.no-ip. 3900 :ECB (424274) : &old_key_copied);
3901 : :
3902 : : /* NO EREPORT(ERROR) from here till changes are logged */
8493 tgl@sss.pgh.pa.us 3903 :CBC 276978 : START_CRIT_SECTION();
3904 : :
3905 : : /*
3906 : : * If this transaction commits, the old tuple will become DEAD sooner or
3907 : : * later. Set flag that this page is a candidate for pruning once our xid
3908 : : * falls below the OldestXmin horizon. If the transaction finally aborts,
3909 : : * the subsequent page pruning will be a no-op and the hint will be
3910 : : * cleared.
3911 : : *
3912 : : * XXX Should we set hint on newbuf as well? If the transaction aborts,
3913 : : * there would be a prunable tuple in the newbuf; but for now we choose
3914 : : * not to optimize for aborts. Note that heap_xlog_update must be kept in
3915 : : * sync if this decision changes.
3916 : : */
5754 3917 [ - + + + : 276978 : PageSetPrunable(page, xid);
+ + ]
3918 : :
6051 3919 [ + + ]: 276978 : if (use_hot_update)
3920 : : {
3921 : : /* Mark the old tuple as HOT-updated */
3922 : 126107 : HeapTupleSetHotUpdated(&oldtup);
3923 : : /* And mark the new tuple as heap-only */
3924 : 126107 : HeapTupleSetHeapOnly(heaptup);
3925 : : /* Mark the caller's copy too, in case different from heaptup */
3926 : 126107 : HeapTupleSetHeapOnly(newtup);
3927 : : }
3928 : : else
3929 : : {
3930 : : /* Make sure tuples are correctly marked as not-HOT */
3931 : 150871 : HeapTupleClearHotUpdated(&oldtup);
3932 : 150871 : HeapTupleClearHeapOnly(heaptup);
3933 : 150871 : HeapTupleClearHeapOnly(newtup);
3934 : : }
3935 : :
2489 3936 : 276978 : RelationPutHeapTuple(relation, newbuf, heaptup, false); /* insert new tuple */
3937 : :
3938 : :
3939 : : /* Clear obsolete visibility flags, possibly set by ourselves above... */
2830 andres@anarazel.de 3940 : 276978 : oldtup.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
3941 : 276978 : oldtup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3942 : : /* ... and store info about transaction updating this tuple */
3943 [ - + ]: 276978 : Assert(TransactionIdIsValid(xmax_old_tuple));
3944 : 276978 : HeapTupleHeaderSetXmax(oldtup.t_data, xmax_old_tuple);
3945 : 276978 : oldtup.t_data->t_infomask |= infomask_old_tuple;
3946 : 276978 : oldtup.t_data->t_infomask2 |= infomask2_old_tuple;
3947 [ - + + + ]: 276978 : HeapTupleHeaderSetCmax(oldtup.t_data, cid, iscombo);
3948 : :
3949 : : /* record address of new tuple in t_ctid of old one */
6720 tgl@sss.pgh.pa.us 3950 : 276978 : oldtup.t_data->t_ctid = heaptup->t_self;
3951 : :
3952 : : /* clear PD_ALL_VISIBLE flags, reset all visibilitymap bits */
2916 kgrittn@postgresql.o 3953 [ + + ]: 276978 : if (PageIsAllVisible(BufferGetPage(buffer)))
3954 : : {
5347 tgl@sss.pgh.pa.us 3955 : 1176 : all_visible_cleared = true;
2916 kgrittn@postgresql.o 3956 : 1176 : PageClearAllVisible(BufferGetPage(buffer));
4675 rhaas@postgresql.org 3957 : 1176 : visibilitymap_clear(relation, BufferGetBlockNumber(buffer),
3958 : : vmbuffer, VISIBILITYMAP_VALID_BITS);
3959 : : }
2916 kgrittn@postgresql.o 3960 [ + + + + ]: 276978 : if (newbuf != buffer && PageIsAllVisible(BufferGetPage(newbuf)))
3961 : : {
5347 tgl@sss.pgh.pa.us 3962 : 815 : all_visible_cleared_new = true;
2916 kgrittn@postgresql.o 3963 : 815 : PageClearAllVisible(BufferGetPage(newbuf));
4675 rhaas@postgresql.org 3964 : 815 : visibilitymap_clear(relation, BufferGetBlockNumber(newbuf),
3965 : : vmbuffer_new, VISIBILITYMAP_VALID_BITS);
3966 : : }
3967 : :
6589 tgl@sss.pgh.pa.us 3968 [ + + ]: 276978 : if (newbuf != buffer)
3969 : 140150 : MarkBufferDirty(newbuf);
3970 : 276978 : MarkBufferDirty(buffer);
3971 : :
3972 : : /* XLOG stuff */
4871 rhaas@postgresql.org 3973 [ + + + + : 276978 : if (RelationNeedsWAL(relation))
+ + + + ]
3974 : : {
3975 : : XLogRecPtr recptr;
3976 : :
3977 : : /*
3978 : : * For logical decoding we need combo CIDs to properly decode the
3979 : : * catalog.
3980 : : */
3778 3981 [ + + + - : 265654 : if (RelationIsAccessibleInLogicalDecoding(relation))
- + - - -
- + + + +
- + - - +
- ]
3982 : : {
3983 : 2423 : log_heap_new_cid(relation, &oldtup);
3984 : 2423 : log_heap_new_cid(relation, heaptup);
3985 : : }
3986 : :
3987 : 265654 : recptr = log_heap_update(relation, buffer,
3988 : : newbuf, &oldtup, heaptup,
3989 : : old_key_tuple,
3990 : : all_visible_cleared,
3991 : : all_visible_cleared_new);
8686 vadim4o@yahoo.com 3992 [ + + ]: 265654 : if (newbuf != buffer)
3993 : : {
2916 kgrittn@postgresql.o 3994 : 130024 : PageSetLSN(BufferGetPage(newbuf), recptr);
3995 : : }
3996 : 265654 : PageSetLSN(BufferGetPage(buffer), recptr);
3997 : : }
3998 : :
8493 tgl@sss.pgh.pa.us 3999 [ - + ]: 276978 : END_CRIT_SECTION();
4000 : :
8686 vadim4o@yahoo.com 4001 [ + + ]: 276978 : if (newbuf != buffer)
4002 : 140150 : LockBuffer(newbuf, BUFFER_LOCK_UNLOCK);
9252 4003 : 276978 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
4004 : :
4005 : : /*
4006 : : * Mark old tuple for invalidation from system caches at next command
4007 : : * boundary, and mark the new tuple for invalidation in case we abort. We
4008 : : * have to do this before releasing the buffer because oldtup is in the
4009 : : * buffer. (heaptup is all in local memory, but it's necessary to process
4010 : : * both tuple versions in one call to inval.c so we can avoid redundant
4011 : : * sinval messages.)
4012 : : */
4625 tgl@sss.pgh.pa.us 4013 : 276978 : CacheInvalidateHeapTuple(relation, &oldtup, heaptup);
4014 : :
4015 : : /* Now we can release the buffer(s) */
8498 4016 [ + + ]: 276978 : if (newbuf != buffer)
6589 4017 : 140150 : ReleaseBuffer(newbuf);
3 akorotkov@postgresql 4018 : 276978 : ReleaseBuffer(buffer);
4681 rhaas@postgresql.org 4019 [ + + ]: 276978 : if (BufferIsValid(vmbuffer_new))
4020 : 815 : ReleaseBuffer(vmbuffer_new);
4021 [ + + ]: 276978 : if (BufferIsValid(vmbuffer))
4022 : 1176 : ReleaseBuffer(vmbuffer);
4023 : :
4024 : : /*
4025 : : * Release the lmgr tuple lock, if we had it.
4026 : : */
6924 tgl@sss.pgh.pa.us 4027 [ + + ]: 276978 : if (have_tuple_lock)
2194 simon@2ndQuadrant.co 4028 : 12 : UnlockTupleTuplock(relation, &(oldtup.t_self), *lockmode);
4029 : :
388 pg@bowt.ie 4030 : 276978 : pgstat_count_heap_update(relation, use_hot_update, newbuf != buffer);
4031 : :
4032 : : /*
4033 : : * If heaptup is a private copy, release it. Don't forget to copy t_self
4034 : : * back to the caller's image, too.
4035 : : */
6720 tgl@sss.pgh.pa.us 4036 [ + + ]: 276978 : if (heaptup != newtup)
4037 : : {
4038 : 1049 : newtup->t_self = heaptup->t_self;
4039 : 1049 : heap_freetuple(heaptup);
4040 : : }
4041 : :
4042 : : /*
4043 : : * If it is a HOT update, the update may still need to update summarized
4044 : : * indexes, lest we fail to update those summaries and get incorrect
4045 : : * results (for example, minmax bounds of the block may change with this
4046 : : * update).
4047 : : */
391 tomas.vondra@postgre 4048 [ + + ]: 276978 : if (use_hot_update)
4049 : : {
4050 [ + + ]: 126107 : if (summarized_update)
4051 : 1641 : *update_indexes = TU_Summarizing;
4052 : : else
4053 : 124466 : *update_indexes = TU_None;
4054 : : }
4055 : : else
4056 : 150871 : *update_indexes = TU_All;
4057 : :
3778 rhaas@postgresql.org 4058 [ + + + + ]: 276978 : if (old_key_tuple != NULL && old_key_copied)
4059 : 80 : heap_freetuple(old_key_tuple);
4060 : :
6051 tgl@sss.pgh.pa.us 4061 : 276978 : bms_free(hot_attrs);
391 tomas.vondra@postgre 4062 : 276978 : bms_free(sum_attrs);
4099 alvherre@alvh.no-ip. 4063 : 276978 : bms_free(key_attrs);
2790 tgl@sss.pgh.pa.us 4064 : 276978 : bms_free(id_attrs);
2573 alvherre@alvh.no-ip. 4065 : 276978 : bms_free(modified_attrs);
4066 : 276978 : bms_free(interesting_attrs);
4067 : :
1849 andres@anarazel.de 4068 : 276978 : return TM_Ok;
4069 : : }
4070 : :
4071 : : /*
4072 : : * Check if the specified attribute's values are the same. Subroutine for
4073 : : * HeapDetermineColumnsInfo.
4074 : : */
4075 : : static bool
790 akapila@postgresql.o 4076 : 619520 : heap_attr_equals(TupleDesc tupdesc, int attrnum, Datum value1, Datum value2,
4077 : : bool isnull1, bool isnull2)
4078 : : {
4079 : : Form_pg_attribute att;
4080 : :
4081 : : /*
4082 : : * If one value is NULL and other is not, then they are certainly not
4083 : : * equal
4084 : : */
6051 tgl@sss.pgh.pa.us 4085 [ + + ]: 619520 : if (isnull1 != isnull2)
4086 : 3 : return false;
4087 : :
4088 : : /*
4089 : : * If both are NULL, they can be considered equal.
4090 : : */
4091 [ + + ]: 619517 : if (isnull1)
4092 : 4991 : return true;
4093 : :
4094 : : /*
4095 : : * We do simple binary comparison of the two datums. This may be overly
4096 : : * strict because there can be multiple binary representations for the
4097 : : * same logical value. But we should be OK as long as there are no false
4098 : : * positives. Using a type-specific equality operator is messy because
4099 : : * there could be multiple notions of equality in different operator
4100 : : * classes; furthermore, we cannot safely invoke user-defined functions
4101 : : * while holding exclusive buffer lock.
4102 : : */
4103 [ - + ]: 614526 : if (attrnum <= 0)
4104 : : {
4105 : : /* The only allowed system columns are OIDs, so do this */
6051 tgl@sss.pgh.pa.us 4106 :UBC 0 : return (DatumGetObjectId(value1) == DatumGetObjectId(value2));
4107 : : }
4108 : : else
4109 : : {
6051 tgl@sss.pgh.pa.us 4110 [ - + ]:CBC 614526 : Assert(attrnum <= tupdesc->natts);
2429 andres@anarazel.de 4111 : 614526 : att = TupleDescAttr(tupdesc, attrnum - 1);
6051 tgl@sss.pgh.pa.us 4112 : 614526 : return datumIsEqual(value1, value2, att->attbyval, att->attlen);
4113 : : }
4114 : : }
4115 : :
4116 : : /*
4117 : : * Check which columns are being updated.
4118 : : *
4119 : : * Given an updated tuple, determine (and return into the output bitmapset),
4120 : : * from those listed as interesting, the set of columns that changed.
4121 : : *
4122 : : * has_external indicates if any of the unmodified attributes (from those
4123 : : * listed as interesting) of the old tuple is a member of external_cols and is
4124 : : * stored externally.
4125 : : */
4126 : : static Bitmapset *
790 akapila@postgresql.o 4127 : 277144 : HeapDetermineColumnsInfo(Relation relation,
4128 : : Bitmapset *interesting_cols,
4129 : : Bitmapset *external_cols,
4130 : : HeapTuple oldtup, HeapTuple newtup,
4131 : : bool *has_external)
4132 : : {
4133 : : int attidx;
2524 bruce@momjian.us 4134 : 277144 : Bitmapset *modified = NULL;
790 akapila@postgresql.o 4135 : 277144 : TupleDesc tupdesc = RelationGetDescr(relation);
4136 : :
409 tgl@sss.pgh.pa.us 4137 : 277144 : attidx = -1;
4138 [ + + ]: 896664 : while ((attidx = bms_next_member(interesting_cols, attidx)) >= 0)
4139 : : {
4140 : : /* attidx is zero-based, attrnum is the normal attribute number */
4141 : 619520 : AttrNumber attrnum = attidx + FirstLowInvalidHeapAttributeNumber;
4142 : : Datum value1,
4143 : : value2;
4144 : : bool isnull1,
4145 : : isnull2;
4146 : :
4147 : : /*
4148 : : * If it's a whole-tuple reference, say "not equal". It's not really
4149 : : * worth supporting this case, since it could only succeed after a
4150 : : * no-op update, which is hardly a case worth optimizing for.
4151 : : */
790 akapila@postgresql.o 4152 [ - + ]: 619520 : if (attrnum == 0)
4153 : : {
409 tgl@sss.pgh.pa.us 4154 :UBC 0 : modified = bms_add_member(modified, attidx);
790 akapila@postgresql.o 4155 : 0 : continue;
4156 : : }
4157 : :
4158 : : /*
4159 : : * Likewise, automatically say "not equal" for any system attribute
4160 : : * other than tableOID; we cannot expect these to be consistent in a
4161 : : * HOT chain, or even to be set correctly yet in the new tuple.
4162 : : */
790 akapila@postgresql.o 4163 [ - + ]:CBC 619520 : if (attrnum < 0)
4164 : : {
790 akapila@postgresql.o 4165 [ # # ]:UBC 0 : if (attrnum != TableOidAttributeNumber)
4166 : : {
409 tgl@sss.pgh.pa.us 4167 : 0 : modified = bms_add_member(modified, attidx);
790 akapila@postgresql.o 4168 : 0 : continue;
4169 : : }
4170 : : }
4171 : :
4172 : : /*
4173 : : * Extract the corresponding values. XXX this is pretty inefficient
4174 : : * if there are many indexed columns. Should we do a single
4175 : : * heap_deform_tuple call on each tuple, instead? But that doesn't
4176 : : * work for system columns ...
4177 : : */
790 akapila@postgresql.o 4178 :CBC 619520 : value1 = heap_getattr(oldtup, attrnum, tupdesc, &isnull1);
4179 : 619520 : value2 = heap_getattr(newtup, attrnum, tupdesc, &isnull2);
4180 : :
4181 [ + + ]: 619520 : if (!heap_attr_equals(tupdesc, attrnum, value1,
4182 : : value2, isnull1, isnull2))
4183 : : {
409 tgl@sss.pgh.pa.us 4184 : 25188 : modified = bms_add_member(modified, attidx);
790 akapila@postgresql.o 4185 : 25188 : continue;
4186 : : }
4187 : :
4188 : : /*
4189 : : * No need to check attributes that can't be stored externally. Note
4190 : : * that system attributes can't be stored externally.
4191 : : */
4192 [ + - + + ]: 594332 : if (attrnum < 0 || isnull1 ||
4193 [ + + ]: 589341 : TupleDescAttr(tupdesc, attrnum - 1)->attlen != -1)
4194 : 565087 : continue;
4195 : :
4196 : : /*
4197 : : * Check if the old tuple's attribute is stored externally and is a
4198 : : * member of external_cols.
4199 : : */
4200 [ + + + + ]: 29250 : if (VARATT_IS_EXTERNAL((struct varlena *) DatumGetPointer(value1)) &&
409 tgl@sss.pgh.pa.us 4201 : 5 : bms_is_member(attidx, external_cols))
790 akapila@postgresql.o 4202 : 2 : *has_external = true;
4203 : : }
4204 : :
2573 alvherre@alvh.no-ip. 4205 : 277144 : return modified;
4206 : : }
4207 : :
4208 : : /*
4209 : : * simple_heap_update - replace a tuple
4210 : : *
4211 : : * This routine may be used to update a tuple when concurrent updates of
4212 : : * the target tuple are not expected (for example, because we have a lock
4213 : : * on the relation associated with the tuple). Any failure is reported
4214 : : * via ereport().
4215 : : */
4216 : : void
391 tomas.vondra@postgre 4217 : 88998 : simple_heap_update(Relation relation, ItemPointer otid, HeapTuple tup,
4218 : : TU_UpdateIndexes *update_indexes)
4219 : : {
4220 : : TM_Result result;
4221 : : TM_FailureData tmfd;
4222 : : LockTupleMode lockmode;
4223 : :
7517 tgl@sss.pgh.pa.us 4224 : 88998 : result = heap_update(relation, otid, tup,
4225 : : GetCurrentCommandId(true), InvalidSnapshot,
4226 : : true /* wait for commit */ ,
4227 : : &tmfd, &lockmode, update_indexes);
8482 4228 [ - + - - : 88998 : switch (result)
- ]
4229 : : {
1849 andres@anarazel.de 4230 :UBC 0 : case TM_SelfModified:
4231 : : /* Tuple was already updated in current command? */
7573 tgl@sss.pgh.pa.us 4232 [ # # ]: 0 : elog(ERROR, "tuple already updated by self");
4233 : : break;
4234 : :
1849 andres@anarazel.de 4235 :CBC 88998 : case TM_Ok:
4236 : : /* done successfully */
8482 tgl@sss.pgh.pa.us 4237 : 88998 : break;
4238 : :
1849 andres@anarazel.de 4239 :UBC 0 : case TM_Updated:
7573 tgl@sss.pgh.pa.us 4240 [ # # ]: 0 : elog(ERROR, "tuple concurrently updated");
4241 : : break;
4242 : :
1849 andres@anarazel.de 4243 : 0 : case TM_Deleted:
4244 [ # # ]: 0 : elog(ERROR, "tuple concurrently deleted");
4245 : : break;
4246 : :
8482 tgl@sss.pgh.pa.us 4247 : 0 : default:
7573 4248 [ # # ]: 0 : elog(ERROR, "unrecognized heap_update status: %u", result);
4249 : : break;
4250 : : }
8482 tgl@sss.pgh.pa.us 4251 :CBC 88998 : }
4252 : :
4253 : :
4254 : : /*
4255 : : * Return the MultiXactStatus corresponding to the given tuple lock mode.
4256 : : */
4257 : : static MultiXactStatus
4099 alvherre@alvh.no-ip. 4258 : 1196 : get_mxact_status_for_lock(LockTupleMode mode, bool is_update)
4259 : : {
4260 : : int retval;
4261 : :
4262 [ + + ]: 1196 : if (is_update)
4263 : 96 : retval = tupleLockExtraInfo[mode].updstatus;
4264 : : else
4265 : 1100 : retval = tupleLockExtraInfo[mode].lockstatus;
4266 : :
4267 [ - + ]: 1196 : if (retval == -1)
4099 alvherre@alvh.no-ip. 4268 [ # # # # ]:UBC 0 : elog(ERROR, "invalid lock tuple mode %d/%s", mode,
4269 : : is_update ? "true" : "false");
4270 : :
3946 alvherre@alvh.no-ip. 4271 :CBC 1196 : return (MultiXactStatus) retval;
4272 : : }
4273 : :
4274 : : /*
4275 : : * heap_lock_tuple - lock a tuple in shared or exclusive mode
4276 : : *
4277 : : * Note that this acquires a buffer pin, which the caller must release.
4278 : : *
4279 : : * Input parameters:
4280 : : * relation: relation containing tuple (caller must hold suitable lock)
4281 : : * tid: TID of tuple to lock
4282 : : * cid: current command ID (used for visibility test, and stored into
4283 : : * tuple's cmax if lock is successful)
4284 : : * mode: indicates if shared or exclusive tuple lock is desired
4285 : : * wait_policy: what to do if tuple lock is not available
4286 : : * follow_updates: if true, follow the update chain to also lock descendant
4287 : : * tuples.
4288 : : *
4289 : : * Output parameters:
4290 : : * *tuple: all fields filled in
4291 : : * *buffer: set to buffer holding tuple (pinned but not locked at exit)
4292 : : * *tmfd: filled in failure cases (see below)
4293 : : *
4294 : : * Function results are the same as the ones for table_tuple_lock().
4295 : : *
4296 : : * In the failure cases other than TM_Invisible, the routine fills
4297 : : * *tmfd with the tuple's t_ctid, t_xmax (resolving a possible MultiXact,
4298 : : * if necessary), and t_cmax (the last only for TM_SelfModified,
4299 : : * since we cannot obtain cmax from a combo CID generated by another
4300 : : * transaction).
4301 : : * See comments for struct TM_FailureData for additional info.
4302 : : *
4303 : : * See README.tuplock for a thorough explanation of this mechanism.
4304 : : */
4305 : : TM_Result
3 akorotkov@postgresql 4306 : 82709 : heap_lock_tuple(Relation relation, HeapTuple tuple,
4307 : : CommandId cid, LockTupleMode mode, LockWaitPolicy wait_policy,
4308 : : bool follow_updates,
4309 : : Buffer *buffer, TM_FailureData *tmfd)
4310 : : {
4311 : : TM_Result result;
4312 : 82709 : ItemPointer tid = &(tuple->t_self);
4313 : : ItemId lp;
4314 : : Page page;
2827 andres@anarazel.de 4315 : 82709 : Buffer vmbuffer = InvalidBuffer;
4316 : : BlockNumber block;
4317 : : TransactionId xid,
4318 : : xmax;
4319 : : uint16 old_infomask,
4320 : : new_infomask,
4321 : : new_infomask2;
3292 alvherre@alvh.no-ip. 4322 : 82709 : bool first_time = true;
1762 4323 : 82709 : bool skip_tuple_lock = false;
6924 tgl@sss.pgh.pa.us 4324 : 82709 : bool have_tuple_lock = false;
2827 andres@anarazel.de 4325 : 82709 : bool cleared_all_frozen = false;
4326 : :
3 akorotkov@postgresql 4327 : 82709 : *buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
2827 andres@anarazel.de 4328 : 82709 : block = ItemPointerGetBlockNumber(tid);
4329 : :
4330 : : /*
4331 : : * Before locking the buffer, pin the visibility map page if it appears to
4332 : : * be necessary. Since we haven't got the lock yet, someone else might be
4333 : : * in the middle of changing this, so we'll need to recheck after we have
4334 : : * the lock.
4335 : : */
3 akorotkov@postgresql 4336 [ + + ]: 82709 : if (PageIsAllVisible(BufferGetPage(*buffer)))
2827 andres@anarazel.de 4337 : 1658 : visibilitymap_pin(relation, block, &vmbuffer);
4338 : :
3 akorotkov@postgresql 4339 : 82709 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4340 : :
4341 : 82709 : page = BufferGetPage(*buffer);
5754 tgl@sss.pgh.pa.us 4342 : 82709 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
6059 4343 [ - + ]: 82709 : Assert(ItemIdIsNormal(lp));
4344 : :
5754 4345 : 82709 : tuple->t_data = (HeapTupleHeader) PageGetItem(page, lp);
9252 vadim4o@yahoo.com 4346 : 82709 : tuple->t_len = ItemIdGetLength(lp);
6812 tgl@sss.pgh.pa.us 4347 : 82709 : tuple->t_tableOid = RelationGetRelid(relation);
4348 : :
9252 vadim4o@yahoo.com 4349 : 13 : l3:
3 akorotkov@postgresql 4350 : 82722 : result = HeapTupleSatisfiesUpdate(tuple, cid, *buffer);
4351 : :
1849 andres@anarazel.de 4352 [ + + ]: 82722 : if (result == TM_Invisible)
4353 : : {
4354 : : /*
4355 : : * This is possible, but only when locking a tuple for ON CONFLICT
4356 : : * UPDATE. We return this value here rather than throwing an error in
4357 : : * order to give that case the opportunity to throw a more specific
4358 : : * error.
4359 : : */
4360 : 12 : result = TM_Invisible;
2827 4361 : 12 : goto out_locked;
4362 : : }
1849 4363 [ + + + + ]: 82710 : else if (result == TM_BeingModified ||
4364 [ + + ]: 75903 : result == TM_Updated ||
4365 : : result == TM_Deleted)
4366 : : {
4367 : : TransactionId xwait;
4368 : : uint16 infomask;
4369 : : uint16 infomask2;
4370 : : bool require_sleep;
4371 : : ItemPointerData t_ctid;
4372 : :
4373 : : /* must copy state data before unlocking buffer */
4099 alvherre@alvh.no-ip. 4374 : 6808 : xwait = HeapTupleHeaderGetRawXmax(tuple->t_data);
6924 tgl@sss.pgh.pa.us 4375 : 6808 : infomask = tuple->t_data->t_infomask;
4099 alvherre@alvh.no-ip. 4376 : 6808 : infomask2 = tuple->t_data->t_infomask2;
4377 : 6808 : ItemPointerCopy(&tuple->t_data->t_ctid, &t_ctid);
4378 : :
3 akorotkov@postgresql 4379 : 6808 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4380 : :
4381 : : /*
4382 : : * If any subtransaction of the current top transaction already holds
4383 : : * a lock as strong as or stronger than what we're requesting, we
4384 : : * effectively hold the desired lock already. We *must* succeed
4385 : : * without trying to take the tuple lock, else we will deadlock
4386 : : * against anyone wanting to acquire a stronger lock.
4387 : : *
4388 : : * Note we only do this the first time we loop on the HTSU result;
4389 : : * there is no point in testing in subsequent passes, because
4390 : : * evidently our own transaction cannot have acquired a new lock after
4391 : : * the first time we checked.
4392 : : */
3292 alvherre@alvh.no-ip. 4393 [ + + ]: 6808 : if (first_time)
4394 : : {
4395 : 6799 : first_time = false;
4396 : :
4397 [ + + ]: 6799 : if (infomask & HEAP_XMAX_IS_MULTI)
4398 : : {
4399 : : int i;
4400 : : int nmembers;
4401 : : MultiXactMember *members;
4402 : :
4403 : : /*
4404 : : * We don't need to allow old multixacts here; if that had
4405 : : * been the case, HeapTupleSatisfiesUpdate would have returned
4406 : : * MayBeUpdated and we wouldn't be here.
4407 : : */
4408 : : nmembers =
4409 : 84 : GetMultiXactIdMembers(xwait, &members, false,
4410 [ + + - + ]: 84 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
4411 : :
4412 [ + + ]: 251 : for (i = 0; i < nmembers; i++)
4413 : : {
4414 : : /* only consider members of our own transaction */
4415 [ + + ]: 181 : if (!TransactionIdIsCurrentTransactionId(members[i].xid))
4416 : 132 : continue;
4417 : :
4418 [ + + ]: 49 : if (TUPLOCK_from_mxstatus(members[i].status) >= mode)
4419 : : {
4099 4420 : 14 : pfree(members);
1849 andres@anarazel.de 4421 : 14 : result = TM_Ok;
2827 4422 : 14 : goto out_unlocked;
4423 : : }
4424 : : else
4425 : : {
4426 : : /*
4427 : : * Disable acquisition of the heavyweight tuple lock.
4428 : : * Otherwise, when promoting a weaker lock, we might
4429 : : * deadlock with another locker that has acquired the
4430 : : * heavyweight tuple lock and is waiting for our
4431 : : * transaction to finish.
4432 : : *
4433 : : * Note that in this case we still need to wait for
4434 : : * the multixact if required, to avoid acquiring
4435 : : * conflicting locks.
4436 : : */
1762 alvherre@alvh.no-ip. 4437 : 35 : skip_tuple_lock = true;
4438 : : }
4439 : : }
4440 : :
3292 4441 [ + - ]: 70 : if (members)
4442 : 70 : pfree(members);
4443 : : }
4444 [ + + ]: 6715 : else if (TransactionIdIsCurrentTransactionId(xwait))
4445 : : {
4446 [ + + + + : 5495 : switch (mode)
- ]
4447 : : {
4448 : 138 : case LockTupleKeyShare:
4449 [ - + - - : 138 : Assert(HEAP_XMAX_IS_KEYSHR_LOCKED(infomask) ||
- - ]
4450 : : HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4451 : : HEAP_XMAX_IS_EXCL_LOCKED(infomask));
1849 andres@anarazel.de 4452 : 138 : result = TM_Ok;
2827 4453 : 138 : goto out_unlocked;
3292 alvherre@alvh.no-ip. 4454 : 6 : case LockTupleShare:
4455 [ + - ]: 6 : if (HEAP_XMAX_IS_SHR_LOCKED(infomask) ||
4456 [ - + ]: 6 : HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4457 : : {
1849 andres@anarazel.de 4458 :LBC (110) : result = TM_Ok;
2827 4459 : (110) : goto out_unlocked;
4460 : : }
3292 alvherre@alvh.no-ip. 4461 :CBC 6 : break;
4462 : 61 : case LockTupleNoKeyExclusive:
4463 [ + + ]: 61 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4464 : : {
1849 andres@anarazel.de 4465 : 50 : result = TM_Ok;
2827 4466 : 50 : goto out_unlocked;
4467 : : }
3292 alvherre@alvh.no-ip. 4468 : 11 : break;
4469 : 5290 : case LockTupleExclusive:
4470 [ + + ]: 5290 : if (HEAP_XMAX_IS_EXCL_LOCKED(infomask) &&
4471 [ + + ]: 251 : infomask2 & HEAP_KEYS_UPDATED)
4472 : : {
1849 andres@anarazel.de 4473 : 230 : result = TM_Ok;
2827 4474 : 230 : goto out_unlocked;
4475 : : }
3292 alvherre@alvh.no-ip. 4476 : 5060 : break;
4477 : : }
4478 : : }
4479 : : }
4480 : :
4481 : : /*
4482 : : * Initially assume that we will have to wait for the locking
4483 : : * transaction(s) to finish. We check various cases below in which
4484 : : * this can be turned off.
4485 : : */
4099 4486 : 6376 : require_sleep = true;
4487 [ + + ]: 6376 : if (mode == LockTupleKeyShare)
4488 : : {
4489 : : /*
4490 : : * If we're requesting KeyShare, and there's no update present, we
4491 : : * don't need to wait. Even if there is an update, we can still
4492 : : * continue if the key hasn't been modified.
4493 : : *
4494 : : * However, if there are updates, we need to walk the update chain
4495 : : * to mark future versions of the row as locked, too. That way,
4496 : : * if somebody deletes that future version, we're protected
4497 : : * against the key going away. This locking of future versions
4498 : : * could block momentarily, if a concurrent transaction is
4499 : : * deleting a key; or it could return a value to the effect that
4500 : : * the transaction deleting the key has already committed. So we
4501 : : * do this before re-locking the buffer; otherwise this would be
4502 : : * prone to deadlocks.
4503 : : *
4504 : : * Note that the TID we're locking was grabbed before we unlocked
4505 : : * the buffer. For it to change while we're not looking, the
4506 : : * other properties we're testing for below after re-locking the
4507 : : * buffer would also change, in which case we would restart this
4508 : : * loop above.
4509 : : */
4510 [ + + ]: 578 : if (!(infomask2 & HEAP_KEYS_UPDATED))
4511 : : {
4512 : : bool updated;
4513 : :
4514 [ + + + - ]: 547 : updated = !HEAP_XMAX_IS_LOCKED_ONLY(infomask);
4515 : :
4516 : : /*
4517 : : * If there are updates, follow the update chain; bail out if
4518 : : * that cannot be done.
4519 : : */
4520 [ + - + + ]: 547 : if (follow_updates && updated)
4521 : : {
4522 : : TM_Result res;
4523 : :
4524 : 50 : res = heap_lock_updated_tuple(relation, tuple, &t_ctid,
4525 : : GetCurrentTransactionId(),
4526 : : mode);
1849 andres@anarazel.de 4527 [ + + ]: 50 : if (res != TM_Ok)
4528 : : {
4099 alvherre@alvh.no-ip. 4529 : 6 : result = res;
4530 : : /* recovery code expects to have buffer lock held */
3 akorotkov@postgresql 4531 : 6 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4099 alvherre@alvh.no-ip. 4532 : 182 : goto failed;
4533 : : }
4534 : : }
4535 : :
3 akorotkov@postgresql 4536 : 541 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4537 : :
4538 : : /*
4539 : : * Make sure it's still an appropriate lock, else start over.
4540 : : * Also, if it wasn't updated before we released the lock, but
4541 : : * is updated now, we start over too; the reason is that we
4542 : : * now need to follow the update chain to lock the new
4543 : : * versions.
4544 : : */
4099 alvherre@alvh.no-ip. 4545 [ + + ]: 541 : if (!HeapTupleHeaderIsOnlyLocked(tuple->t_data) &&
4546 [ + - ]: 43 : ((tuple->t_data->t_infomask2 & HEAP_KEYS_UPDATED) ||
4547 [ - + ]: 43 : !updated))
4099 alvherre@alvh.no-ip. 4548 :UBC 0 : goto l3;
4549 : :
4550 : : /* Things look okay, so we can skip sleeping */
4099 alvherre@alvh.no-ip. 4551 :CBC 541 : require_sleep = false;
4552 : :
4553 : : /*
4554 : : * Note we allow Xmax to change here; other updaters/lockers
4555 : : * could have modified it before we grabbed the buffer lock.
4556 : : * However, this is not a problem, because with the recheck we
4557 : : * just did we ensure that they still don't conflict with the
4558 : : * lock we want.
4559 : : */
4560 : : }
4561 : : }
4562 [ + + ]: 5798 : else if (mode == LockTupleShare)
4563 : : {
4564 : : /*
4565 : : * If we're requesting Share, we can similarly avoid sleeping if
4566 : : * there's no update and no exclusive lock present.
4567 : : */
4568 [ - + - - ]: 441 : if (HEAP_XMAX_IS_LOCKED_ONLY(infomask) &&
4569 [ + + ]: 441 : !HEAP_XMAX_IS_EXCL_LOCKED(infomask))
4570 : : {
3 akorotkov@postgresql 4571 : 435 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4572 : :
4573 : : /*
4574 : : * Make sure it's still an appropriate lock, else start over.
4575 : : * See above about allowing xmax to change.
4576 : : */
4099 alvherre@alvh.no-ip. 4577 [ - + - - ]: 435 : if (!HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
4578 [ - + ]: 435 : HEAP_XMAX_IS_EXCL_LOCKED(tuple->t_data->t_infomask))
4099 alvherre@alvh.no-ip. 4579 :UBC 0 : goto l3;
4099 alvherre@alvh.no-ip. 4580 :CBC 435 : require_sleep = false;
4581 : : }
4582 : : }
4583 [ + + ]: 5357 : else if (mode == LockTupleNoKeyExclusive)
4584 : : {
4585 : : /*
4586 : : * If we're requesting NoKeyExclusive, we might also be able to
4587 : : * avoid sleeping; just ensure that there no conflicting lock
4588 : : * already acquired.
4589 : : */
4590 [ + + ]: 155 : if (infomask & HEAP_XMAX_IS_MULTI)
4591 : : {
3397 4592 [ + + ]: 26 : if (!DoesMultiXactIdConflict((MultiXactId) xwait, infomask,
4593 : : mode, NULL))
4594 : : {
4595 : : /*
4596 : : * No conflict, but if the xmax changed under us in the
4597 : : * meantime, start over.
4598 : : */
3 akorotkov@postgresql 4599 : 13 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
3397 alvherre@alvh.no-ip. 4600 [ + - ]: 13 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4601 [ - + ]: 13 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4602 : : xwait))
3397 alvherre@alvh.no-ip. 4603 :UBC 0 : goto l3;
4604 : :
4605 : : /* otherwise, we're good */
3397 alvherre@alvh.no-ip. 4606 :CBC 13 : require_sleep = false;
4607 : : }
4608 : : }
4099 4609 [ + + ]: 129 : else if (HEAP_XMAX_IS_KEYSHR_LOCKED(infomask))
4610 : : {
3 akorotkov@postgresql 4611 : 15 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4612 : :
4613 : : /* if the xmax changed in the meantime, start over */
3643 alvherre@alvh.no-ip. 4614 [ + - ]: 15 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
1536 4615 [ - + ]: 15 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4616 : : xwait))
4099 alvherre@alvh.no-ip. 4617 :UBC 0 : goto l3;
4618 : : /* otherwise, we're good */
4099 alvherre@alvh.no-ip. 4619 :CBC 15 : require_sleep = false;
4620 : : }
4621 : : }
4622 : :
4623 : : /*
4624 : : * As a check independent from those above, we can also avoid sleeping
4625 : : * if the current transaction is the sole locker of the tuple. Note
4626 : : * that the strength of the lock already held is irrelevant; this is
4627 : : * not about recording the lock in Xmax (which will be done regardless
4628 : : * of this optimization, below). Also, note that the cases where we
4629 : : * hold a lock stronger than we are requesting are already handled
4630 : : * above by not doing anything.
4631 : : *
4632 : : * Note we only deal with the non-multixact case here; MultiXactIdWait
4633 : : * is well equipped to deal with this situation on its own.
4634 : : */
3292 4635 [ + + + + : 11695 : if (require_sleep && !(infomask & HEAP_XMAX_IS_MULTI) &&
+ + ]
4636 : 5325 : TransactionIdIsCurrentTransactionId(xwait))
4637 : : {
4638 : : /* ... but if the xmax changed in the meantime, start over */
3 akorotkov@postgresql 4639 : 5060 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
3292 alvherre@alvh.no-ip. 4640 [ + - ]: 5060 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4641 [ - + ]: 5060 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4642 : : xwait))
3292 alvherre@alvh.no-ip. 4643 :UBC 0 : goto l3;
3292 alvherre@alvh.no-ip. 4644 [ - + - - ]:CBC 5060 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask));
4645 : 5060 : require_sleep = false;
4646 : : }
4647 : :
4648 : : /*
4649 : : * Time to sleep on the other transaction/multixact, if necessary.
4650 : : *
4651 : : * If the other transaction is an update/delete that's already
4652 : : * committed, then sleeping cannot possibly do any good: if we're
4653 : : * required to sleep, get out to raise an error instead.
4654 : : *
4655 : : * By here, we either have already acquired the buffer exclusive lock,
4656 : : * or we must wait for the locking transaction or multixact; so below
4657 : : * we ensure that we grab buffer lock after the sleep.
4658 : : */
1849 andres@anarazel.de 4659 [ + + + + : 6370 : if (require_sleep && (result == TM_Updated || result == TM_Deleted))
+ + ]
4660 : : {
3 akorotkov@postgresql 4661 : 138 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
2830 alvherre@alvh.no-ip. 4662 : 138 : goto failed;
4663 : : }
4664 [ + + ]: 6232 : else if (require_sleep)
4665 : : {
4666 : : /*
4667 : : * Acquire tuple lock to establish our priority for the tuple, or
4668 : : * die trying. LockTuple will release us when we are next-in-line
4669 : : * for the tuple. We must do this even if we are share-locking,
4670 : : * but not if we already have a weaker lock on the tuple.
4671 : : *
4672 : : * If we are forced to "start over" below, we keep the tuple lock;
4673 : : * this arranges that we stay at the head of the line while
4674 : : * rechecking tuple state.
4675 : : */
1762 4676 [ + + ]: 168 : if (!skip_tuple_lock &&
4677 [ + + ]: 152 : !heap_acquire_tuplock(relation, tid, mode, wait_policy,
4678 : : &have_tuple_lock))
4679 : : {
4680 : : /*
4681 : : * This can only happen if wait_policy is Skip and the lock
4682 : : * couldn't be obtained.
4683 : : */
1849 andres@anarazel.de 4684 : 1 : result = TM_WouldBlock;
4685 : : /* recovery code expects to have buffer lock held */
3 akorotkov@postgresql 4686 : 1 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
3397 alvherre@alvh.no-ip. 4687 : 1 : goto failed;
4688 : : }
4689 : :
4099 4690 [ + + ]: 166 : if (infomask & HEAP_XMAX_IS_MULTI)
4691 : : {
4692 : 40 : MultiXactStatus status = get_mxact_status_for_lock(mode, false);
4693 : :
4694 : : /* We only ever lock tuples, never update them */
4695 [ - + ]: 40 : if (status >= MultiXactStatusNoKeyUpdate)
4099 alvherre@alvh.no-ip. 4696 [ # # ]:UBC 0 : elog(ERROR, "invalid lock mode in heap_lock_tuple");
4697 : :
4698 : : /* wait for multixact to end, or die trying */
3477 alvherre@alvh.no-ip. 4699 [ + + + - ]:CBC 40 : switch (wait_policy)
4700 : : {
4701 : 36 : case LockWaitBlock:
4702 : 36 : MultiXactIdWait((MultiXactId) xwait, status, infomask,
4703 : : relation, &tuple->t_self, XLTW_Lock, NULL);
4704 : 36 : break;
4705 : 2 : case LockWaitSkip:
4706 [ + - ]: 2 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
4707 : : status, infomask, relation,
4708 : : NULL))
4709 : : {
1849 andres@anarazel.de 4710 : 2 : result = TM_WouldBlock;
4711 : : /* recovery code expects to have buffer lock held */
3 akorotkov@postgresql 4712 : 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
3477 alvherre@alvh.no-ip. 4713 : 2 : goto failed;
4714 : : }
3477 alvherre@alvh.no-ip. 4715 :UBC 0 : break;
3477 alvherre@alvh.no-ip. 4716 :CBC 2 : case LockWaitError:
4717 [ + - ]: 2 : if (!ConditionalMultiXactIdWait((MultiXactId) xwait,
4718 : : status, infomask, relation,
4719 : : NULL))
4720 [ + - ]: 2 : ereport(ERROR,
4721 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
4722 : : errmsg("could not obtain lock on row in relation \"%s\"",
4723 : : RelationGetRelationName(relation))));
4724 : :
3477 alvherre@alvh.no-ip. 4725 :UBC 0 : break;
4726 : : }
4727 : :
4728 : : /*
4729 : : * Of course, the multixact might not be done here: if we're
4730 : : * requesting a light lock mode, other transactions with light
4731 : : * locks could still be alive, as well as locks owned by our
4732 : : * own xact or other subxacts of this backend. We need to
4733 : : * preserve the surviving MultiXact members. Note that it
4734 : : * isn't absolutely necessary in the latter case, but doing so
4735 : : * is simpler.
4736 : : */
4737 : : }
4738 : : else
4739 : : {
4740 : : /* wait for regular transaction to end, or die trying */
3477 alvherre@alvh.no-ip. 4741 [ + + + - ]:CBC 126 : switch (wait_policy)
4742 : : {
4743 : 87 : case LockWaitBlock:
3357 heikki.linnakangas@i 4744 : 87 : XactLockTableWait(xwait, relation, &tuple->t_self,
4745 : : XLTW_Lock);
3477 alvherre@alvh.no-ip. 4746 : 87 : break;
4747 : 33 : case LockWaitSkip:
4748 [ + - ]: 33 : if (!ConditionalXactLockTableWait(xwait))
4749 : : {
1849 andres@anarazel.de 4750 : 33 : result = TM_WouldBlock;
4751 : : /* recovery code expects to have buffer lock held */
3 akorotkov@postgresql 4752 : 33 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
3477 alvherre@alvh.no-ip. 4753 : 33 : goto failed;
4754 : : }
3477 alvherre@alvh.no-ip. 4755 :UBC 0 : break;
3477 alvherre@alvh.no-ip. 4756 :CBC 6 : case LockWaitError:
4757 [ + - ]: 6 : if (!ConditionalXactLockTableWait(xwait))
4758 [ + - ]: 6 : ereport(ERROR,
4759 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
4760 : : errmsg("could not obtain lock on row in relation \"%s\"",
4761 : : RelationGetRelationName(relation))));
3477 alvherre@alvh.no-ip. 4762 :UBC 0 : break;
4763 : : }
4764 : : }
4765 : :
4766 : : /* if there are updates, follow the update chain */
3292 alvherre@alvh.no-ip. 4767 [ + + + + :CBC 123 : if (follow_updates && !HEAP_XMAX_IS_LOCKED_ONLY(infomask))
+ - ]
4768 : : {
4769 : : TM_Result res;
4770 : :
4771 : 39 : res = heap_lock_updated_tuple(relation, tuple, &t_ctid,
4772 : : GetCurrentTransactionId(),
4773 : : mode);
1849 andres@anarazel.de 4774 [ + + ]: 39 : if (res != TM_Ok)
4775 : : {
3292 alvherre@alvh.no-ip. 4776 : 2 : result = res;
4777 : : /* recovery code expects to have buffer lock held */
3 akorotkov@postgresql 4778 : 2 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
3292 alvherre@alvh.no-ip. 4779 : 2 : goto failed;
4780 : : }
4781 : : }
4782 : :
3 akorotkov@postgresql 4783 : 121 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
4784 : :
4785 : : /*
4786 : : * xwait is done, but if xwait had just locked the tuple then some
4787 : : * other xact could update this tuple before we get to this point.
4788 : : * Check for xmax change, and start over if so.
4789 : : */
3292 alvherre@alvh.no-ip. 4790 [ + + ]: 121 : if (xmax_infomask_changed(tuple->t_data->t_infomask, infomask) ||
4791 [ + + ]: 110 : !TransactionIdEquals(HeapTupleHeaderGetRawXmax(tuple->t_data),
4792 : : xwait))
4793 : 13 : goto l3;
4794 : :
4795 [ + + ]: 108 : if (!(infomask & HEAP_XMAX_IS_MULTI))
4796 : : {
4797 : : /*
4798 : : * Otherwise check if it committed or aborted. Note we cannot
4799 : : * be here if the tuple was only locked by somebody who didn't
4800 : : * conflict with us; that would have been handled above. So
4801 : : * that transaction must necessarily be gone by now. But
4802 : : * don't check for this in the multixact case, because some
4803 : : * locker transactions might still be running.
4804 : : */
3 akorotkov@postgresql 4805 : 76 : UpdateXmaxHintBits(tuple->t_data, *buffer, xwait);
4806 : : }
4807 : : }
4808 : :
4809 : : /* By here, we're certain that we hold buffer exclusive lock again */
4810 : :
4811 : : /*
4812 : : * We may lock if previous xmax aborted, or if it committed but only
4813 : : * locked the tuple without updating it; or if we didn't have to wait
4814 : : * at all for whatever reason.
4815 : : */
4099 alvherre@alvh.no-ip. 4816 [ + + ]: 6172 : if (!require_sleep ||
4817 [ + + ]: 108 : (tuple->t_data->t_infomask & HEAP_XMAX_INVALID) ||
4818 [ + + + - : 141 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_data->t_infomask) ||
+ + ]
4819 : 63 : HeapTupleHeaderIsOnlyLocked(tuple->t_data))
1849 andres@anarazel.de 4820 : 6116 : result = TM_Ok;
1147 alvherre@alvh.no-ip. 4821 [ + + ]: 56 : else if (!ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid))
1849 andres@anarazel.de 4822 : 45 : result = TM_Updated;
4823 : : else
4824 : 11 : result = TM_Deleted;
4825 : : }
4826 : :
4099 alvherre@alvh.no-ip. 4827 : 75902 : failed:
1849 andres@anarazel.de 4828 [ + + ]: 82256 : if (result != TM_Ok)
4829 : : {
4830 [ + + + + : 244 : Assert(result == TM_SelfModified || result == TM_Updated ||
+ + - + ]
4831 : : result == TM_Deleted || result == TM_WouldBlock);
4832 : :
4833 : : /*
4834 : : * When locking a tuple under LockWaitSkip semantics and we fail with
4835 : : * TM_WouldBlock above, it's possible for concurrent transactions to
4836 : : * release the lock and set HEAP_XMAX_INVALID in the meantime. So
4837 : : * this assert is slightly different from the equivalent one in
4838 : : * heap_delete and heap_update.
4839 : : */
831 alvherre@alvh.no-ip. 4840 [ + + - + ]: 244 : Assert((result == TM_WouldBlock) ||
4841 : : !(tuple->t_data->t_infomask & HEAP_XMAX_INVALID));
1849 andres@anarazel.de 4842 [ + + - + ]: 244 : Assert(result != TM_Updated ||
4843 : : !ItemPointerEquals(&tuple->t_self, &tuple->t_data->t_ctid));
4844 : 244 : tmfd->ctid = tuple->t_data->t_ctid;
4845 [ + - + + : 244 : tmfd->xmax = HeapTupleHeaderGetUpdateXid(tuple->t_data);
+ + ]
4846 [ + + ]: 244 : if (result == TM_SelfModified)
4847 : 6 : tmfd->cmax = HeapTupleHeaderGetCmax(tuple->t_data);
4848 : : else
4849 : 238 : tmfd->cmax = InvalidCommandId;
2827 4850 : 244 : goto out_locked;
4851 : : }
4852 : :
4853 : : /*
4854 : : * If we didn't pin the visibility map page and the page has become all
4855 : : * visible while we were busy locking the buffer, or during some
4856 : : * subsequent window during which we had it unlocked, we'll have to unlock
4857 : : * and re-lock, to avoid holding the buffer lock across I/O. That's a bit
4858 : : * unfortunate, especially since we'll now have to recheck whether the
4859 : : * tuple has been locked or updated under us, but hopefully it won't
4860 : : * happen very often.
4861 : : */
2810 4862 [ + + - + ]: 82012 : if (vmbuffer == InvalidBuffer && PageIsAllVisible(page))
4863 : : {
3 akorotkov@postgresql 4864 :UBC 0 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
2810 andres@anarazel.de 4865 : 0 : visibilitymap_pin(relation, block, &vmbuffer);
3 akorotkov@postgresql 4866 : 0 : LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
2810 andres@anarazel.de 4867 : 0 : goto l3;
4868 : : }
4869 : :
4099 alvherre@alvh.no-ip. 4870 :CBC 82012 : xmax = HeapTupleHeaderGetRawXmax(tuple->t_data);
4871 : 82012 : old_infomask = tuple->t_data->t_infomask;
4872 : :
4873 : : /*
4874 : : * If this is the first possibly-multixact-able operation in the current
4875 : : * transaction, set my per-backend OldestMemberMXactId setting. We can be
4876 : : * certain that the transaction will never become a member of any older
4877 : : * MultiXactIds than that. (We have to do this even if we end up just
4878 : : * using our own TransactionId below, since some other backend could
4879 : : * incorporate our XID into a MultiXact immediately afterwards.)
4880 : : */
4881 : 82012 : MultiXactIdSetOldestMember();
4882 : :
4883 : : /*
4884 : : * Compute the new xmax and infomask to store into the tuple. Note we do
4885 : : * not modify the tuple just yet, because that would leave it in the wrong
4886 : : * state if multixact.c elogs.
4887 : : */
4888 : 82012 : compute_new_xmax_infomask(xmax, old_infomask, tuple->t_data->t_infomask2,
4889 : : GetCurrentTransactionId(), mode, false,
4890 : : &xid, &new_infomask, &new_infomask2);
4891 : :
6926 tgl@sss.pgh.pa.us 4892 : 82012 : START_CRIT_SECTION();
4893 : :
4894 : : /*
4895 : : * Store transaction information of xact locking the tuple.
4896 : : *
4897 : : * Note: Cmax is meaningless in this context, so don't set it; this avoids
4898 : : * possibly generating a useless combo CID. Moreover, if we're locking a
4899 : : * previously updated tuple, it's important to preserve the Cmax.
4900 : : *
4901 : : * Also reset the HOT UPDATE bit, but only if there's no update; otherwise
4902 : : * we would break the HOT chain.
4903 : : */
4099 alvherre@alvh.no-ip. 4904 : 82012 : tuple->t_data->t_infomask &= ~HEAP_XMAX_BITS;
4905 : 82012 : tuple->t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
4906 : 82012 : tuple->t_data->t_infomask |= new_infomask;
4907 : 82012 : tuple->t_data->t_infomask2 |= new_infomask2;
4908 [ + + - + ]: 82012 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
4909 : 81973 : HeapTupleHeaderClearHotUpdated(tuple->t_data);
7150 tgl@sss.pgh.pa.us 4910 : 82012 : HeapTupleHeaderSetXmax(tuple->t_data, xid);
4911 : :
4912 : : /*
4913 : : * Make sure there is no forward chain link in t_ctid. Note that in the
4914 : : * cases where the tuple has been updated, we must not overwrite t_ctid,
4915 : : * because it was set by the updater. Moreover, if the tuple has been
4916 : : * updated, we need to follow the update chain to lock the new versions of
4917 : : * the tuple as well.
4918 : : */
4099 alvherre@alvh.no-ip. 4919 [ + + - + ]: 82012 : if (HEAP_XMAX_IS_LOCKED_ONLY(new_infomask))
4920 : 81973 : tuple->t_data->t_ctid = *tid;
4921 : :
4922 : : /* Clear only the all-frozen bit on visibility map if needed */
2827 andres@anarazel.de 4923 [ + + + + ]: 83670 : if (PageIsAllVisible(page) &&
4924 : 1658 : visibilitymap_clear(relation, block, vmbuffer,
4925 : : VISIBILITYMAP_ALL_FROZEN))
4926 : 14 : cleared_all_frozen = true;
4927 : :
4928 : :
3 akorotkov@postgresql 4929 : 82012 : MarkBufferDirty(*buffer);
4930 : :
4931 : : /*
4932 : : * XLOG stuff. You might think that we don't need an XLOG record because
4933 : : * there is no state change worth restoring after a crash. You would be
4934 : : * wrong however: we have just written either a TransactionId or a
4935 : : * MultiXactId that may never have been seen on disk before, and we need
4936 : : * to make sure that there are XLOG entries covering those ID numbers.
4937 : : * Else the same IDs might be re-used after a crash, which would be
4938 : : * disastrous if this page made it to disk before the crash. Essentially
4939 : : * we have to enforce the WAL log-before-data rule even in this case.
4940 : : * (Also, in a PITR log-shipping or 2PC environment, we have to have XLOG
4941 : : * entries for everything anyway.)
4942 : : */
4871 rhaas@postgresql.org 4943 [ + + + + : 82012 : if (RelationNeedsWAL(relation))
+ - + - ]
4944 : : {
4945 : : xl_heap_lock xlrec;
4946 : : XLogRecPtr recptr;
4947 : :
3433 heikki.linnakangas@i 4948 : 81451 : XLogBeginInsert();
3 akorotkov@postgresql 4949 : 81451 : XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD);
4950 : :
3433 heikki.linnakangas@i 4951 : 81451 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
369 pg@bowt.ie 4952 : 81451 : xlrec.xmax = xid;
4099 alvherre@alvh.no-ip. 4953 : 162902 : xlrec.infobits_set = compute_infobits(new_infomask,
4954 : 81451 : tuple->t_data->t_infomask2);
2827 andres@anarazel.de 4955 : 81451 : xlrec.flags = cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
3433 heikki.linnakangas@i 4956 : 81451 : XLogRegisterData((char *) &xlrec, SizeOfHeapLock);
4957 : :
4958 : : /* we don't decode row locks atm, so no need to log the origin */
4959 : :
4960 : 81451 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_LOCK);
4961 : :
5754 tgl@sss.pgh.pa.us 4962 : 81451 : PageSetLSN(page, recptr);
4963 : : }
4964 : :
6926 4965 [ - + ]: 82012 : END_CRIT_SECTION();
4966 : :
1849 andres@anarazel.de 4967 : 82012 : result = TM_Ok;
4968 : :
2827 4969 : 82268 : out_locked:
3 akorotkov@postgresql 4970 : 82268 : LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
4971 : :
2827 andres@anarazel.de 4972 : 82700 : out_unlocked:
4973 [ + + ]: 82700 : if (BufferIsValid(vmbuffer))
4974 : 1658 : ReleaseBuffer(vmbuffer);
4975 : :
4976 : : /*
4977 : : * Don't update the visibility map here. Locking a tuple doesn't change
4978 : : * visibility info.
4979 : : */
4980 : :
4981 : : /*
4982 : : * Now that we have successfully marked the tuple as locked, we can
4983 : : * release the lmgr tuple lock, if we had it.
4984 : : */
6924 tgl@sss.pgh.pa.us 4985 [ + + ]: 82700 : if (have_tuple_lock)
4099 alvherre@alvh.no-ip. 4986 : 137 : UnlockTupleTuplock(relation, tid, mode);
4987 : :
2827 andres@anarazel.de 4988 : 82700 : return result;
4989 : : }
4990 : :
4991 : : /*
4992 : : * Acquire heavyweight lock on the given tuple, in preparation for acquiring
4993 : : * its normal, Xmax-based tuple lock.
4994 : : *
4995 : : * have_tuple_lock is an input and output parameter: on input, it indicates
4996 : : * whether the lock has previously been acquired (and this function does
4997 : : * nothing in that case). If this function returns success, have_tuple_lock
4998 : : * has been flipped to true.
4999 : : *
5000 : : * Returns false if it was unable to obtain the lock; this can only happen if
5001 : : * wait_policy is Skip.
5002 : : */
5003 : : static bool
3397 alvherre@alvh.no-ip. 5004 : 260 : heap_acquire_tuplock(Relation relation, ItemPointer tid, LockTupleMode mode,
5005 : : LockWaitPolicy wait_policy, bool *have_tuple_lock)
5006 : : {
5007 [ + + ]: 260 : if (*have_tuple_lock)
5008 : 9 : return true;
5009 : :
5010 [ + + + - ]: 251 : switch (wait_policy)
5011 : : {
5012 : 210 : case LockWaitBlock:
5013 : 210 : LockTupleTuplock(relation, tid, mode);
5014 : 210 : break;
5015 : :
5016 : 34 : case LockWaitSkip:
5017 [ + + ]: 34 : if (!ConditionalLockTupleTuplock(relation, tid, mode))
5018 : 1 : return false;
5019 : 33 : break;
5020 : :
5021 : 7 : case LockWaitError:
5022 [ + + ]: 7 : if (!ConditionalLockTupleTuplock(relation, tid, mode))
5023 [ + - ]: 1 : ereport(ERROR,
5024 : : (errcode(ERRCODE_LOCK_NOT_AVAILABLE),
5025 : : errmsg("could not obtain lock on row in relation \"%s\"",
5026 : : RelationGetRelationName(relation))));
5027 : 6 : break;
5028 : : }
5029 : 249 : *have_tuple_lock = true;
5030 : :
5031 : 249 : return true;
5032 : : }
5033 : :
5034 : : /*
5035 : : * Given an original set of Xmax and infomask, and a transaction (identified by
5036 : : * add_to_xmax) acquiring a new lock of some mode, compute the new Xmax and
5037 : : * corresponding infomasks to use on the tuple.
5038 : : *
5039 : : * Note that this might have side effects such as creating a new MultiXactId.
5040 : : *
5041 : : * Most callers will have called HeapTupleSatisfiesUpdate before this function;
5042 : : * that will have set the HEAP_XMAX_INVALID bit if the xmax was a MultiXactId
5043 : : * but it was not running anymore. There is a race condition, which is that the
5044 : : * MultiXactId may have finished since then, but that uncommon case is handled
5045 : : * either here, or within MultiXactIdExpand.
5046 : : *
5047 : : * There is a similar race condition possible when the old xmax was a regular
5048 : : * TransactionId. We test TransactionIdIsInProgress again just to narrow the
5049 : : * window, but it's still possible to end up creating an unnecessary
5050 : : * MultiXactId. Fortunately this is harmless.
5051 : : */
5052 : : static void
4099 5053 : 1932106 : compute_new_xmax_infomask(TransactionId xmax, uint16 old_infomask,
5054 : : uint16 old_infomask2, TransactionId add_to_xmax,
5055 : : LockTupleMode mode, bool is_update,
5056 : : TransactionId *result_xmax, uint16 *result_infomask,
5057 : : uint16 *result_infomask2)
5058 : : {
5059 : : TransactionId new_xmax;
5060 : : uint16 new_infomask,
5061 : : new_infomask2;
5062 : :
3769 5063 [ + - ]: 1932106 : Assert(TransactionIdIsCurrentTransactionId(add_to_xmax));
5064 : :
4099 5065 : 2036120 : l5:
5066 : 2036120 : new_infomask = 0;
5067 : 2036120 : new_infomask2 = 0;
5068 [ + + ]: 2036120 : if (old_infomask & HEAP_XMAX_INVALID)
5069 : : {
5070 : : /*
5071 : : * No previous locker; we just insert our own TransactionId.
5072 : : *
5073 : : * Note that it's critical that this case be the first one checked,
5074 : : * because there are several blocks below that come back to this one
5075 : : * to implement certain optimizations; old_infomask might contain
5076 : : * other dirty bits in those cases, but we don't really care.
5077 : : */
5078 [ + + ]: 1930982 : if (is_update)
5079 : : {
5080 : 1709479 : new_xmax = add_to_xmax;
5081 [ + + ]: 1709479 : if (mode == LockTupleExclusive)
5082 : 1468351 : new_infomask2 |= HEAP_KEYS_UPDATED;
5083 : : }
5084 : : else
5085 : : {
5086 : 221503 : new_infomask |= HEAP_XMAX_LOCK_ONLY;
5087 [ + + + + : 221503 : switch (mode)
- ]
5088 : : {
5089 : 2471 : case LockTupleKeyShare:
5090 : 2471 : new_xmax = add_to_xmax;
5091 : 2471 : new_infomask |= HEAP_XMAX_KEYSHR_LOCK;
5092 : 2471 : break;
5093 : 708 : case LockTupleShare:
5094 : 708 : new_xmax = add_to_xmax;
5095 : 708 : new_infomask |= HEAP_XMAX_SHR_LOCK;
5096 : 708 : break;
5097 : 123687 : case LockTupleNoKeyExclusive:
5098 : 123687 : new_xmax = add_to_xmax;
5099 : 123687 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5100 : 123687 : break;
5101 : 94637 : case LockTupleExclusive:
5102 : 94637 : new_xmax = add_to_xmax;
5103 : 94637 : new_infomask |= HEAP_XMAX_EXCL_LOCK;
5104 : 94637 : new_infomask2 |= HEAP_KEYS_UPDATED;
5105 : 94637 : break;
4099 alvherre@alvh.no-ip. 5106 :UBC 0 : default:
5107 : 0 : new_xmax = InvalidTransactionId; /* silence compiler */
5108 [ # # ]: 0 : elog(ERROR, "invalid lock mode");
5109 : : }
5110 : : }
5111 : : }
4099 alvherre@alvh.no-ip. 5112 [ + + ]:CBC 105138 : else if (old_infomask & HEAP_XMAX_IS_MULTI)
5113 : : {
5114 : : MultiXactStatus new_status;
5115 : :
5116 : : /*
5117 : : * Currently we don't allow XMAX_COMMITTED to be set for multis, so
5118 : : * cross-check.
5119 : : */
5120 [ - + ]: 121 : Assert(!(old_infomask & HEAP_XMAX_COMMITTED));
5121 : :
5122 : : /*
5123 : : * A multixact together with LOCK_ONLY set but neither lock bit set
5124 : : * (i.e. a pg_upgraded share locked tuple) cannot possibly be running
5125 : : * anymore. This check is critical for databases upgraded by
5126 : : * pg_upgrade; both MultiXactIdIsRunning and MultiXactIdExpand assume
5127 : : * that such multis are never passed.
5128 : : */
2851 5129 [ + - + + : 121 : if (HEAP_LOCKED_UPGRADED(old_infomask))
- + ]
5130 : : {
4099 alvherre@alvh.no-ip. 5131 :UBC 0 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5132 : 0 : old_infomask |= HEAP_XMAX_INVALID;
5133 : 0 : goto l5;
5134 : : }
5135 : :
5136 : : /*
5137 : : * If the XMAX is already a MultiXactId, then we need to expand it to
5138 : : * include add_to_xmax; but if all the members were lockers and are
5139 : : * all gone, we can do away with the IS_MULTI bit and just set
5140 : : * add_to_xmax as the only locker/updater. If all lockers are gone
5141 : : * and we have an updater that aborted, we can also do without a
5142 : : * multi.
5143 : : *
5144 : : * The cost of doing GetMultiXactIdMembers would be paid by
5145 : : * MultiXactIdExpand if we weren't to do this, so this check is not
5146 : : * incurring extra work anyhow.
5147 : : */
3547 alvherre@alvh.no-ip. 5148 [ + + - + :CBC 121 : if (!MultiXactIdIsRunning(xmax, HEAP_XMAX_IS_LOCKED_ONLY(old_infomask)))
+ + ]
5149 : : {
4099 5150 [ + + + - ]: 24 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) ||
3292 5151 [ + - ]: 9 : !TransactionIdDidCommit(MultiXactIdGetUpdateXid(xmax,
5152 : : old_infomask)))
5153 : : {
5154 : : /*
5155 : : * Reset these bits and restart; otherwise fall through to
5156 : : * create a new multi below.
5157 : : */
4099 5158 : 24 : old_infomask &= ~HEAP_XMAX_IS_MULTI;
5159 : 24 : old_infomask |= HEAP_XMAX_INVALID;
5160 : 24 : goto l5;
5161 : : }
5162 : : }
5163 : :
5164 : 97 : new_status = get_mxact_status_for_lock(mode, is_update);
5165 : :
5166 : 97 : new_xmax = MultiXactIdExpand((MultiXactId) xmax, add_to_xmax,
5167 : : new_status);
5168 : 97 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5169 : : }
5170 [ + + ]: 105017 : else if (old_infomask & HEAP_XMAX_COMMITTED)
5171 : : {
5172 : : /*
5173 : : * It's a committed update, so we need to preserve him as updater of
5174 : : * the tuple.
5175 : : */
5176 : : MultiXactStatus status;
5177 : : MultiXactStatus new_status;
5178 : :
5179 [ - + ]: 13 : if (old_infomask2 & HEAP_KEYS_UPDATED)
4099 alvherre@alvh.no-ip. 5180 :UBC 0 : status = MultiXactStatusUpdate;
5181 : : else
4099 alvherre@alvh.no-ip. 5182 :CBC 13 : status = MultiXactStatusNoKeyUpdate;
5183 : :
5184 : 13 : new_status = get_mxact_status_for_lock(mode, is_update);
5185 : :
5186 : : /*
5187 : : * since it's not running, it's obviously impossible for the old
5188 : : * updater to be identical to the current one, so we need not check
5189 : : * for that case as we do in the block above.
5190 : : */
5191 : 13 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5192 : 13 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5193 : : }
5194 [ + + ]: 105004 : else if (TransactionIdIsInProgress(xmax))
5195 : : {
5196 : : /*
5197 : : * If the XMAX is a valid, in-progress TransactionId, then we need to
5198 : : * create a new MultiXactId that includes both the old locker or
5199 : : * updater and our own TransactionId.
5200 : : */
5201 : : MultiXactStatus new_status;
5202 : : MultiXactStatus old_status;
5203 : : LockTupleMode old_mode;
5204 : :
5205 [ + + - + ]: 104995 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5206 : : {
5207 [ + + ]: 104969 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
3769 5208 : 5615 : old_status = MultiXactStatusForKeyShare;
4099 5209 [ + + ]: 99354 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
3769 5210 : 431 : old_status = MultiXactStatusForShare;
4099 5211 [ + - ]: 98923 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5212 : : {
5213 [ + + ]: 98923 : if (old_infomask2 & HEAP_KEYS_UPDATED)
3769 5214 : 92817 : old_status = MultiXactStatusForUpdate;
5215 : : else
5216 : 6106 : old_status = MultiXactStatusForNoKeyUpdate;
5217 : : }
5218 : : else
5219 : : {
5220 : : /*
5221 : : * LOCK_ONLY can be present alone only when a page has been
5222 : : * upgraded by pg_upgrade. But in that case,
5223 : : * TransactionIdIsInProgress() should have returned false. We
5224 : : * assume it's no longer locked in this case.
5225 : : */
4099 alvherre@alvh.no-ip. 5226 [ # # ]:UBC 0 : elog(WARNING, "LOCK_ONLY found for Xid in progress %u", xmax);
5227 : 0 : old_infomask |= HEAP_XMAX_INVALID;
5228 : 0 : old_infomask &= ~HEAP_XMAX_LOCK_ONLY;
5229 : 0 : goto l5;
5230 : : }
5231 : : }
5232 : : else
5233 : : {
5234 : : /* it's an update, but which kind? */
4099 alvherre@alvh.no-ip. 5235 [ - + ]:CBC 26 : if (old_infomask2 & HEAP_KEYS_UPDATED)
3769 alvherre@alvh.no-ip. 5236 :UBC 0 : old_status = MultiXactStatusUpdate;
5237 : : else
3769 alvherre@alvh.no-ip. 5238 :CBC 26 : old_status = MultiXactStatusNoKeyUpdate;
5239 : : }
5240 : :
5241 : 104995 : old_mode = TUPLOCK_from_mxstatus(old_status);
5242 : :
5243 : : /*
5244 : : * If the lock to be acquired is for the same TransactionId as the
5245 : : * existing lock, there's an optimization possible: consider only the
5246 : : * strongest of both locks as the only one present, and restart.
5247 : : */
4099 5248 [ + + ]: 104995 : if (xmax == add_to_xmax)
5249 : : {
5250 : : /*
5251 : : * Note that it's not possible for the original tuple to be
5252 : : * updated: we wouldn't be here because the tuple would have been
5253 : : * invisible and we wouldn't try to update it. As a subtlety,
5254 : : * this code can also run when traversing an update chain to lock
5255 : : * future versions of a tuple. But we wouldn't be here either,
5256 : : * because the add_to_xmax would be different from the original
5257 : : * updater.
5258 : : */
3769 5259 [ - + - - ]: 103982 : Assert(HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
5260 : :
5261 : : /* acquire the strongest of both */
5262 [ + + ]: 103982 : if (mode < old_mode)
5263 : 52257 : mode = old_mode;
5264 : : /* mustn't touch is_update */
5265 : :
5266 : 103982 : old_infomask |= HEAP_XMAX_INVALID;
5267 : 103982 : goto l5;
5268 : : }
5269 : :
5270 : : /* otherwise, just fall back to creating a new multixact */
5271 : 1013 : new_status = get_mxact_status_for_lock(mode, is_update);
5272 : 1013 : new_xmax = MultiXactIdCreate(xmax, old_status,
5273 : : add_to_xmax, new_status);
4099 5274 : 1013 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5275 : : }
5276 [ + + + - : 14 : else if (!HEAP_XMAX_IS_LOCKED_ONLY(old_infomask) &&
+ + ]
5277 : 5 : TransactionIdDidCommit(xmax))
5278 : 1 : {
5279 : : /*
5280 : : * It's a committed update, so we gotta preserve him as updater of the
5281 : : * tuple.
5282 : : */
5283 : : MultiXactStatus status;
5284 : : MultiXactStatus new_status;
5285 : :
5286 [ - + ]: 1 : if (old_infomask2 & HEAP_KEYS_UPDATED)
4099 alvherre@alvh.no-ip. 5287 :UBC 0 : status = MultiXactStatusUpdate;
5288 : : else
4099 alvherre@alvh.no-ip. 5289 :CBC 1 : status = MultiXactStatusNoKeyUpdate;
5290 : :
5291 : 1 : new_status = get_mxact_status_for_lock(mode, is_update);
5292 : :
5293 : : /*
5294 : : * since it's not running, it's obviously impossible for the old
5295 : : * updater to be identical to the current one, so we need not check
5296 : : * for that case as we do in the block above.
5297 : : */
5298 : 1 : new_xmax = MultiXactIdCreate(xmax, status, add_to_xmax, new_status);
5299 : 1 : GetMultiXactIdHintBits(new_xmax, &new_infomask, &new_infomask2);
5300 : : }
5301 : : else
5302 : : {
5303 : : /*
5304 : : * Can get here iff the locking/updating transaction was running when
5305 : : * the infomask was extracted from the tuple, but finished before
5306 : : * TransactionIdIsInProgress got to run. Deal with it as if there was
5307 : : * no locker at all in the first place.
5308 : : */
5309 : 8 : old_infomask |= HEAP_XMAX_INVALID;
5310 : 8 : goto l5;
5311 : : }
5312 : :
5313 : 1932106 : *result_infomask = new_infomask;
5314 : 1932106 : *result_infomask2 = new_infomask2;
5315 : 1932106 : *result_xmax = new_xmax;
5316 : 1932106 : }
5317 : :
5318 : : /*
5319 : : * Subroutine for heap_lock_updated_tuple_rec.
5320 : : *
5321 : : * Given a hypothetical multixact status held by the transaction identified
5322 : : * with the given xid, does the current transaction need to wait, fail, or can
5323 : : * it continue if it wanted to acquire a lock of the given mode? "needwait"
5324 : : * is set to true if waiting is necessary; if it can continue, then TM_Ok is
5325 : : * returned. If the lock is already held by the current transaction, return
5326 : : * TM_SelfModified. In case of a conflict with another transaction, a
5327 : : * different HeapTupleSatisfiesUpdate return code is returned.
5328 : : *
5329 : : * The held status is said to be hypothetical because it might correspond to a
5330 : : * lock held by a single Xid, i.e. not a real MultiXactId; we express it this
5331 : : * way for simplicity of API.
5332 : : */
5333 : : static TM_Result
3791 5334 : 32 : test_lockmode_for_conflict(MultiXactStatus status, TransactionId xid,
5335 : : LockTupleMode mode, HeapTuple tup,
5336 : : bool *needwait)
5337 : : {
5338 : : MultiXactStatus wantedstatus;
5339 : :
5340 : 32 : *needwait = false;
5341 : 32 : wantedstatus = get_mxact_status_for_lock(mode, false);
5342 : :
5343 : : /*
5344 : : * Note: we *must* check TransactionIdIsInProgress before
5345 : : * TransactionIdDidAbort/Commit; see comment at top of heapam_visibility.c
5346 : : * for an explanation.
5347 : : */
5348 [ - + ]: 32 : if (TransactionIdIsCurrentTransactionId(xid))
5349 : : {
5350 : : /*
5351 : : * The tuple has already been locked by our own transaction. This is
5352 : : * very rare but can happen if multiple transactions are trying to
5353 : : * lock an ancient version of the same tuple.
5354 : : */
1849 andres@anarazel.de 5355 :UBC 0 : return TM_SelfModified;
5356 : : }
3791 alvherre@alvh.no-ip. 5357 [ + + ]:CBC 32 : else if (TransactionIdIsInProgress(xid))
5358 : : {
5359 : : /*
5360 : : * If the locking transaction is running, what we do depends on
5361 : : * whether the lock modes conflict: if they do, then we must wait for
5362 : : * it to finish; otherwise we can fall through to lock this tuple
5363 : : * version without waiting.
5364 : : */
5365 [ + + ]: 16 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5366 : 16 : LOCKMODE_from_mxstatus(wantedstatus)))
5367 : : {
5368 : 8 : *needwait = true;
5369 : : }
5370 : :
5371 : : /*
5372 : : * If we set needwait above, then this value doesn't matter;
5373 : : * otherwise, this value signals to caller that it's okay to proceed.
5374 : : */
1849 andres@anarazel.de 5375 : 16 : return TM_Ok;
5376 : : }
3791 alvherre@alvh.no-ip. 5377 [ + + ]: 16 : else if (TransactionIdDidAbort(xid))
1849 andres@anarazel.de 5378 : 3 : return TM_Ok;
3791 alvherre@alvh.no-ip. 5379 [ + - ]: 13 : else if (TransactionIdDidCommit(xid))
5380 : : {
5381 : : /*
5382 : : * The other transaction committed. If it was only a locker, then the
5383 : : * lock is completely gone now and we can return success; but if it
5384 : : * was an update, then what we do depends on whether the two lock
5385 : : * modes conflict. If they conflict, then we must report error to
5386 : : * caller. But if they don't, we can fall through to allow the current
5387 : : * transaction to lock the tuple.
5388 : : *
5389 : : * Note: the reason we worry about ISUPDATE here is because as soon as
5390 : : * a transaction ends, all its locks are gone and meaningless, and
5391 : : * thus we can ignore them; whereas its updates persist. In the
5392 : : * TransactionIdIsInProgress case, above, we don't need to check
5393 : : * because we know the lock is still "alive" and thus a conflict needs
5394 : : * always be checked.
5395 : : */
3783 5396 [ + + ]: 13 : if (!ISUPDATE_from_mxstatus(status))
1849 andres@anarazel.de 5397 : 4 : return TM_Ok;
5398 : :
3791 alvherre@alvh.no-ip. 5399 [ + + ]: 9 : if (DoLockModesConflict(LOCKMODE_from_mxstatus(status),
5400 : 9 : LOCKMODE_from_mxstatus(wantedstatus)))
5401 : : {
5402 : : /* bummer */
1147 5403 [ + + ]: 8 : if (!ItemPointerEquals(&tup->t_self, &tup->t_data->t_ctid))
1849 andres@anarazel.de 5404 : 6 : return TM_Updated;
5405 : : else
5406 : 2 : return TM_Deleted;
5407 : : }
5408 : :
5409 : 1 : return TM_Ok;
5410 : : }
5411 : :
5412 : : /* Not in progress, not aborted, not committed -- must have crashed */
1849 andres@anarazel.de 5413 :UBC 0 : return TM_Ok;
5414 : : }
5415 : :
5416 : :
5417 : : /*
5418 : : * Recursive part of heap_lock_updated_tuple
5419 : : *
5420 : : * Fetch the tuple pointed to by tid in rel, and mark it as locked by the given
5421 : : * xid with the given mode; if this tuple is updated, recurse to lock the new
5422 : : * version as well.
5423 : : */
5424 : : static TM_Result
4099 alvherre@alvh.no-ip. 5425 :CBC 80 : heap_lock_updated_tuple_rec(Relation rel, ItemPointer tid, TransactionId xid,
5426 : : LockTupleMode mode)
5427 : : {
5428 : : TM_Result result;
5429 : : ItemPointerData tupid;
5430 : : HeapTupleData mytup;
5431 : : Buffer buf;
5432 : : uint16 new_infomask,
5433 : : new_infomask2,
5434 : : old_infomask,
5435 : : old_infomask2;
5436 : : TransactionId xmax,
5437 : : new_xmax;
3791 5438 : 80 : TransactionId priorXmax = InvalidTransactionId;
2827 andres@anarazel.de 5439 : 80 : bool cleared_all_frozen = false;
5440 : : bool pinned_desired_page;
5441 : 80 : Buffer vmbuffer = InvalidBuffer;
5442 : : BlockNumber block;
5443 : :
4099 alvherre@alvh.no-ip. 5444 : 80 : ItemPointerCopy(tid, &tupid);
5445 : :
5446 : : for (;;)
5447 : : {
5448 : 83 : new_infomask = 0;
5449 : 83 : new_xmax = InvalidTransactionId;
2827 andres@anarazel.de 5450 : 83 : block = ItemPointerGetBlockNumber(&tupid);
4099 alvherre@alvh.no-ip. 5451 : 83 : ItemPointerCopy(&tupid, &(mytup.t_self));
5452 : :
732 tgl@sss.pgh.pa.us 5453 [ + - ]: 83 : if (!heap_fetch(rel, SnapshotAny, &mytup, &buf, false))
5454 : : {
5455 : : /*
5456 : : * if we fail to find the updated version of the tuple, it's
5457 : : * because it was vacuumed/pruned away after its creator
5458 : : * transaction aborted. So behave as if we got to the end of the
5459 : : * chain, and there's no further tuple to lock: return success to
5460 : : * caller.
5461 : : */
1849 andres@anarazel.de 5462 :UBC 0 : result = TM_Ok;
2235 tgl@sss.pgh.pa.us 5463 : 0 : goto out_unlocked;
5464 : : }
5465 : :
4099 alvherre@alvh.no-ip. 5466 :CBC 83 : l4:
5467 [ - + ]: 91 : CHECK_FOR_INTERRUPTS();
5468 : :
5469 : : /*
5470 : : * Before locking the buffer, pin the visibility map page if it
5471 : : * appears to be necessary. Since we haven't got the lock yet,
5472 : : * someone else might be in the middle of changing this, so we'll need
5473 : : * to recheck after we have the lock.
5474 : : */
2827 andres@anarazel.de 5475 [ - + ]: 91 : if (PageIsAllVisible(BufferGetPage(buf)))
5476 : : {
2827 andres@anarazel.de 5477 :UBC 0 : visibilitymap_pin(rel, block, &vmbuffer);
2235 tgl@sss.pgh.pa.us 5478 : 0 : pinned_desired_page = true;
5479 : : }
5480 : : else
2235 tgl@sss.pgh.pa.us 5481 :CBC 91 : pinned_desired_page = false;
5482 : :
4099 alvherre@alvh.no-ip. 5483 : 91 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5484 : :
5485 : : /*
5486 : : * If we didn't pin the visibility map page and the page has become
5487 : : * all visible while we were busy locking the buffer, we'll have to
5488 : : * unlock and re-lock, to avoid holding the buffer lock across I/O.
5489 : : * That's a bit unfortunate, but hopefully shouldn't happen often.
5490 : : *
5491 : : * Note: in some paths through this function, we will reach here
5492 : : * holding a pin on a vm page that may or may not be the one matching
5493 : : * this page. If this page isn't all-visible, we won't use the vm
5494 : : * page, but we hold onto such a pin till the end of the function.
5495 : : */
2235 tgl@sss.pgh.pa.us 5496 [ + - - + ]: 91 : if (!pinned_desired_page && PageIsAllVisible(BufferGetPage(buf)))
5497 : : {
2810 andres@anarazel.de 5498 :UBC 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
5499 : 0 : visibilitymap_pin(rel, block, &vmbuffer);
5500 : 0 : LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
5501 : : }
5502 : :
5503 : : /*
5504 : : * Check the tuple XMIN against prior XMAX, if any. If we reached the
5505 : : * end of the chain, we're done, so return success.
5506 : : */
3791 alvherre@alvh.no-ip. 5507 [ + + - + ]:CBC 94 : if (TransactionIdIsValid(priorXmax) &&
2355 5508 [ + - ]: 3 : !TransactionIdEquals(HeapTupleHeaderGetXmin(mytup.t_data),
5509 : : priorXmax))
5510 : : {
1849 andres@anarazel.de 5511 :UBC 0 : result = TM_Ok;
2827 5512 : 0 : goto out_locked;
5513 : : }
5514 : :
5515 : : /*
5516 : : * Also check Xmin: if this tuple was created by an aborted
5517 : : * (sub)transaction, then we already locked the last live one in the
5518 : : * chain, thus we're done, so return success.
5519 : : */
2774 alvherre@alvh.no-ip. 5520 [ + - + + ]:CBC 91 : if (TransactionIdDidAbort(HeapTupleHeaderGetXmin(mytup.t_data)))
5521 : : {
1849 andres@anarazel.de 5522 : 13 : result = TM_Ok;
2235 tgl@sss.pgh.pa.us 5523 : 13 : goto out_locked;
5524 : : }
5525 : :
4099 alvherre@alvh.no-ip. 5526 : 78 : old_infomask = mytup.t_data->t_infomask;
3791 5527 : 78 : old_infomask2 = mytup.t_data->t_infomask2;
4099 5528 : 78 : xmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5529 : :
5530 : : /*
5531 : : * If this tuple version has been updated or locked by some concurrent
5532 : : * transaction(s), what we do depends on whether our lock mode
5533 : : * conflicts with what those other transactions hold, and also on the
5534 : : * status of them.
5535 : : */
3791 5536 [ + + ]: 78 : if (!(old_infomask & HEAP_XMAX_INVALID))
5537 : : {
5538 : : TransactionId rawxmax;
5539 : : bool needwait;
5540 : :
5541 : 30 : rawxmax = HeapTupleHeaderGetRawXmax(mytup.t_data);
5542 [ + + ]: 30 : if (old_infomask & HEAP_XMAX_IS_MULTI)
5543 : : {
5544 : : int nmembers;
5545 : : int i;
5546 : : MultiXactMember *members;
5547 : :
5548 : : /*
5549 : : * We don't need a test for pg_upgrade'd tuples: this is only
5550 : : * applied to tuples after the first in an update chain. Said
5551 : : * first tuple in the chain may well be locked-in-9.2-and-
5552 : : * pg_upgraded, but that one was already locked by our caller,
5553 : : * not us; and any subsequent ones cannot be because our
5554 : : * caller must necessarily have obtained a snapshot later than
5555 : : * the pg_upgrade itself.
5556 : : */
2851 5557 [ + - - + : 1 : Assert(!HEAP_LOCKED_UPGRADED(mytup.t_data->t_infomask));
- - ]
5558 : :
3547 5559 : 1 : nmembers = GetMultiXactIdMembers(rawxmax, &members, false,
2489 tgl@sss.pgh.pa.us 5560 [ + - - + ]: 1 : HEAP_XMAX_IS_LOCKED_ONLY(old_infomask));
3791 alvherre@alvh.no-ip. 5561 [ + + ]: 4 : for (i = 0; i < nmembers; i++)
5562 : : {
2827 andres@anarazel.de 5563 : 3 : result = test_lockmode_for_conflict(members[i].status,
5564 : 3 : members[i].xid,
5565 : : mode,
5566 : : &mytup,
5567 : : &needwait);
5568 : :
5569 : : /*
5570 : : * If the tuple was already locked by ourselves in a
5571 : : * previous iteration of this (say heap_lock_tuple was
5572 : : * forced to restart the locking loop because of a change
5573 : : * in xmax), then we hold the lock already on this tuple
5574 : : * version and we don't need to do anything; and this is
5575 : : * not an error condition either. We just need to skip
5576 : : * this tuple and continue locking the next version in the
5577 : : * update chain.
5578 : : */
1849 5579 [ - + ]: 3 : if (result == TM_SelfModified)
5580 : : {
2454 alvherre@alvh.no-ip. 5581 :UBC 0 : pfree(members);
5582 : 0 : goto next;
5583 : : }
5584 : :
3791 alvherre@alvh.no-ip. 5585 [ - + ]:CBC 3 : if (needwait)
5586 : : {
3791 alvherre@alvh.no-ip. 5587 :UBC 0 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
3679 5588 : 0 : XactLockTableWait(members[i].xid, rel,
5589 : : &mytup.t_self,
5590 : : XLTW_LockUpdated);
3791 5591 : 0 : pfree(members);
5592 : 0 : goto l4;
5593 : : }
1849 andres@anarazel.de 5594 [ - + ]:CBC 3 : if (result != TM_Ok)
5595 : : {
3791 alvherre@alvh.no-ip. 5596 :UBC 0 : pfree(members);
2827 andres@anarazel.de 5597 : 0 : goto out_locked;
5598 : : }
5599 : : }
3791 alvherre@alvh.no-ip. 5600 [ + - ]:CBC 1 : if (members)
5601 : 1 : pfree(members);
5602 : : }
5603 : : else
5604 : : {
5605 : : MultiXactStatus status;
5606 : :
5607 : : /*
5608 : : * For a non-multi Xmax, we first need to compute the
5609 : : * corresponding MultiXactStatus by using the infomask bits.
5610 : : */
5611 [ + + - + ]: 29 : if (HEAP_XMAX_IS_LOCKED_ONLY(old_infomask))
5612 : : {
5613 [ + - ]: 10 : if (HEAP_XMAX_IS_KEYSHR_LOCKED(old_infomask))
5614 : 10 : status = MultiXactStatusForKeyShare;
3791 alvherre@alvh.no-ip. 5615 [ # # ]:UBC 0 : else if (HEAP_XMAX_IS_SHR_LOCKED(old_infomask))
5616 : 0 : status = MultiXactStatusForShare;
5617 [ # # ]: 0 : else if (HEAP_XMAX_IS_EXCL_LOCKED(old_infomask))
5618 : : {
5619 [ # # ]: 0 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5620 : 0 : status = MultiXactStatusForUpdate;
5621 : : else
5622 : 0 : status = MultiXactStatusForNoKeyUpdate;
5623 : : }
5624 : : else
5625 : : {
5626 : : /*
5627 : : * LOCK_ONLY present alone (a pg_upgraded tuple marked
5628 : : * as share-locked in the old cluster) shouldn't be
5629 : : * seen in the middle of an update chain.
5630 : : */
5631 [ # # ]: 0 : elog(ERROR, "invalid lock status in tuple");
5632 : : }
5633 : : }
5634 : : else
5635 : : {
5636 : : /* it's an update, but which kind? */
3791 alvherre@alvh.no-ip. 5637 [ + + ]:CBC 19 : if (old_infomask2 & HEAP_KEYS_UPDATED)
5638 : 14 : status = MultiXactStatusUpdate;
5639 : : else
5640 : 5 : status = MultiXactStatusNoKeyUpdate;
5641 : : }
5642 : :
2827 andres@anarazel.de 5643 : 29 : result = test_lockmode_for_conflict(status, rawxmax, mode,
5644 : : &mytup, &needwait);
5645 : :
5646 : : /*
5647 : : * If the tuple was already locked by ourselves in a previous
5648 : : * iteration of this (say heap_lock_tuple was forced to
5649 : : * restart the locking loop because of a change in xmax), then
5650 : : * we hold the lock already on this tuple version and we don't
5651 : : * need to do anything; and this is not an error condition
5652 : : * either. We just need to skip this tuple and continue
5653 : : * locking the next version in the update chain.
5654 : : */
1849 5655 [ - + ]: 29 : if (result == TM_SelfModified)
2454 alvherre@alvh.no-ip. 5656 :UBC 0 : goto next;
5657 : :
3791 alvherre@alvh.no-ip. 5658 [ + + ]:CBC 29 : if (needwait)
5659 : : {
5660 : 8 : LockBuffer(buf, BUFFER_LOCK_UNLOCK);
3357 heikki.linnakangas@i 5661 : 8 : XactLockTableWait(rawxmax, rel, &mytup.t_self,
5662 : : XLTW_LockUpdated);
3791 alvherre@alvh.no-ip. 5663 : 8 : goto l4;
5664 : : }
1849 andres@anarazel.de 5665 [ + + ]: 21 : if (result != TM_Ok)
5666 : : {
2827 5667 : 8 : goto out_locked;
5668 : : }
5669 : : }
5670 : : }
5671 : :
5672 : : /* compute the new Xmax and infomask values for the tuple ... */
4099 alvherre@alvh.no-ip. 5673 : 62 : compute_new_xmax_infomask(xmax, old_infomask, mytup.t_data->t_infomask2,
5674 : : xid, mode, false,
5675 : : &new_xmax, &new_infomask, &new_infomask2);
5676 : :
2827 andres@anarazel.de 5677 [ - + - - ]: 62 : if (PageIsAllVisible(BufferGetPage(buf)) &&
2827 andres@anarazel.de 5678 :UBC 0 : visibilitymap_clear(rel, block, vmbuffer,
5679 : : VISIBILITYMAP_ALL_FROZEN))
5680 : 0 : cleared_all_frozen = true;
5681 : :
4099 alvherre@alvh.no-ip. 5682 :CBC 62 : START_CRIT_SECTION();
5683 : :
5684 : : /* ... and set them */
5685 : 62 : HeapTupleHeaderSetXmax(mytup.t_data, new_xmax);
5686 : 62 : mytup.t_data->t_infomask &= ~HEAP_XMAX_BITS;
5687 : 62 : mytup.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5688 : 62 : mytup.t_data->t_infomask |= new_infomask;
5689 : 62 : mytup.t_data->t_infomask2 |= new_infomask2;
5690 : :
5691 : 62 : MarkBufferDirty(buf);
5692 : :
5693 : : /* XLOG stuff */
5694 [ + - - + : 62 : if (RelationNeedsWAL(rel))
- - - - ]
5695 : : {
5696 : : xl_heap_lock_updated xlrec;
5697 : : XLogRecPtr recptr;
2916 kgrittn@postgresql.o 5698 : 62 : Page page = BufferGetPage(buf);
5699 : :
3433 heikki.linnakangas@i 5700 : 62 : XLogBeginInsert();
5701 : 62 : XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
5702 : :
5703 : 62 : xlrec.offnum = ItemPointerGetOffsetNumber(&mytup.t_self);
4099 alvherre@alvh.no-ip. 5704 : 62 : xlrec.xmax = new_xmax;
5705 : 62 : xlrec.infobits_set = compute_infobits(new_infomask, new_infomask2);
2827 andres@anarazel.de 5706 : 62 : xlrec.flags =
5707 : 62 : cleared_all_frozen ? XLH_LOCK_ALL_FROZEN_CLEARED : 0;
5708 : :
3433 heikki.linnakangas@i 5709 : 62 : XLogRegisterData((char *) &xlrec, SizeOfHeapLockUpdated);
5710 : :
5711 : 62 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_LOCK_UPDATED);
5712 : :
4099 alvherre@alvh.no-ip. 5713 : 62 : PageSetLSN(page, recptr);
5714 : : }
5715 : :
5716 [ - + ]: 62 : END_CRIT_SECTION();
5717 : :
2454 5718 : 62 : next:
5719 : : /* if we find the end of update chain, we're done. */
4099 5720 [ + - + - ]: 124 : if (mytup.t_data->t_infomask & HEAP_XMAX_INVALID ||
2199 andres@anarazel.de 5721 [ + + ]: 124 : HeapTupleHeaderIndicatesMovedPartitions(mytup.t_data) ||
3973 bruce@momjian.us 5722 [ + + ]: 66 : ItemPointerEquals(&mytup.t_self, &mytup.t_data->t_ctid) ||
4099 alvherre@alvh.no-ip. 5723 : 4 : HeapTupleHeaderIsOnlyLocked(mytup.t_data))
5724 : : {
1849 andres@anarazel.de 5725 : 59 : result = TM_Ok;
2827 5726 : 59 : goto out_locked;
5727 : : }
5728 : :
5729 : : /* tail recursion */
3791 alvherre@alvh.no-ip. 5730 [ + - + - : 3 : priorXmax = HeapTupleHeaderGetUpdateXid(mytup.t_data);
+ - ]
4099 5731 : 3 : ItemPointerCopy(&(mytup.t_data->t_ctid), &tupid);
5732 : 3 : UnlockReleaseBuffer(buf);
5733 : : }
5734 : :
5735 : : result = TM_Ok;
5736 : :
2827 andres@anarazel.de 5737 : 80 : out_locked:
5738 : 80 : UnlockReleaseBuffer(buf);
5739 : :
2235 tgl@sss.pgh.pa.us 5740 : 80 : out_unlocked:
2827 andres@anarazel.de 5741 [ - + ]: 80 : if (vmbuffer != InvalidBuffer)
2827 andres@anarazel.de 5742 :UBC 0 : ReleaseBuffer(vmbuffer);
5743 : :
2827 andres@anarazel.de 5744 :CBC 80 : return result;
5745 : : }
5746 : :
5747 : : /*
5748 : : * heap_lock_updated_tuple
5749 : : * Follow update chain when locking an updated tuple, acquiring locks (row
5750 : : * marks) on the updated versions.
5751 : : *
5752 : : * The initial tuple is assumed to be already locked.
5753 : : *
5754 : : * This function doesn't check visibility, it just unconditionally marks the
5755 : : * tuple(s) as locked. If any tuple in the updated chain is being deleted
5756 : : * concurrently (or updated with the key being modified), sleep until the
5757 : : * transaction doing it is finished.
5758 : : *
5759 : : * Note that we don't acquire heavyweight tuple locks on the tuples we walk
5760 : : * when we have to wait for other transactions to release them, as opposed to
5761 : : * what heap_lock_tuple does. The reason is that having more than one
5762 : : * transaction walking the chain is probably uncommon enough that risk of
5763 : : * starvation is not likely: one of the preconditions for being here is that
5764 : : * the snapshot in use predates the update that created this tuple (because we
5765 : : * started at an earlier version of the tuple), but at the same time such a
5766 : : * transaction cannot be using repeatable read or serializable isolation
5767 : : * levels, because that would lead to a serializability failure.
5768 : : */
5769 : : static TM_Result
4099 alvherre@alvh.no-ip. 5770 : 89 : heap_lock_updated_tuple(Relation rel, HeapTuple tuple, ItemPointer ctid,
5771 : : TransactionId xid, LockTupleMode mode)
5772 : : {
5773 : : /*
5774 : : * If the tuple has not been updated, or has moved into another partition
5775 : : * (effectively a delete) stop here.
5776 : : */
2199 andres@anarazel.de 5777 [ + + ]: 89 : if (!HeapTupleHeaderIndicatesMovedPartitions(tuple->t_data) &&
5778 [ + + ]: 87 : !ItemPointerEquals(&tuple->t_self, ctid))
5779 : : {
5780 : : /*
5781 : : * If this is the first possibly-multixact-able operation in the
5782 : : * current transaction, set my per-backend OldestMemberMXactId
5783 : : * setting. We can be certain that the transaction will never become a
5784 : : * member of any older MultiXactIds than that. (We have to do this
5785 : : * even if we end up just using our own TransactionId below, since
5786 : : * some other backend could incorporate our XID into a MultiXact
5787 : : * immediately afterwards.)
5788 : : */
4099 alvherre@alvh.no-ip. 5789 : 80 : MultiXactIdSetOldestMember();
5790 : :
5791 : 80 : return heap_lock_updated_tuple_rec(rel, ctid, xid, mode);
5792 : : }
5793 : :
5794 : : /* nothing to lock */
1849 andres@anarazel.de 5795 : 9 : return TM_Ok;
5796 : : }
5797 : :
5798 : : /*
5799 : : * heap_finish_speculative - mark speculative insertion as successful
5800 : : *
5801 : : * To successfully finish a speculative insertion we have to clear speculative
5802 : : * token from tuple. To do so the t_ctid field, which will contain a
5803 : : * speculative token value, is modified in place to point to the tuple itself,
5804 : : * which is characteristic of a newly inserted ordinary tuple.
5805 : : *
5806 : : * NB: It is not ok to commit without either finishing or aborting a
5807 : : * speculative insertion. We could treat speculative tuples of committed
5808 : : * transactions implicitly as completed, but then we would have to be prepared
5809 : : * to deal with speculative tokens on committed tuples. That wouldn't be
5810 : : * difficult - no-one looks at the ctid field of a tuple with invalid xmax -
5811 : : * but clearing the token at completion isn't very expensive either.
5812 : : * An explicit confirmation WAL record also makes logical decoding simpler.
5813 : : */
5814 : : void
5815 : 2005 : heap_finish_speculative(Relation relation, ItemPointer tid)
5816 : : {
5817 : : Buffer buffer;
5818 : : Page page;
5819 : : OffsetNumber offnum;
3264 5820 : 2005 : ItemId lp = NULL;
5821 : : HeapTupleHeader htup;
5822 : :
1849 5823 : 2005 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(tid));
3264 5824 : 2005 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2916 kgrittn@postgresql.o 5825 : 2005 : page = (Page) BufferGetPage(buffer);
5826 : :
1849 andres@anarazel.de 5827 : 2005 : offnum = ItemPointerGetOffsetNumber(tid);
3264 5828 [ + - ]: 2005 : if (PageGetMaxOffsetNumber(page) >= offnum)
5829 : 2005 : lp = PageGetItemId(page, offnum);
5830 : :
5831 [ + - - + ]: 2005 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
3069 andres@anarazel.de 5832 [ # # ]:UBC 0 : elog(ERROR, "invalid lp");
5833 : :
3264 andres@anarazel.de 5834 :CBC 2005 : htup = (HeapTupleHeader) PageGetItem(page, lp);
5835 : :
5836 : : /* NO EREPORT(ERROR) from here till changes are logged */
5837 : 2005 : START_CRIT_SECTION();
5838 : :
1849 5839 [ - + ]: 2005 : Assert(HeapTupleHeaderIsSpeculative(htup));
5840 : :
3264 5841 : 2005 : MarkBufferDirty(buffer);
5842 : :
5843 : : /*
5844 : : * Replace the speculative insertion token with a real t_ctid, pointing to
5845 : : * itself like it does on regular tuples.
5846 : : */
1849 5847 : 2005 : htup->t_ctid = *tid;
5848 : :
5849 : : /* XLOG stuff */
3264 5850 [ + + + + : 2005 : if (RelationNeedsWAL(relation))
+ - + - ]
5851 : : {
5852 : : xl_heap_confirm xlrec;
5853 : : XLogRecPtr recptr;
5854 : :
1849 5855 : 1999 : xlrec.offnum = ItemPointerGetOffsetNumber(tid);
5856 : :
3264 5857 : 1999 : XLogBeginInsert();
5858 : :
5859 : : /* We want the same filtering on this as on a plain insert */
2670 5860 : 1999 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
5861 : :
3264 5862 : 1999 : XLogRegisterData((char *) &xlrec, SizeOfHeapConfirm);
5863 : 1999 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
5864 : :
5865 : 1999 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_CONFIRM);
5866 : :
5867 : 1999 : PageSetLSN(page, recptr);
5868 : : }
5869 : :
5870 [ - + ]: 2005 : END_CRIT_SECTION();
5871 : :
5872 : 2005 : UnlockReleaseBuffer(buffer);
5873 : 2005 : }
5874 : :
5875 : : /*
5876 : : * heap_abort_speculative - kill a speculatively inserted tuple
5877 : : *
5878 : : * Marks a tuple that was speculatively inserted in the same command as dead,
5879 : : * by setting its xmin as invalid. That makes it immediately appear as dead
5880 : : * to all transactions, including our own. In particular, it makes
5881 : : * HeapTupleSatisfiesDirty() regard the tuple as dead, so that another backend
5882 : : * inserting a duplicate key value won't unnecessarily wait for our whole
5883 : : * transaction to finish (it'll just wait for our speculative insertion to
5884 : : * finish).
5885 : : *
5886 : : * Killing the tuple prevents "unprincipled deadlocks", which are deadlocks
5887 : : * that arise due to a mutual dependency that is not user visible. By
5888 : : * definition, unprincipled deadlocks cannot be prevented by the user
5889 : : * reordering lock acquisition in client code, because the implementation level
5890 : : * lock acquisitions are not under the user's direct control. If speculative
5891 : : * inserters did not take this precaution, then under high concurrency they
5892 : : * could deadlock with each other, which would not be acceptable.
5893 : : *
5894 : : * This is somewhat redundant with heap_delete, but we prefer to have a
5895 : : * dedicated routine with stripped down requirements. Note that this is also
5896 : : * used to delete the TOAST tuples created during speculative insertion.
5897 : : *
5898 : : * This routine does not affect logical decoding as it only looks at
5899 : : * confirmation records.
5900 : : */
5901 : : void
1849 5902 : 10 : heap_abort_speculative(Relation relation, ItemPointer tid)
5903 : : {
3264 5904 : 10 : TransactionId xid = GetCurrentTransactionId();
5905 : : ItemId lp;
5906 : : HeapTupleData tp;
5907 : : Page page;
5908 : : BlockNumber block;
5909 : : Buffer buffer;
5910 : : TransactionId prune_xid;
5911 : :
5912 [ - + ]: 10 : Assert(ItemPointerIsValid(tid));
5913 : :
5914 : 10 : block = ItemPointerGetBlockNumber(tid);
5915 : 10 : buffer = ReadBuffer(relation, block);
2916 kgrittn@postgresql.o 5916 : 10 : page = BufferGetPage(buffer);
5917 : :
3264 andres@anarazel.de 5918 : 10 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
5919 : :
5920 : : /*
5921 : : * Page can't be all visible, we just inserted into it, and are still
5922 : : * running.
5923 : : */
5924 [ - + ]: 10 : Assert(!PageIsAllVisible(page));
5925 : :
5926 : 10 : lp = PageGetItemId(page, ItemPointerGetOffsetNumber(tid));
5927 [ - + ]: 10 : Assert(ItemIdIsNormal(lp));
5928 : :
5929 : 10 : tp.t_tableOid = RelationGetRelid(relation);
5930 : 10 : tp.t_data = (HeapTupleHeader) PageGetItem(page, lp);
5931 : 10 : tp.t_len = ItemIdGetLength(lp);
5932 : 10 : tp.t_self = *tid;
5933 : :
5934 : : /*
5935 : : * Sanity check that the tuple really is a speculatively inserted tuple,
5936 : : * inserted by us.
5937 : : */
5938 [ - + ]: 10 : if (tp.t_data->t_choice.t_heap.t_xmin != xid)
3264 andres@anarazel.de 5939 [ # # ]:UBC 0 : elog(ERROR, "attempted to kill a tuple inserted by another transaction");
2797 andres@anarazel.de 5940 [ + + - + ]:CBC 10 : if (!(IsToastRelation(relation) || HeapTupleHeaderIsSpeculative(tp.t_data)))
3264 andres@anarazel.de 5941 [ # # ]:UBC 0 : elog(ERROR, "attempted to kill a non-speculative tuple");
3264 andres@anarazel.de 5942 [ - + ]:CBC 10 : Assert(!HeapTupleHeaderIsHeapOnly(tp.t_data));
5943 : :
5944 : : /*
5945 : : * No need to check for serializable conflicts here. There is never a
5946 : : * need for a combo CID, either. No need to extract replica identity, or
5947 : : * do anything special with infomask bits.
5948 : : */
5949 : :
5950 : 10 : START_CRIT_SECTION();
5951 : :
5952 : : /*
5953 : : * The tuple will become DEAD immediately. Flag that this page is a
5954 : : * candidate for pruning by setting xmin to TransactionXmin. While not
5955 : : * immediately prunable, it is the oldest xid we can cheaply determine
5956 : : * that's safe against wraparound / being older than the table's
5957 : : * relfrozenxid. To defend against the unlikely case of a new relation
5958 : : * having a newer relfrozenxid than our TransactionXmin, use relfrozenxid
5959 : : * if so (vacuum can't subsequently move relfrozenxid to beyond
5960 : : * TransactionXmin, so there's no race here).
5961 : : */
1470 5962 [ - + ]: 10 : Assert(TransactionIdIsValid(TransactionXmin));
5963 [ - + ]: 10 : if (TransactionIdPrecedes(TransactionXmin, relation->rd_rel->relfrozenxid))
1470 andres@anarazel.de 5964 :UBC 0 : prune_xid = relation->rd_rel->relfrozenxid;
5965 : : else
1470 andres@anarazel.de 5966 :CBC 10 : prune_xid = TransactionXmin;
5967 [ - + + + : 10 : PageSetPrunable(page, prune_xid);
- + ]
5968 : :
5969 : : /* store transaction information of xact deleting the tuple */
3264 5970 : 10 : tp.t_data->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
5971 : 10 : tp.t_data->t_infomask2 &= ~HEAP_KEYS_UPDATED;
5972 : :
5973 : : /*
5974 : : * Set the tuple header xmin to InvalidTransactionId. This makes the
5975 : : * tuple immediately invisible everyone. (In particular, to any
5976 : : * transactions waiting on the speculative token, woken up later.)
5977 : : */
5978 : 10 : HeapTupleHeaderSetXmin(tp.t_data, InvalidTransactionId);
5979 : :
5980 : : /* Clear the speculative insertion token too */
5981 : 10 : tp.t_data->t_ctid = tp.t_self;
5982 : :
5983 : 10 : MarkBufferDirty(buffer);
5984 : :
5985 : : /*
5986 : : * XLOG stuff
5987 : : *
5988 : : * The WAL records generated here match heap_delete(). The same recovery
5989 : : * routines are used.
5990 : : */
5991 [ + - - + : 10 : if (RelationNeedsWAL(relation))
- - - - ]
5992 : : {
5993 : : xl_heap_delete xlrec;
5994 : : XLogRecPtr recptr;
5995 : :
5996 : 10 : xlrec.flags = XLH_DELETE_IS_SUPER;
5997 : 20 : xlrec.infobits_set = compute_infobits(tp.t_data->t_infomask,
5998 : 10 : tp.t_data->t_infomask2);
5999 : 10 : xlrec.offnum = ItemPointerGetOffsetNumber(&tp.t_self);
6000 : 10 : xlrec.xmax = xid;
6001 : :
6002 : 10 : XLogBeginInsert();
6003 : 10 : XLogRegisterData((char *) &xlrec, SizeOfHeapDelete);
6004 : 10 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6005 : :
6006 : : /* No replica identity & replication origin logged */
6007 : :
6008 : 10 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_DELETE);
6009 : :
6010 : 10 : PageSetLSN(page, recptr);
6011 : : }
6012 : :
6013 [ - + ]: 10 : END_CRIT_SECTION();
6014 : :
6015 : 10 : LockBuffer(buffer, BUFFER_LOCK_UNLOCK);
6016 : :
6017 [ + + ]: 10 : if (HeapTupleHasExternal(&tp))
6018 : : {
2797 6019 [ - + ]: 1 : Assert(!IsToastRelation(relation));
1654 rhaas@postgresql.org 6020 : 1 : heap_toast_delete(relation, &tp, true);
6021 : : }
6022 : :
6023 : : /*
6024 : : * Never need to mark tuple for invalidation, since catalogs don't support
6025 : : * speculative insertion
6026 : : */
6027 : :
6028 : : /* Now we can release the buffer */
3264 andres@anarazel.de 6029 : 10 : ReleaseBuffer(buffer);
6030 : :
6031 : : /* count deletion, as we counted the insertion too */
6032 : 10 : pgstat_count_heap_delete(relation);
6033 : 10 : }
6034 : :
6035 : : /*
6036 : : * heap_inplace_update - update a tuple "in place" (ie, overwrite it)
6037 : : *
6038 : : * Overwriting violates both MVCC and transactional safety, so the uses
6039 : : * of this function in Postgres are extremely limited. Nonetheless we
6040 : : * find some places to use it.
6041 : : *
6042 : : * The tuple cannot change size, and therefore it's reasonable to assume
6043 : : * that its null bitmap (if any) doesn't change either. So we just
6044 : : * overwrite the data portion of the tuple without touching the null
6045 : : * bitmap or any of the header fields.
6046 : : *
6047 : : * tuple is an in-memory tuple structure containing the data to be written
6048 : : * over the target tuple. Also, tuple->t_self identifies the target tuple.
6049 : : *
6050 : : * Note that the tuple updated here had better not come directly from the
6051 : : * syscache if the relation has a toast relation as this tuple could
6052 : : * include toast values that have been expanded, causing a failure here.
6053 : : */
6054 : : void
6549 tgl@sss.pgh.pa.us 6055 : 71114 : heap_inplace_update(Relation relation, HeapTuple tuple)
6056 : : {
6057 : : Buffer buffer;
6058 : : Page page;
6059 : : OffsetNumber offnum;
6060 : 71114 : ItemId lp = NULL;
6061 : : HeapTupleHeader htup;
6062 : : uint32 oldlen;
6063 : : uint32 newlen;
6064 : :
6065 : : /*
6066 : : * For now, we don't allow parallel updates. Unlike a regular update,
6067 : : * this should never create a combo CID, so it might be possible to relax
6068 : : * this restriction, but not without more thought and testing. It's not
6069 : : * clear that it would be useful, anyway.
6070 : : */
3272 rhaas@postgresql.org 6071 [ - + ]: 71114 : if (IsInParallelMode())
3272 rhaas@postgresql.org 6072 [ # # ]:UBC 0 : ereport(ERROR,
6073 : : (errcode(ERRCODE_INVALID_TRANSACTION_STATE),
6074 : : errmsg("cannot update tuples during a parallel operation")));
6075 : :
6549 tgl@sss.pgh.pa.us 6076 :CBC 71114 : buffer = ReadBuffer(relation, ItemPointerGetBlockNumber(&(tuple->t_self)));
6077 : 71114 : LockBuffer(buffer, BUFFER_LOCK_EXCLUSIVE);
2916 kgrittn@postgresql.o 6078 : 71114 : page = (Page) BufferGetPage(buffer);
6079 : :
6549 tgl@sss.pgh.pa.us 6080 : 71114 : offnum = ItemPointerGetOffsetNumber(&(tuple->t_self));
6081 [ + - ]: 71114 : if (PageGetMaxOffsetNumber(page) >= offnum)
6082 : 71114 : lp = PageGetItemId(page, offnum);
6083 : :
6059 6084 [ + - - + ]: 71114 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
3069 andres@anarazel.de 6085 [ # # ]:UBC 0 : elog(ERROR, "invalid lp");
6086 : :
6549 tgl@sss.pgh.pa.us 6087 :CBC 71114 : htup = (HeapTupleHeader) PageGetItem(page, lp);
6088 : :
6089 : 71114 : oldlen = ItemIdGetLength(lp) - htup->t_hoff;
6090 : 71114 : newlen = tuple->t_len - tuple->t_data->t_hoff;
6091 [ + - - + ]: 71114 : if (oldlen != newlen || htup->t_hoff != tuple->t_data->t_hoff)
3069 andres@anarazel.de 6092 [ # # ]:UBC 0 : elog(ERROR, "wrong tuple length");
6093 : :
6094 : : /* NO EREPORT(ERROR) from here till changes are logged */
6549 tgl@sss.pgh.pa.us 6095 :CBC 71114 : START_CRIT_SECTION();
6096 : :
6097 : 71114 : memcpy((char *) htup + htup->t_hoff,
6098 : 71114 : (char *) tuple->t_data + tuple->t_data->t_hoff,
6099 : : newlen);
6100 : :
6101 : 71114 : MarkBufferDirty(buffer);
6102 : :
6103 : : /* XLOG stuff */
4871 rhaas@postgresql.org 6104 [ + - + + : 71114 : if (RelationNeedsWAL(relation))
+ - + + ]
6105 : : {
6106 : : xl_heap_inplace xlrec;
6107 : : XLogRecPtr recptr;
6108 : :
3433 heikki.linnakangas@i 6109 : 71106 : xlrec.offnum = ItemPointerGetOffsetNumber(&tuple->t_self);
6110 : :
6111 : 71106 : XLogBeginInsert();
6112 : 71106 : XLogRegisterData((char *) &xlrec, SizeOfHeapInplace);
6113 : :
6114 : 71106 : XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
6115 : 71106 : XLogRegisterBufData(0, (char *) htup + htup->t_hoff, newlen);
6116 : :
6117 : : /* inplace updates aren't decoded atm, don't log the origin */
6118 : :
6119 : 71106 : recptr = XLogInsert(RM_HEAP_ID, XLOG_HEAP_INPLACE);
6120 : :
6549 tgl@sss.pgh.pa.us 6121 : 71106 : PageSetLSN(page, recptr);
6122 : : }
6123 : :
6124 [ - + ]: 71114 : END_CRIT_SECTION();
6125 : :
6126 : 71114 : UnlockReleaseBuffer(buffer);
6127 : :
6128 : : /*
6129 : : * Send out shared cache inval if necessary. Note that because we only
6130 : : * pass the new version of the tuple, this mustn't be used for any
6131 : : * operations that could change catcache lookup keys. But we aren't
6132 : : * bothering with index updates either, so that's true a fortiori.
6133 : : */
6134 [ + + ]: 71114 : if (!IsBootstrapProcessingMode())
4625 6135 : 59570 : CacheInvalidateHeapTuple(relation, tuple, NULL);
6549 6136 : 71114 : }
6137 : :
6138 : : #define FRM_NOOP 0x0001
6139 : : #define FRM_INVALIDATE_XMAX 0x0002
6140 : : #define FRM_RETURN_IS_XID 0x0004
6141 : : #define FRM_RETURN_IS_MULTI 0x0008
6142 : : #define FRM_MARK_COMMITTED 0x0010
6143 : :
6144 : : /*
6145 : : * FreezeMultiXactId
6146 : : * Determine what to do during freezing when a tuple is marked by a
6147 : : * MultiXactId.
6148 : : *
6149 : : * "flags" is an output value; it's used to tell caller what to do on return.
6150 : : * "pagefrz" is an input/output value, used to manage page level freezing.
6151 : : *
6152 : : * Possible values that we can set in "flags":
6153 : : * FRM_NOOP
6154 : : * don't do anything -- keep existing Xmax
6155 : : * FRM_INVALIDATE_XMAX
6156 : : * mark Xmax as InvalidTransactionId and set XMAX_INVALID flag.
6157 : : * FRM_RETURN_IS_XID
6158 : : * The Xid return value is a single update Xid to set as xmax.
6159 : : * FRM_MARK_COMMITTED
6160 : : * Xmax can be marked as HEAP_XMAX_COMMITTED
6161 : : * FRM_RETURN_IS_MULTI
6162 : : * The return value is a new MultiXactId to set as new Xmax.
6163 : : * (caller must obtain proper infomask bits using GetMultiXactIdHintBits)
6164 : : *
6165 : : * Caller delegates control of page freezing to us. In practice we always
6166 : : * force freezing of caller's page unless FRM_NOOP processing is indicated.
6167 : : * We help caller ensure that XIDs < FreezeLimit and MXIDs < MultiXactCutoff
6168 : : * can never be left behind. We freely choose when and how to process each
6169 : : * Multi, without ever violating the cutoff postconditions for freezing.
6170 : : *
6171 : : * It's useful to remove Multis on a proactive timeline (relative to freezing
6172 : : * XIDs) to keep MultiXact member SLRU buffer misses to a minimum. It can also
6173 : : * be cheaper in the short run, for us, since we too can avoid SLRU buffer
6174 : : * misses through eager processing.
6175 : : *
6176 : : * NB: Creates a _new_ MultiXactId when FRM_RETURN_IS_MULTI is set, though only
6177 : : * when FreezeLimit and/or MultiXactCutoff cutoffs leave us with no choice.
6178 : : * This can usually be put off, which is usually enough to avoid it altogether.
6179 : : * Allocating new multis during VACUUM should be avoided on general principle;
6180 : : * only VACUUM can advance relminmxid, so allocating new Multis here comes with
6181 : : * its own special risks.
6182 : : *
6183 : : * NB: Caller must maintain "no freeze" NewRelfrozenXid/NewRelminMxid trackers
6184 : : * using heap_tuple_should_freeze when we haven't forced page-level freezing.
6185 : : *
6186 : : * NB: Caller should avoid needlessly calling heap_tuple_should_freeze when we
6187 : : * have already forced page-level freezing, since that might incur the same
6188 : : * SLRU buffer misses that we specifically intended to avoid by freezing.
6189 : : */
6190 : : static TransactionId
3772 alvherre@alvh.no-ip. 6191 : 9 : FreezeMultiXactId(MultiXactId multi, uint16 t_infomask,
6192 : : const struct VacuumCutoffs *cutoffs, uint16 *flags,
6193 : : HeapPageFreeze *pagefrz)
6194 : : {
6195 : : TransactionId newxmax;
6196 : : MultiXactMember *members;
6197 : : int nmembers;
6198 : : bool need_replace;
6199 : : int nnewmembers;
6200 : : MultiXactMember *newmembers;
6201 : : bool has_lockers;
6202 : : TransactionId update_xid;
6203 : : bool update_committed;
6204 : : TransactionId FreezePageRelfrozenXid;
6205 : :
6206 : 9 : *flags = 0;
6207 : :
6208 : : /* We should only be called in Multis */
6209 [ - + ]: 9 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
6210 : :
2851 6211 [ + - ]: 9 : if (!MultiXactIdIsValid(multi) ||
6212 [ + - + - : 9 : HEAP_LOCKED_UPGRADED(t_infomask))
- + ]
6213 : : {
3772 alvherre@alvh.no-ip. 6214 :UBC 0 : *flags |= FRM_INVALIDATE_XMAX;
473 pg@bowt.ie 6215 : 0 : pagefrz->freeze_required = true;
3772 alvherre@alvh.no-ip. 6216 : 0 : return InvalidTransactionId;
6217 : : }
479 pg@bowt.ie 6218 [ - + ]:CBC 9 : else if (MultiXactIdPrecedes(multi, cutoffs->relminmxid))
2344 andres@anarazel.de 6219 [ # # ]:UBC 0 : ereport(ERROR,
6220 : : (errcode(ERRCODE_DATA_CORRUPTED),
6221 : : errmsg_internal("found multixact %u from before relminmxid %u",
6222 : : multi, cutoffs->relminmxid)));
473 pg@bowt.ie 6223 [ + + ]:CBC 9 : else if (MultiXactIdPrecedes(multi, cutoffs->OldestMxact))
6224 : : {
6225 : : TransactionId update_xact;
6226 : :
6227 : : /*
6228 : : * This old multi cannot possibly have members still running, but
6229 : : * verify just in case. If it was a locker only, it can be removed
6230 : : * without any further consideration; but if it contained an update,
6231 : : * we might need to preserve it.
6232 : : */
2344 andres@anarazel.de 6233 [ - + ]: 7 : if (MultiXactIdIsRunning(multi,
6234 [ - + - - ]: 7 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask)))
2344 andres@anarazel.de 6235 [ # # ]:UBC 0 : ereport(ERROR,
6236 : : (errcode(ERRCODE_DATA_CORRUPTED),
6237 : : errmsg_internal("multixact %u from before multi freeze cutoff %u found to be still running",
6238 : : multi, cutoffs->OldestMxact)));
6239 : :
3772 alvherre@alvh.no-ip. 6240 [ - + - - ]:CBC 7 : if (HEAP_XMAX_IS_LOCKED_ONLY(t_infomask))
6241 : : {
6242 : 7 : *flags |= FRM_INVALIDATE_XMAX;
473 pg@bowt.ie 6243 : 7 : pagefrz->freeze_required = true;
6244 : 7 : return InvalidTransactionId;
6245 : : }
6246 : :
6247 : : /* replace multi with single XID for its updater? */
473 pg@bowt.ie 6248 :UBC 0 : update_xact = MultiXactIdGetUpdateXid(multi, t_infomask);
6249 [ # # ]: 0 : if (TransactionIdPrecedes(update_xact, cutoffs->relfrozenxid))
6250 [ # # ]: 0 : ereport(ERROR,
6251 : : (errcode(ERRCODE_DATA_CORRUPTED),
6252 : : errmsg_internal("multixact %u contains update XID %u from before relfrozenxid %u",
6253 : : multi, update_xact,
6254 : : cutoffs->relfrozenxid)));
6255 [ # # ]: 0 : else if (TransactionIdPrecedes(update_xact, cutoffs->OldestXmin))
6256 : : {
6257 : : /*
6258 : : * Updater XID has to have aborted (otherwise the tuple would have
6259 : : * been pruned away instead, since updater XID is < OldestXmin).
6260 : : * Just remove xmax.
6261 : : */
467 6262 [ # # ]: 0 : if (TransactionIdDidCommit(update_xact))
473 6263 [ # # ]: 0 : ereport(ERROR,
6264 : : (errcode(ERRCODE_DATA_CORRUPTED),
6265 : : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6266 : : multi, update_xact,
6267 : : cutoffs->OldestXmin)));
6268 : 0 : *flags |= FRM_INVALIDATE_XMAX;
6269 : 0 : pagefrz->freeze_required = true;
6270 : 0 : return InvalidTransactionId;
6271 : : }
6272 : :
6273 : : /* Have to keep updater XID as new xmax */
6274 : 0 : *flags |= FRM_RETURN_IS_XID;
6275 : 0 : pagefrz->freeze_required = true;
6276 : 0 : return update_xact;
6277 : : }
6278 : :
6279 : : /*
6280 : : * Some member(s) of this Multi may be below FreezeLimit xid cutoff, so we
6281 : : * need to walk the whole members array to figure out what to do, if
6282 : : * anything.
6283 : : */
6284 : : nmembers =
2851 alvherre@alvh.no-ip. 6285 :CBC 2 : GetMultiXactIdMembers(multi, &members, false,
3547 6286 [ - + - - ]: 2 : HEAP_XMAX_IS_LOCKED_ONLY(t_infomask));
3772 6287 [ - + ]: 2 : if (nmembers <= 0)
6288 : : {
6289 : : /* Nothing worth keeping */
3772 alvherre@alvh.no-ip. 6290 :UBC 0 : *flags |= FRM_INVALIDATE_XMAX;
473 pg@bowt.ie 6291 : 0 : pagefrz->freeze_required = true;
3772 alvherre@alvh.no-ip. 6292 : 0 : return InvalidTransactionId;
6293 : : }
6294 : :
6295 : : /*
6296 : : * The FRM_NOOP case is the only case where we might need to ratchet back
6297 : : * FreezePageRelfrozenXid or FreezePageRelminMxid. It is also the only
6298 : : * case where our caller might ratchet back its NoFreezePageRelfrozenXid
6299 : : * or NoFreezePageRelminMxid "no freeze" trackers to deal with a multi.
6300 : : * FRM_NOOP handling should result in the NewRelfrozenXid/NewRelminMxid
6301 : : * trackers managed by VACUUM being ratcheting back by xmax to the degree
6302 : : * required to make it safe to leave xmax undisturbed, independent of
6303 : : * whether or not page freezing is triggered somewhere else.
6304 : : *
6305 : : * Our policy is to force freezing in every case other than FRM_NOOP,
6306 : : * which obviates the need to maintain either set of trackers, anywhere.
6307 : : * Every other case will reliably execute a freeze plan for xmax that
6308 : : * either replaces xmax with an XID/MXID >= OldestXmin/OldestMxact, or
6309 : : * sets xmax to an InvalidTransactionId XID, rendering xmax fully frozen.
6310 : : * (VACUUM's NewRelfrozenXid/NewRelminMxid trackers are initialized with
6311 : : * OldestXmin/OldestMxact, so later values never need to be tracked here.)
6312 : : */
3772 alvherre@alvh.no-ip. 6313 :CBC 2 : need_replace = false;
473 pg@bowt.ie 6314 : 2 : FreezePageRelfrozenXid = pagefrz->FreezePageRelfrozenXid;
479 6315 [ + + ]: 4 : for (int i = 0; i < nmembers; i++)
6316 : : {
6317 : 3 : TransactionId xid = members[i].xid;
6318 : :
6319 [ - + ]: 3 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6320 : :
6321 [ + + ]: 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
6322 : : {
6323 : : /* Can't violate the FreezeLimit postcondition */
3772 alvherre@alvh.no-ip. 6324 : 1 : need_replace = true;
6325 : 1 : break;
6326 : : }
473 pg@bowt.ie 6327 [ - + ]: 2 : if (TransactionIdPrecedes(xid, FreezePageRelfrozenXid))
473 pg@bowt.ie 6328 :UBC 0 : FreezePageRelfrozenXid = xid;
6329 : : }
6330 : :
6331 : : /* Can't violate the MultiXactCutoff postcondition, either */
473 pg@bowt.ie 6332 [ + + ]:CBC 2 : if (!need_replace)
6333 : 1 : need_replace = MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff);
6334 : :
3772 alvherre@alvh.no-ip. 6335 [ + + ]: 2 : if (!need_replace)
6336 : : {
6337 : : /*
6338 : : * vacuumlazy.c might ratchet back NewRelminMxid, NewRelfrozenXid, or
6339 : : * both together to make it safe to retain this particular multi after
6340 : : * freezing its page
6341 : : */
6342 : 1 : *flags |= FRM_NOOP;
473 pg@bowt.ie 6343 : 1 : pagefrz->FreezePageRelfrozenXid = FreezePageRelfrozenXid;
6344 [ - + ]: 1 : if (MultiXactIdPrecedes(multi, pagefrz->FreezePageRelminMxid))
473 pg@bowt.ie 6345 :UBC 0 : pagefrz->FreezePageRelminMxid = multi;
3772 alvherre@alvh.no-ip. 6346 :CBC 1 : pfree(members);
742 pg@bowt.ie 6347 : 1 : return multi;
6348 : : }
6349 : :
6350 : : /*
6351 : : * Do a more thorough second pass over the multi to figure out which
6352 : : * member XIDs actually need to be kept. Checking the precise status of
6353 : : * individual members might even show that we don't need to keep anything.
6354 : : * That is quite possible even though the Multi must be >= OldestMxact,
6355 : : * since our second pass only keeps member XIDs when it's truly necessary;
6356 : : * even member XIDs >= OldestXmin often won't be kept by second pass.
6357 : : */
3772 alvherre@alvh.no-ip. 6358 : 1 : nnewmembers = 0;
6359 : 1 : newmembers = palloc(sizeof(MultiXactMember) * nmembers);
6360 : 1 : has_lockers = false;
6361 : 1 : update_xid = InvalidTransactionId;
6362 : 1 : update_committed = false;
6363 : :
6364 : : /*
6365 : : * Determine whether to keep each member xid, or to ignore it instead
6366 : : */
479 pg@bowt.ie 6367 [ + + ]: 3 : for (int i = 0; i < nmembers; i++)
6368 : : {
6369 : 2 : TransactionId xid = members[i].xid;
6370 : 2 : MultiXactStatus mstatus = members[i].status;
6371 : :
6372 [ - + ]: 2 : Assert(!TransactionIdPrecedes(xid, cutoffs->relfrozenxid));
6373 : :
6374 [ + - ]: 2 : if (!ISUPDATE_from_mxstatus(mstatus))
6375 : : {
6376 : : /*
6377 : : * Locker XID (not updater XID). We only keep lockers that are
6378 : : * still running.
6379 : : */
6380 [ + - + + ]: 4 : if (TransactionIdIsCurrentTransactionId(xid) ||
6381 : 2 : TransactionIdIsInProgress(xid))
6382 : : {
473 6383 [ - + ]: 1 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
473 pg@bowt.ie 6384 [ # # ]:UBC 0 : ereport(ERROR,
6385 : : (errcode(ERRCODE_DATA_CORRUPTED),
6386 : : errmsg_internal("multixact %u contains running locker XID %u from before removable cutoff %u",
6387 : : multi, xid,
6388 : : cutoffs->OldestXmin)));
479 pg@bowt.ie 6389 :CBC 1 : newmembers[nnewmembers++] = members[i];
6390 : 1 : has_lockers = true;
6391 : : }
6392 : :
6393 : 2 : continue;
6394 : : }
6395 : :
6396 : : /*
6397 : : * Updater XID (not locker XID). Should we keep it?
6398 : : *
6399 : : * Since the tuple wasn't totally removed when vacuum pruned, the
6400 : : * update Xid cannot possibly be older than OldestXmin cutoff unless
6401 : : * the updater XID aborted. If the updater transaction is known
6402 : : * aborted or crashed then it's okay to ignore it, otherwise not.
6403 : : *
6404 : : * In any case the Multi should never contain two updaters, whatever
6405 : : * their individual commit status. Check for that first, in passing.
6406 : : */
479 pg@bowt.ie 6407 [ # # ]:UBC 0 : if (TransactionIdIsValid(update_xid))
6408 [ # # ]: 0 : ereport(ERROR,
6409 : : (errcode(ERRCODE_DATA_CORRUPTED),
6410 : : errmsg_internal("multixact %u has two or more updating members",
6411 : : multi),
6412 : : errdetail_internal("First updater XID=%u second updater XID=%u.",
6413 : : update_xid, xid)));
6414 : :
6415 : : /*
6416 : : * As with all tuple visibility routines, it's critical to test
6417 : : * TransactionIdIsInProgress before TransactionIdDidCommit, because of
6418 : : * race conditions explained in detail in heapam_visibility.c.
6419 : : */
6420 [ # # # # ]: 0 : if (TransactionIdIsCurrentTransactionId(xid) ||
6421 : 0 : TransactionIdIsInProgress(xid))
6422 : 0 : update_xid = xid;
6423 [ # # ]: 0 : else if (TransactionIdDidCommit(xid))
6424 : : {
6425 : : /*
6426 : : * The transaction committed, so we can tell caller to set
6427 : : * HEAP_XMAX_COMMITTED. (We can only do this because we know the
6428 : : * transaction is not running.)
6429 : : */
6430 : 0 : update_committed = true;
6431 : 0 : update_xid = xid;
6432 : : }
6433 : : else
6434 : : {
6435 : : /*
6436 : : * Not in progress, not committed -- must be aborted or crashed;
6437 : : * we can ignore it.
6438 : : */
6439 : 0 : continue;
6440 : : }
6441 : :
6442 : : /*
6443 : : * We determined that updater must be kept -- add it to pending new
6444 : : * members list
6445 : : */
473 6446 [ # # ]: 0 : if (TransactionIdPrecedes(xid, cutoffs->OldestXmin))
6447 [ # # ]: 0 : ereport(ERROR,
6448 : : (errcode(ERRCODE_DATA_CORRUPTED),
6449 : : errmsg_internal("multixact %u contains committed update XID %u from before removable cutoff %u",
6450 : : multi, xid, cutoffs->OldestXmin)));
479 6451 : 0 : newmembers[nnewmembers++] = members[i];
6452 : : }
6453 : :
3772 alvherre@alvh.no-ip. 6454 :CBC 1 : pfree(members);
6455 : :
6456 : : /*
6457 : : * Determine what to do with caller's multi based on information gathered
6458 : : * during our second pass
6459 : : */
6460 [ - + ]: 1 : if (nnewmembers == 0)
6461 : : {
6462 : : /* Nothing worth keeping */
3772 alvherre@alvh.no-ip. 6463 :UBC 0 : *flags |= FRM_INVALIDATE_XMAX;
479 pg@bowt.ie 6464 : 0 : newxmax = InvalidTransactionId;
6465 : : }
3772 alvherre@alvh.no-ip. 6466 [ - + - - ]:CBC 1 : else if (TransactionIdIsValid(update_xid) && !has_lockers)
6467 : : {
6468 : : /*
6469 : : * If there's a single member and it's an update, pass it back alone
6470 : : * without creating a new Multi. (XXX we could do this when there's a
6471 : : * single remaining locker, too, but that would complicate the API too
6472 : : * much; moreover, the case with the single updater is more
6473 : : * interesting, because those are longer-lived.)
6474 : : */
3772 alvherre@alvh.no-ip. 6475 [ # # ]:UBC 0 : Assert(nnewmembers == 1);
6476 : 0 : *flags |= FRM_RETURN_IS_XID;
6477 [ # # ]: 0 : if (update_committed)
6478 : 0 : *flags |= FRM_MARK_COMMITTED;
479 pg@bowt.ie 6479 : 0 : newxmax = update_xid;
6480 : : }
6481 : : else
6482 : : {
6483 : : /*
6484 : : * Create a new multixact with the surviving members of the previous
6485 : : * one, to set as new Xmax in the tuple
6486 : : */
479 pg@bowt.ie 6487 :CBC 1 : newxmax = MultiXactIdCreateFromMembers(nnewmembers, newmembers);
3772 alvherre@alvh.no-ip. 6488 : 1 : *flags |= FRM_RETURN_IS_MULTI;
6489 : : }
6490 : :
6491 : 1 : pfree(newmembers);
6492 : :
473 pg@bowt.ie 6493 : 1 : pagefrz->freeze_required = true;
479 6494 : 1 : return newxmax;
6495 : : }
6496 : :
6497 : : /*
6498 : : * heap_prepare_freeze_tuple
6499 : : *
6500 : : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
6501 : : * are older than the OldestXmin and/or OldestMxact freeze cutoffs. If so,
6502 : : * setup enough state (in the *frz output argument) to enable caller to
6503 : : * process this tuple as part of freezing its page, and return true. Return
6504 : : * false if nothing can be changed about the tuple right now.
6505 : : *
6506 : : * Also sets *totally_frozen to true if the tuple will be totally frozen once
6507 : : * caller executes returned freeze plan (or if the tuple was already totally
6508 : : * frozen by an earlier VACUUM). This indicates that there are no remaining
6509 : : * XIDs or MultiXactIds that will need to be processed by a future VACUUM.
6510 : : *
6511 : : * VACUUM caller must assemble HeapTupleFreeze freeze plan entries for every
6512 : : * tuple that we returned true for, and then execute freezing. Caller must
6513 : : * initialize pagefrz fields for page as a whole before first call here for
6514 : : * each heap page.
6515 : : *
6516 : : * VACUUM caller decides on whether or not to freeze the page as a whole.
6517 : : * We'll often prepare freeze plans for a page that caller just discards.
6518 : : * However, VACUUM doesn't always get to make a choice; it must freeze when
6519 : : * pagefrz.freeze_required is set, to ensure that any XIDs < FreezeLimit (and
6520 : : * MXIDs < MultiXactCutoff) can never be left behind. We help to make sure
6521 : : * that VACUUM always follows that rule.
6522 : : *
6523 : : * We sometimes force freezing of xmax MultiXactId values long before it is
6524 : : * strictly necessary to do so just to ensure the FreezeLimit postcondition.
6525 : : * It's worth processing MultiXactIds proactively when it is cheap to do so,
6526 : : * and it's convenient to make that happen by piggy-backing it on the "force
6527 : : * freezing" mechanism. Conversely, we sometimes delay freezing MultiXactIds
6528 : : * because it is expensive right now (though only when it's still possible to
6529 : : * do so without violating the FreezeLimit/MultiXactCutoff postcondition).
6530 : : *
6531 : : * It is assumed that the caller has checked the tuple with
6532 : : * HeapTupleSatisfiesVacuum() and determined that it is not HEAPTUPLE_DEAD
6533 : : * (else we should be removing the tuple, not freezing it).
6534 : : *
6535 : : * NB: This function has side effects: it might allocate a new MultiXactId.
6536 : : * It will be set as tuple's new xmax when our *frz output is processed within
6537 : : * heap_execute_freeze_tuple later on. If the tuple is in a shared buffer
6538 : : * then caller had better have an exclusive lock on it already.
6539 : : */
6540 : : bool
2344 andres@anarazel.de 6541 : 15681373 : heap_prepare_freeze_tuple(HeapTupleHeader tuple,
6542 : : const struct VacuumCutoffs *cutoffs,
6543 : : HeapPageFreeze *pagefrz,
6544 : : HeapTupleFreeze *frz, bool *totally_frozen)
6545 : : {
479 pg@bowt.ie 6546 : 15681373 : bool xmin_already_frozen = false,
6547 : 15681373 : xmax_already_frozen = false;
6548 : 15681373 : bool freeze_xmin = false,
6549 : 15681373 : replace_xvac = false,
6550 : 15681373 : replace_xmax = false,
6551 : 15681373 : freeze_xmax = false;
6552 : : TransactionId xid;
6553 : :
467 6554 : 15681373 : frz->xmax = HeapTupleHeaderGetRawXmax(tuple);
3772 alvherre@alvh.no-ip. 6555 : 15681373 : frz->t_infomask2 = tuple->t_infomask2;
6556 : 15681373 : frz->t_infomask = tuple->t_infomask;
467 pg@bowt.ie 6557 : 15681373 : frz->frzflags = 0;
6558 : 15681373 : frz->checkflags = 0;
6559 : :
6560 : : /*
6561 : : * Process xmin, while keeping track of whether it's already frozen, or
6562 : : * will become frozen iff our freeze plan is executed by caller (could be
6563 : : * neither).
6564 : : */
6370 tgl@sss.pgh.pa.us 6565 [ + + ]: 15681373 : xid = HeapTupleHeaderGetXmin(tuple);
1809 alvherre@alvh.no-ip. 6566 [ + + ]: 15681373 : if (!TransactionIdIsNormal(xid))
479 pg@bowt.ie 6567 : 12902131 : xmin_already_frozen = true;
6568 : : else
6569 : : {
6570 [ - + ]: 2779242 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
2344 andres@anarazel.de 6571 [ # # ]:UBC 0 : ereport(ERROR,
6572 : : (errcode(ERRCODE_DATA_CORRUPTED),
6573 : : errmsg_internal("found xmin %u from before relfrozenxid %u",
6574 : : xid, cutoffs->relfrozenxid)));
6575 : :
6576 : : /* Will set freeze_xmin flags in freeze plan below */
473 pg@bowt.ie 6577 :CBC 2779242 : freeze_xmin = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
6578 : :
6579 : : /* Verify that xmin committed if and when freeze plan is executed */
467 6580 [ + + ]: 2779242 : if (freeze_xmin)
6581 : 2157917 : frz->checkflags |= HEAP_FREEZE_CHECK_XMIN_COMMITTED;
6582 : : }
6583 : :
6584 : : /*
6585 : : * Old-style VACUUM FULL is gone, but we have to process xvac for as long
6586 : : * as we support having MOVED_OFF/MOVED_IN tuples in the database
6587 : : */
479 6588 [ - + ]: 15681373 : xid = HeapTupleHeaderGetXvac(tuple);
6589 [ - + ]: 15681373 : if (TransactionIdIsNormal(xid))
6590 : : {
479 pg@bowt.ie 6591 [ # # ]:UBC 0 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
6592 [ # # ]: 0 : Assert(TransactionIdPrecedes(xid, cutoffs->OldestXmin));
6593 : :
6594 : : /*
6595 : : * For Xvac, we always freeze proactively. This allows totally_frozen
6596 : : * tracking to ignore xvac.
6597 : : */
473 6598 : 0 : replace_xvac = pagefrz->freeze_required = true;
6599 : :
6600 : : /* Will set replace_xvac flags in freeze plan below */
6601 : : }
6602 : :
6603 : : /* Now process xmax */
467 pg@bowt.ie 6604 :CBC 15681373 : xid = frz->xmax;
3790 alvherre@alvh.no-ip. 6605 [ + + ]: 15681373 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
6606 : : {
6607 : : /* Raw xmax is a MultiXactId */
6608 : : TransactionId newxmax;
6609 : : uint16 flags;
6610 : :
6611 : : /*
6612 : : * We will either remove xmax completely (in the "freeze_xmax" path),
6613 : : * process xmax by replacing it (in the "replace_xmax" path), or
6614 : : * perform no-op xmax processing. The only constraint is that the
6615 : : * FreezeLimit/MultiXactCutoff postcondition must never be violated.
6616 : : */
479 pg@bowt.ie 6617 : 9 : newxmax = FreezeMultiXactId(xid, tuple->t_infomask, cutoffs,
6618 : : &flags, pagefrz);
6619 : :
473 6620 [ + + ]: 9 : if (flags & FRM_NOOP)
6621 : : {
6622 : : /*
6623 : : * xmax is a MultiXactId, and nothing about it changes for now.
6624 : : * This is the only case where 'freeze_required' won't have been
6625 : : * set for us by FreezeMultiXactId, as well as the only case where
6626 : : * neither freeze_xmax nor replace_xmax are set (given a multi).
6627 : : *
6628 : : * This is a no-op, but the call to FreezeMultiXactId might have
6629 : : * ratcheted back NewRelfrozenXid and/or NewRelminMxid trackers
6630 : : * for us (the "freeze page" variants, specifically). That'll
6631 : : * make it safe for our caller to freeze the page later on, while
6632 : : * leaving this particular xmax undisturbed.
6633 : : *
6634 : : * FreezeMultiXactId is _not_ responsible for the "no freeze"
6635 : : * NewRelfrozenXid/NewRelminMxid trackers, though -- that's our
6636 : : * job. A call to heap_tuple_should_freeze for this same tuple
6637 : : * will take place below if 'freeze_required' isn't set already.
6638 : : * (This repeats work from FreezeMultiXactId, but allows "no
6639 : : * freeze" tracker maintenance to happen in only one place.)
6640 : : */
6641 [ - + ]: 1 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->MultiXactCutoff));
6642 [ + - - + ]: 1 : Assert(MultiXactIdIsValid(newxmax) && xid == newxmax);
6643 : : }
6644 [ - + ]: 8 : else if (flags & FRM_RETURN_IS_XID)
6645 : : {
6646 : : /*
6647 : : * xmax will become an updater Xid (original MultiXact's updater
6648 : : * member Xid will be carried forward as a simple Xid in Xmax).
6649 : : */
479 pg@bowt.ie 6650 [ # # ]:UBC 0 : Assert(!TransactionIdPrecedes(newxmax, cutoffs->OldestXmin));
6651 : :
6652 : : /*
6653 : : * NB -- some of these transformations are only valid because we
6654 : : * know the return Xid is a tuple updater (i.e. not merely a
6655 : : * locker.) Also note that the only reason we don't explicitly
6656 : : * worry about HEAP_KEYS_UPDATED is because it lives in
6657 : : * t_infomask2 rather than t_infomask.
6658 : : */
3772 alvherre@alvh.no-ip. 6659 : 0 : frz->t_infomask &= ~HEAP_XMAX_BITS;
6660 : 0 : frz->xmax = newxmax;
6661 [ # # ]: 0 : if (flags & FRM_MARK_COMMITTED)
2474 teodor@sigaev.ru 6662 : 0 : frz->t_infomask |= HEAP_XMAX_COMMITTED;
479 pg@bowt.ie 6663 : 0 : replace_xmax = true;
6664 : : }
3772 alvherre@alvh.no-ip. 6665 [ + + ]:CBC 8 : else if (flags & FRM_RETURN_IS_MULTI)
6666 : : {
6667 : : uint16 newbits;
6668 : : uint16 newbits2;
6669 : :
6670 : : /*
6671 : : * xmax is an old MultiXactId that we have to replace with a new
6672 : : * MultiXactId, to carry forward two or more original member XIDs.
6673 : : */
479 pg@bowt.ie 6674 [ - + ]: 1 : Assert(!MultiXactIdPrecedes(newxmax, cutoffs->OldestMxact));
6675 : :
6676 : : /*
6677 : : * We can't use GetMultiXactIdHintBits directly on the new multi
6678 : : * here; that routine initializes the masks to all zeroes, which
6679 : : * would lose other bits we need. Doing it this way ensures all
6680 : : * unrelated bits remain untouched.
6681 : : */
3772 alvherre@alvh.no-ip. 6682 : 1 : frz->t_infomask &= ~HEAP_XMAX_BITS;
6683 : 1 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6684 : 1 : GetMultiXactIdHintBits(newxmax, &newbits, &newbits2);
6685 : 1 : frz->t_infomask |= newbits;
6686 : 1 : frz->t_infomask2 |= newbits2;
6687 : 1 : frz->xmax = newxmax;
479 pg@bowt.ie 6688 : 1 : replace_xmax = true;
6689 : : }
6690 : : else
6691 : : {
6692 : : /*
6693 : : * Freeze plan for tuple "freezes xmax" in the strictest sense:
6694 : : * it'll leave nothing in xmax (neither an Xid nor a MultiXactId).
6695 : : */
6696 [ - + ]: 7 : Assert(flags & FRM_INVALIDATE_XMAX);
742 6697 [ - + ]: 7 : Assert(!TransactionIdIsValid(newxmax));
6698 : :
6699 : : /* Will set freeze_xmax flags in freeze plan below */
479 6700 : 7 : freeze_xmax = true;
6701 : : }
6702 : :
6703 : : /* MultiXactId processing forces freezing (barring FRM_NOOP case) */
473 6704 [ - + - - : 9 : Assert(pagefrz->freeze_required || (!freeze_xmax && !replace_xmax));
- - ]
6705 : : }
2860 rhaas@postgresql.org 6706 [ + + ]: 15681364 : else if (TransactionIdIsNormal(xid))
6707 : : {
6708 : : /* Raw xmax is normal XID */
479 pg@bowt.ie 6709 [ - + ]: 3929888 : if (TransactionIdPrecedes(xid, cutoffs->relfrozenxid))
2344 andres@anarazel.de 6710 [ # # ]:UBC 0 : ereport(ERROR,
6711 : : (errcode(ERRCODE_DATA_CORRUPTED),
6712 : : errmsg_internal("found xmax %u from before relfrozenxid %u",
6713 : : xid, cutoffs->relfrozenxid)));
6714 : :
6715 : : /* Will set freeze_xmax flags in freeze plan below */
467 pg@bowt.ie 6716 :CBC 3929888 : freeze_xmax = TransactionIdPrecedes(xid, cutoffs->OldestXmin);
6717 : :
6718 : : /*
6719 : : * Verify that xmax aborted if and when freeze plan is executed,
6720 : : * provided it's from an update. (A lock-only xmax can be removed
6721 : : * independent of this, since the lock is released at xact end.)
6722 : : */
6723 [ + + + + : 3929888 : if (freeze_xmax && !HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask))
+ - ]
6724 : 1402 : frz->checkflags |= HEAP_FREEZE_CHECK_XMAX_ABORTED;
6725 : : }
508 6726 [ + - ]: 11751476 : else if (!TransactionIdIsValid(xid))
6727 : : {
6728 : : /* Raw xmax is InvalidTransactionId XID */
6729 [ - + ]: 11751476 : Assert((tuple->t_infomask & HEAP_XMAX_IS_MULTI) == 0);
2172 alvherre@alvh.no-ip. 6730 : 11751476 : xmax_already_frozen = true;
6731 : : }
6732 : : else
2172 alvherre@alvh.no-ip. 6733 [ # # ]:UBC 0 : ereport(ERROR,
6734 : : (errcode(ERRCODE_DATA_CORRUPTED),
6735 : : errmsg_internal("found raw xmax %u (infomask 0x%04x) not invalid and not multi",
6736 : : xid, tuple->t_infomask)));
6737 : :
479 pg@bowt.ie 6738 [ + + ]:CBC 15681373 : if (freeze_xmin)
6739 : : {
6740 [ - + ]: 2157917 : Assert(!xmin_already_frozen);
6741 : :
6742 : 2157917 : frz->t_infomask |= HEAP_XMIN_FROZEN;
6743 : : }
6744 [ - + ]: 15681373 : if (replace_xvac)
6745 : : {
6746 : : /*
6747 : : * If a MOVED_OFF tuple is not dead, the xvac transaction must have
6748 : : * failed; whereas a non-dead MOVED_IN tuple must mean the xvac
6749 : : * transaction succeeded.
6750 : : */
473 pg@bowt.ie 6751 [ # # ]:UBC 0 : Assert(pagefrz->freeze_required);
479 6752 [ # # ]: 0 : if (tuple->t_infomask & HEAP_MOVED_OFF)
6753 : 0 : frz->frzflags |= XLH_INVALID_XVAC;
6754 : : else
6755 : 0 : frz->frzflags |= XLH_FREEZE_XVAC;
6756 : : }
479 pg@bowt.ie 6757 [ + + ]:CBC 15681373 : if (replace_xmax)
6758 : : {
6759 [ + - - + ]: 1 : Assert(!xmax_already_frozen && !freeze_xmax);
473 6760 [ - + ]: 1 : Assert(pagefrz->freeze_required);
6761 : :
6762 : : /* Already set replace_xmax flags in freeze plan earlier */
6763 : : }
3790 alvherre@alvh.no-ip. 6764 [ + + ]: 15681373 : if (freeze_xmax)
6765 : : {
479 pg@bowt.ie 6766 [ + - - + ]: 2165 : Assert(!xmax_already_frozen && !replace_xmax);
6767 : :
3772 alvherre@alvh.no-ip. 6768 : 2165 : frz->xmax = InvalidTransactionId;
6769 : :
6770 : : /*
6771 : : * The tuple might be marked either XMAX_INVALID or XMAX_COMMITTED +
6772 : : * LOCKED. Normalize to INVALID just to be sure no one gets confused.
6773 : : * Also get rid of the HEAP_KEYS_UPDATED bit.
6774 : : */
6775 : 2165 : frz->t_infomask &= ~HEAP_XMAX_BITS;
6776 : 2165 : frz->t_infomask |= HEAP_XMAX_INVALID;
6777 : 2165 : frz->t_infomask2 &= ~HEAP_HOT_UPDATED;
6778 : 2165 : frz->t_infomask2 &= ~HEAP_KEYS_UPDATED;
6779 : : }
6780 : :
6781 : : /*
6782 : : * Determine if this tuple is already totally frozen, or will become
6783 : : * totally frozen (provided caller executes freeze plans for the page)
6784 : : */
479 pg@bowt.ie 6785 [ + + + + : 30739256 : *totally_frozen = ((freeze_xmin || xmin_already_frozen) &&
+ + ]
6786 [ + + ]: 15057883 : (freeze_xmax || xmax_already_frozen));
6787 : :
473 6788 [ + + + + : 15681373 : if (!pagefrz->freeze_required && !(xmin_already_frozen &&
+ + ]
6789 : : xmax_already_frozen))
6790 : : {
6791 : : /*
6792 : : * So far no previous tuple from the page made freezing mandatory.
6793 : : * Does this tuple force caller to freeze the entire page?
6794 : : */
6795 : 5483586 : pagefrz->freeze_required =
6796 : 5483586 : heap_tuple_should_freeze(tuple, cutoffs,
6797 : : &pagefrz->NoFreezePageRelfrozenXid,
6798 : : &pagefrz->NoFreezePageRelminMxid);
6799 : : }
6800 : :
6801 : : /* Tell caller if this tuple has a usable freeze plan set in *frz */
479 6802 [ + + + - : 15681373 : return freeze_xmin || replace_xvac || replace_xmax || freeze_xmax;
+ - + + ]
6803 : : }
6804 : :
6805 : : /*
6806 : : * heap_execute_freeze_tuple
6807 : : * Execute the prepared freezing of a tuple with caller's freeze plan.
6808 : : *
6809 : : * Caller is responsible for ensuring that no other backend can access the
6810 : : * storage underlying this tuple, either by holding an exclusive lock on the
6811 : : * buffer containing it (which is what lazy VACUUM does), or by having it be
6812 : : * in private storage (which is what CLUSTER and friends do).
6813 : : */
6814 : : static inline void
516 6815 : 894218 : heap_execute_freeze_tuple(HeapTupleHeader tuple, HeapTupleFreeze *frz)
6816 : : {
3772 alvherre@alvh.no-ip. 6817 : 894218 : HeapTupleHeaderSetXmax(tuple, frz->xmax);
6818 : :
6819 [ - + ]: 894218 : if (frz->frzflags & XLH_FREEZE_XVAC)
3772 alvherre@alvh.no-ip. 6820 [ # # ]:UBC 0 : HeapTupleHeaderSetXvac(tuple, FrozenTransactionId);
6821 : :
3772 alvherre@alvh.no-ip. 6822 [ - + ]:CBC 894218 : if (frz->frzflags & XLH_INVALID_XVAC)
3772 alvherre@alvh.no-ip. 6823 [ # # ]:UBC 0 : HeapTupleHeaderSetXvac(tuple, InvalidTransactionId);
6824 : :
3772 alvherre@alvh.no-ip. 6825 :CBC 894218 : tuple->t_infomask = frz->t_infomask;
6826 : 894218 : tuple->t_infomask2 = frz->t_infomask2;
6827 : 894218 : }
6828 : :
6829 : : /*
6830 : : * Perform xmin/xmax XID status sanity checks before actually executing freeze
6831 : : * plans.
6832 : : *
6833 : : * heap_prepare_freeze_tuple doesn't perform these checks directly because
6834 : : * pg_xact lookups are relatively expensive. They shouldn't be repeated by
6835 : : * successive VACUUMs that each decide against freezing the same page.
6836 : : */
6837 : : void
11 heikki.linnakangas@i 6838 :GNC 14927 : heap_pre_freeze_checks(Buffer buffer,
6839 : : HeapTupleFreeze *tuples, int ntuples)
6840 : : {
516 pg@bowt.ie 6841 :CBC 14927 : Page page = BufferGetPage(buffer);
6842 : :
467 6843 [ + + ]: 637991 : for (int i = 0; i < ntuples; i++)
6844 : : {
6845 : 623064 : HeapTupleFreeze *frz = tuples + i;
6846 : 623064 : ItemId itemid = PageGetItemId(page, frz->offset);
6847 : : HeapTupleHeader htup;
6848 : :
6849 : 623064 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
6850 : :
6851 : : /* Deliberately avoid relying on tuple hint bits here */
6852 [ + + ]: 623064 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMIN_COMMITTED)
6853 : : {
6854 : 623063 : TransactionId xmin = HeapTupleHeaderGetRawXmin(htup);
6855 : :
6856 [ - + ]: 623063 : Assert(!HeapTupleHeaderXminFrozen(htup));
6857 [ - + ]: 623063 : if (unlikely(!TransactionIdDidCommit(xmin)))
467 pg@bowt.ie 6858 [ # # ]:UBC 0 : ereport(ERROR,
6859 : : (errcode(ERRCODE_DATA_CORRUPTED),
6860 : : errmsg_internal("uncommitted xmin %u needs to be frozen",
6861 : : xmin)));
6862 : : }
6863 : :
6864 : : /*
6865 : : * TransactionIdDidAbort won't work reliably in the presence of XIDs
6866 : : * left behind by transactions that were in progress during a crash,
6867 : : * so we can only check that xmax didn't commit
6868 : : */
467 pg@bowt.ie 6869 [ + + ]:CBC 623064 : if (frz->checkflags & HEAP_FREEZE_CHECK_XMAX_ABORTED)
6870 : : {
6871 : 513 : TransactionId xmax = HeapTupleHeaderGetRawXmax(htup);
6872 : :
6873 [ - + ]: 513 : Assert(TransactionIdIsNormal(xmax));
6874 [ - + ]: 513 : if (unlikely(TransactionIdDidCommit(xmax)))
467 pg@bowt.ie 6875 [ # # ]:UBC 0 : ereport(ERROR,
6876 : : (errcode(ERRCODE_DATA_CORRUPTED),
6877 : : errmsg_internal("cannot freeze committed xmax %u",
6878 : : xmax)));
6879 : : }
6880 : : }
11 heikki.linnakangas@i 6881 :GNC 14927 : }
6882 : :
6883 : : /*
6884 : : * Helper which executes freezing of one or more heap tuples on a page on
6885 : : * behalf of caller. Caller passes an array of tuple plans from
6886 : : * heap_prepare_freeze_tuple. Caller must set 'offset' in each plan for us.
6887 : : * Must be called in a critical section that also marks the buffer dirty and,
6888 : : * if needed, emits WAL.
6889 : : */
6890 : : void
6891 : 14927 : heap_freeze_prepared_tuples(Buffer buffer, HeapTupleFreeze *tuples, int ntuples)
6892 : : {
6893 : 14927 : Page page = BufferGetPage(buffer);
6894 : :
516 pg@bowt.ie 6895 [ + + ]:CBC 637991 : for (int i = 0; i < ntuples; i++)
6896 : : {
467 6897 : 623064 : HeapTupleFreeze *frz = tuples + i;
6898 : 623064 : ItemId itemid = PageGetItemId(page, frz->offset);
6899 : : HeapTupleHeader htup;
6900 : :
516 6901 : 623064 : htup = (HeapTupleHeader) PageGetItem(page, itemid);
467 6902 : 623064 : heap_execute_freeze_tuple(htup, frz);
6903 : : }
516 pg@bowt.ie 6904 :GIC 14927 : }
6905 : :
6906 : : /*
6907 : : * heap_freeze_tuple
6908 : : * Freeze tuple in place, without WAL logging.
6909 : : *
6910 : : * Useful for callers like CLUSTER that perform their own WAL logging.
6911 : : */
6912 : : bool
2344 andres@anarazel.de 6913 :CBC 375258 : heap_freeze_tuple(HeapTupleHeader tuple,
6914 : : TransactionId relfrozenxid, TransactionId relminmxid,
6915 : : TransactionId FreezeLimit, TransactionId MultiXactCutoff)
6916 : : {
6917 : : HeapTupleFreeze frz;
6918 : : bool do_freeze;
6919 : : bool totally_frozen;
6920 : : struct VacuumCutoffs cutoffs;
6921 : : HeapPageFreeze pagefrz;
6922 : :
479 pg@bowt.ie 6923 : 375258 : cutoffs.relfrozenxid = relfrozenxid;
6924 : 375258 : cutoffs.relminmxid = relminmxid;
6925 : 375258 : cutoffs.OldestXmin = FreezeLimit;
6926 : 375258 : cutoffs.OldestMxact = MultiXactCutoff;
6927 : 375258 : cutoffs.FreezeLimit = FreezeLimit;
6928 : 375258 : cutoffs.MultiXactCutoff = MultiXactCutoff;
6929 : :
473 6930 : 375258 : pagefrz.freeze_required = true;
6931 : 375258 : pagefrz.FreezePageRelfrozenXid = FreezeLimit;
6932 : 375258 : pagefrz.FreezePageRelminMxid = MultiXactCutoff;
6933 : 375258 : pagefrz.NoFreezePageRelfrozenXid = FreezeLimit;
6934 : 375258 : pagefrz.NoFreezePageRelminMxid = MultiXactCutoff;
6935 : :
479 6936 : 375258 : do_freeze = heap_prepare_freeze_tuple(tuple, &cutoffs,
6937 : : &pagefrz, &frz, &totally_frozen);
6938 : :
6939 : : /*
6940 : : * Note that because this is not a WAL-logged operation, we don't need to
6941 : : * fill in the offset in the freeze record.
6942 : : */
6943 : :
3772 alvherre@alvh.no-ip. 6944 [ + + ]: 375258 : if (do_freeze)
6945 : 251087 : heap_execute_freeze_tuple(tuple, &frz);
6946 : 375258 : return do_freeze;
6947 : : }
6948 : :
6949 : : /*
6950 : : * For a given MultiXactId, return the hint bits that should be set in the
6951 : : * tuple's infomask.
6952 : : *
6953 : : * Normally this should be called for a multixact that was just created, and
6954 : : * so is on our local cache, so the GetMembers call is fast.
6955 : : */
6956 : : static void
4099 6957 : 1178 : GetMultiXactIdHintBits(MultiXactId multi, uint16 *new_infomask,
6958 : : uint16 *new_infomask2)
6959 : : {
6960 : : int nmembers;
6961 : : MultiXactMember *members;
6962 : : int i;
3973 bruce@momjian.us 6963 : 1178 : uint16 bits = HEAP_XMAX_IS_MULTI;
6964 : 1178 : uint16 bits2 = 0;
6965 : 1178 : bool has_update = false;
6966 : 1178 : LockTupleMode strongest = LockTupleKeyShare;
6967 : :
6968 : : /*
6969 : : * We only use this in multis we just created, so they cannot be values
6970 : : * pre-pg_upgrade.
6971 : : */
3547 alvherre@alvh.no-ip. 6972 : 1178 : nmembers = GetMultiXactIdMembers(multi, &members, false, false);
6973 : :
4099 6974 [ + + ]: 3600 : for (i = 0; i < nmembers; i++)
6975 : : {
6976 : : LockTupleMode mode;
6977 : :
6978 : : /*
6979 : : * Remember the strongest lock mode held by any member of the
6980 : : * multixact.
6981 : : */
4091 6982 : 2422 : mode = TUPLOCK_from_mxstatus(members[i].status);
6983 [ + + ]: 2422 : if (mode > strongest)
6984 : 656 : strongest = mode;
6985 : :
6986 : : /* See what other bits we need */
4099 6987 [ + + + + : 2422 : switch (members[i].status)
- ]
6988 : : {
6989 : 2231 : case MultiXactStatusForKeyShare:
6990 : : case MultiXactStatusForShare:
6991 : : case MultiXactStatusForNoKeyUpdate:
6992 : 2231 : break;
6993 : :
6994 : 52 : case MultiXactStatusForUpdate:
6995 : 52 : bits2 |= HEAP_KEYS_UPDATED;
6996 : 52 : break;
6997 : :
6998 : 129 : case MultiXactStatusNoKeyUpdate:
6999 : 129 : has_update = true;
7000 : 129 : break;
7001 : :
7002 : 10 : case MultiXactStatusUpdate:
7003 : 10 : bits2 |= HEAP_KEYS_UPDATED;
7004 : 10 : has_update = true;
7005 : 10 : break;
7006 : : }
7007 : : }
7008 : :
4091 7009 [ + + + + ]: 1178 : if (strongest == LockTupleExclusive ||
7010 : : strongest == LockTupleNoKeyExclusive)
7011 : 216 : bits |= HEAP_XMAX_EXCL_LOCK;
7012 [ + + ]: 962 : else if (strongest == LockTupleShare)
7013 : 437 : bits |= HEAP_XMAX_SHR_LOCK;
7014 [ + - ]: 525 : else if (strongest == LockTupleKeyShare)
7015 : 525 : bits |= HEAP_XMAX_KEYSHR_LOCK;
7016 : :
4099 7017 [ + + ]: 1178 : if (!has_update)
7018 : 1039 : bits |= HEAP_XMAX_LOCK_ONLY;
7019 : :
7020 [ + - ]: 1178 : if (nmembers > 0)
7021 : 1178 : pfree(members);
7022 : :
7023 : 1178 : *new_infomask = bits;
7024 : 1178 : *new_infomask2 = bits2;
7025 : 1178 : }
7026 : :
7027 : : /*
7028 : : * MultiXactIdGetUpdateXid
7029 : : *
7030 : : * Given a multixact Xmax and corresponding infomask, which does not have the
7031 : : * HEAP_XMAX_LOCK_ONLY bit set, obtain and return the Xid of the updating
7032 : : * transaction.
7033 : : *
7034 : : * Caller is expected to check the status of the updating transaction, if
7035 : : * necessary.
7036 : : */
7037 : : static TransactionId
7038 : 553 : MultiXactIdGetUpdateXid(TransactionId xmax, uint16 t_infomask)
7039 : : {
3973 bruce@momjian.us 7040 : 553 : TransactionId update_xact = InvalidTransactionId;
7041 : : MultiXactMember *members;
7042 : : int nmembers;
7043 : :
4099 alvherre@alvh.no-ip. 7044 [ - + ]: 553 : Assert(!(t_infomask & HEAP_XMAX_LOCK_ONLY));
7045 [ - + ]: 553 : Assert(t_infomask & HEAP_XMAX_IS_MULTI);
7046 : :
7047 : : /*
7048 : : * Since we know the LOCK_ONLY bit is not set, this cannot be a multi from
7049 : : * pre-pg_upgrade.
7050 : : */
3547 7051 : 553 : nmembers = GetMultiXactIdMembers(xmax, &members, false, false);
7052 : :
4099 7053 [ + - ]: 553 : if (nmembers > 0)
7054 : : {
7055 : : int i;
7056 : :
7057 [ + + ]: 2084 : for (i = 0; i < nmembers; i++)
7058 : : {
7059 : : /* Ignore lockers */
3789 7060 [ + + ]: 1531 : if (!ISUPDATE_from_mxstatus(members[i].status))
4099 7061 : 978 : continue;
7062 : :
7063 : : /* there can be at most one updater */
7064 [ - + ]: 553 : Assert(update_xact == InvalidTransactionId);
7065 : 553 : update_xact = members[i].xid;
7066 : : #ifndef USE_ASSERT_CHECKING
7067 : :
7068 : : /*
7069 : : * in an assert-enabled build, walk the whole array to ensure
7070 : : * there's no other updater.
7071 : : */
7072 : : break;
7073 : : #endif
7074 : : }
7075 : :
7076 : 553 : pfree(members);
7077 : : }
7078 : :
7079 : 553 : return update_xact;
7080 : : }
7081 : :
7082 : : /*
7083 : : * HeapTupleGetUpdateXid
7084 : : * As above, but use a HeapTupleHeader
7085 : : *
7086 : : * See also HeapTupleHeaderGetUpdateXid, which can be used without previously
7087 : : * checking the hint bits.
7088 : : */
7089 : : TransactionId
7090 : 544 : HeapTupleGetUpdateXid(HeapTupleHeader tuple)
7091 : : {
7092 : 1088 : return MultiXactIdGetUpdateXid(HeapTupleHeaderGetRawXmax(tuple),
7093 : 544 : tuple->t_infomask);
7094 : : }
7095 : :
7096 : : /*
7097 : : * Does the given multixact conflict with the current transaction grabbing a
7098 : : * tuple lock of the given strength?
7099 : : *
7100 : : * The passed infomask pairs up with the given multixact in the tuple header.
7101 : : *
7102 : : * If current_is_member is not NULL, it is set to 'true' if the current
7103 : : * transaction is a member of the given multixact.
7104 : : */
7105 : : static bool
3397 7106 : 94 : DoesMultiXactIdConflict(MultiXactId multi, uint16 infomask,
7107 : : LockTupleMode lockmode, bool *current_is_member)
7108 : : {
7109 : : int nmembers;
7110 : : MultiXactMember *members;
3249 bruce@momjian.us 7111 : 94 : bool result = false;
7112 : 94 : LOCKMODE wanted = tupleLockExtraInfo[lockmode].hwlock;
7113 : :
2851 alvherre@alvh.no-ip. 7114 [ + - + + : 94 : if (HEAP_LOCKED_UPGRADED(infomask))
- + ]
2851 alvherre@alvh.no-ip. 7115 :UBC 0 : return false;
7116 : :
2851 alvherre@alvh.no-ip. 7117 :CBC 94 : nmembers = GetMultiXactIdMembers(multi, &members, false,
3397 7118 [ + + - + ]: 94 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7119 [ + - ]: 94 : if (nmembers >= 0)
7120 : : {
7121 : : int i;
7122 : :
7123 [ + + ]: 295 : for (i = 0; i < nmembers; i++)
7124 : : {
7125 : : TransactionId memxid;
7126 : : LOCKMODE memlockmode;
7127 : :
1762 7128 [ + + + + : 207 : if (result && (current_is_member == NULL || *current_is_member))
+ - ]
7129 : : break;
7130 : :
7131 : 201 : memlockmode = LOCKMODE_from_mxstatus(members[i].status);
7132 : :
7133 : : /* ignore members from current xact (but track their presence) */
1764 7134 : 201 : memxid = members[i].xid;
7135 [ + + ]: 201 : if (TransactionIdIsCurrentTransactionId(memxid))
7136 : : {
1762 7137 [ + + ]: 91 : if (current_is_member != NULL)
7138 : 78 : *current_is_member = true;
7139 : 91 : continue;
7140 : : }
7141 [ + + ]: 110 : else if (result)
7142 : 8 : continue;
7143 : :
7144 : : /* ignore members that don't conflict with the lock we want */
7145 [ + + ]: 102 : if (!DoLockModesConflict(memlockmode, wanted))
1764 7146 : 67 : continue;
7147 : :
3397 7148 [ + + ]: 35 : if (ISUPDATE_from_mxstatus(members[i].status))
7149 : : {
7150 : : /* ignore aborted updaters */
7151 [ + + ]: 17 : if (TransactionIdDidAbort(memxid))
7152 : 1 : continue;
7153 : : }
7154 : : else
7155 : : {
7156 : : /* ignore lockers-only that are no longer in progress */
7157 [ + + ]: 18 : if (!TransactionIdIsInProgress(memxid))
7158 : 5 : continue;
7159 : : }
7160 : :
7161 : : /*
7162 : : * Whatever remains are either live lockers that conflict with our
7163 : : * wanted lock, and updaters that are not aborted. Those conflict
7164 : : * with what we want. Set up to return true, but keep going to
7165 : : * look for the current transaction among the multixact members,
7166 : : * if needed.
7167 : : */
7168 : 29 : result = true;
7169 : : }
7170 : 94 : pfree(members);
7171 : : }
7172 : :
7173 : 94 : return result;
7174 : : }
7175 : :
7176 : : /*
7177 : : * Do_MultiXactIdWait
7178 : : * Actual implementation for the two functions below.
7179 : : *
7180 : : * 'multi', 'status' and 'infomask' indicate what to sleep on (the status is
7181 : : * needed to ensure we only sleep on conflicting members, and the infomask is
7182 : : * used to optimize multixact access in case it's a lock-only multi); 'nowait'
7183 : : * indicates whether to use conditional lock acquisition, to allow callers to
7184 : : * fail if lock is unavailable. 'rel', 'ctid' and 'oper' are used to set up
7185 : : * context information for error messages. 'remaining', if not NULL, receives
7186 : : * the number of members that are still running, including any (non-aborted)
7187 : : * subtransactions of our own transaction.
7188 : : *
7189 : : * We do this by sleeping on each member using XactLockTableWait. Any
7190 : : * members that belong to the current backend are *not* waited for, however;
7191 : : * this would not merely be useless but would lead to Assert failure inside
7192 : : * XactLockTableWait. By the time this returns, it is certain that all
7193 : : * transactions *of other backends* that were members of the MultiXactId
7194 : : * that conflict with the requested status are dead (and no new ones can have
7195 : : * been added, since it is not legal to add members to an existing
7196 : : * MultiXactId).
7197 : : *
7198 : : * But by the time we finish sleeping, someone else may have changed the Xmax
7199 : : * of the containing tuple, so the caller needs to iterate on us somehow.
7200 : : *
7201 : : * Note that in case we return false, the number of remaining members is
7202 : : * not to be trusted.
7203 : : */
7204 : : static bool
4099 7205 : 56 : Do_MultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7206 : : uint16 infomask, bool nowait,
7207 : : Relation rel, ItemPointer ctid, XLTW_Oper oper,
7208 : : int *remaining)
7209 : : {
7210 : 56 : bool result = true;
7211 : : MultiXactMember *members;
7212 : : int nmembers;
7213 : 56 : int remain = 0;
7214 : :
7215 : : /* for pre-pg_upgrade tuples, no need to sleep at all */
2851 7216 [ + - + + : 56 : nmembers = HEAP_LOCKED_UPGRADED(infomask) ? -1 :
+ - ]
7217 : 56 : GetMultiXactIdMembers(multi, &members, false,
7218 [ + + - + ]: 56 : HEAP_XMAX_IS_LOCKED_ONLY(infomask));
7219 : :
4099 7220 [ + - ]: 56 : if (nmembers >= 0)
7221 : : {
7222 : : int i;
7223 : :
7224 [ + + ]: 181 : for (i = 0; i < nmembers; i++)
7225 : : {
7226 : 129 : TransactionId memxid = members[i].xid;
7227 : 129 : MultiXactStatus memstatus = members[i].status;
7228 : :
7229 [ + + ]: 129 : if (TransactionIdIsCurrentTransactionId(memxid))
7230 : : {
7231 : 24 : remain++;
7232 : 24 : continue;
7233 : : }
7234 : :
7235 [ + + ]: 105 : if (!DoLockModesConflict(LOCKMODE_from_mxstatus(memstatus),
7236 : 105 : LOCKMODE_from_mxstatus(status)))
7237 : : {
7238 [ + + + - ]: 20 : if (remaining && TransactionIdIsInProgress(memxid))
7239 : 6 : remain++;
7240 : 20 : continue;
7241 : : }
7242 : :
7243 : : /*
7244 : : * This member conflicts with our multi, so we have to sleep (or
7245 : : * return failure, if asked to avoid waiting.)
7246 : : *
7247 : : * Note that we don't set up an error context callback ourselves,
7248 : : * but instead we pass the info down to XactLockTableWait. This
7249 : : * might seem a bit wasteful because the context is set up and
7250 : : * tore down for each member of the multixact, but in reality it
7251 : : * should be barely noticeable, and it avoids duplicate code.
7252 : : */
7253 [ + + ]: 85 : if (nowait)
7254 : : {
7255 : 4 : result = ConditionalXactLockTableWait(memxid);
7256 [ + - ]: 4 : if (!result)
7257 : 4 : break;
7258 : : }
7259 : : else
3679 7260 : 81 : XactLockTableWait(memxid, rel, ctid, oper);
7261 : : }
7262 : :
4099 7263 : 56 : pfree(members);
7264 : : }
7265 : :
7266 [ + + ]: 56 : if (remaining)
7267 : 8 : *remaining = remain;
7268 : :
7269 : 56 : return result;
7270 : : }
7271 : :
7272 : : /*
7273 : : * MultiXactIdWait
7274 : : * Sleep on a MultiXactId.
7275 : : *
7276 : : * By the time we finish sleeping, someone else may have changed the Xmax
7277 : : * of the containing tuple, so the caller needs to iterate on us somehow.
7278 : : *
7279 : : * We return (in *remaining, if not NULL) the number of members that are still
7280 : : * running, including any (non-aborted) subtransactions of our own transaction.
7281 : : */
7282 : : static void
3679 7283 : 52 : MultiXactIdWait(MultiXactId multi, MultiXactStatus status, uint16 infomask,
7284 : : Relation rel, ItemPointer ctid, XLTW_Oper oper,
7285 : : int *remaining)
7286 : : {
7287 : 52 : (void) Do_MultiXactIdWait(multi, status, infomask, false,
7288 : : rel, ctid, oper, remaining);
4099 7289 : 52 : }
7290 : :
7291 : : /*
7292 : : * ConditionalMultiXactIdWait
7293 : : * As above, but only lock if we can get the lock without blocking.
7294 : : *
7295 : : * By the time we finish sleeping, someone else may have changed the Xmax
7296 : : * of the containing tuple, so the caller needs to iterate on us somehow.
7297 : : *
7298 : : * If the multixact is now all gone, return true. Returns false if some
7299 : : * transactions might still be running.
7300 : : *
7301 : : * We return (in *remaining, if not NULL) the number of members that are still
7302 : : * running, including any (non-aborted) subtransactions of our own transaction.
7303 : : */
7304 : : static bool
7305 : 4 : ConditionalMultiXactIdWait(MultiXactId multi, MultiXactStatus status,
7306 : : uint16 infomask, Relation rel, int *remaining)
7307 : : {
3679 7308 : 4 : return Do_MultiXactIdWait(multi, status, infomask, true,
7309 : : rel, NULL, XLTW_None, remaining);
7310 : : }
7311 : :
7312 : : /*
7313 : : * heap_tuple_needs_eventual_freeze
7314 : : *
7315 : : * Check to see whether any of the XID fields of a tuple (xmin, xmax, xvac)
7316 : : * will eventually require freezing (if tuple isn't removed by pruning first).
7317 : : */
7318 : : bool
2966 rhaas@postgresql.org 7319 : 8329478 : heap_tuple_needs_eventual_freeze(HeapTupleHeader tuple)
7320 : : {
7321 : : TransactionId xid;
7322 : :
7323 : : /*
7324 : : * If xmin is a normal transaction ID, this tuple is definitely not
7325 : : * frozen.
7326 : : */
7327 [ + + ]: 8329478 : xid = HeapTupleHeaderGetXmin(tuple);
7328 [ + + ]: 8329478 : if (TransactionIdIsNormal(xid))
7329 : 13340 : return true;
7330 : :
7331 : : /*
7332 : : * If xmax is a valid xact or multixact, this tuple is also not frozen.
7333 : : */
7334 [ + + ]: 8316138 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
7335 : : {
7336 : : MultiXactId multi;
7337 : :
7338 : 2 : multi = HeapTupleHeaderGetRawXmax(tuple);
7339 [ + - ]: 2 : if (MultiXactIdIsValid(multi))
7340 : 2 : return true;
7341 : : }
7342 : : else
7343 : : {
7344 : 8316136 : xid = HeapTupleHeaderGetRawXmax(tuple);
7345 [ + + ]: 8316136 : if (TransactionIdIsNormal(xid))
7346 : 8 : return true;
7347 : : }
7348 : :
7349 [ - + ]: 8316128 : if (tuple->t_infomask & HEAP_MOVED)
7350 : : {
2966 rhaas@postgresql.org 7351 [ # # ]:UBC 0 : xid = HeapTupleHeaderGetXvac(tuple);
7352 [ # # ]: 0 : if (TransactionIdIsNormal(xid))
7353 : 0 : return true;
7354 : : }
7355 : :
2966 rhaas@postgresql.org 7356 :CBC 8316128 : return false;
7357 : : }
7358 : :
7359 : : /*
7360 : : * heap_tuple_should_freeze
7361 : : *
7362 : : * Return value indicates if heap_prepare_freeze_tuple sibling function would
7363 : : * (or should) force freezing of the heap page that contains caller's tuple.
7364 : : * Tuple header XIDs/MXIDs < FreezeLimit/MultiXactCutoff trigger freezing.
7365 : : * This includes (xmin, xmax, xvac) fields, as well as MultiXact member XIDs.
7366 : : *
7367 : : * The *NoFreezePageRelfrozenXid and *NoFreezePageRelminMxid input/output
7368 : : * arguments help VACUUM track the oldest extant XID/MXID remaining in rel.
7369 : : * Our working assumption is that caller won't decide to freeze this tuple.
7370 : : * It's up to caller to only ratchet back its own top-level trackers after the
7371 : : * point that it fully commits to not freezing the tuple/page in question.
7372 : : */
7373 : : bool
473 pg@bowt.ie 7374 : 5484392 : heap_tuple_should_freeze(HeapTupleHeader tuple,
7375 : : const struct VacuumCutoffs *cutoffs,
7376 : : TransactionId *NoFreezePageRelfrozenXid,
7377 : : MultiXactId *NoFreezePageRelminMxid)
7378 : : {
7379 : : TransactionId xid;
7380 : : MultiXactId multi;
479 7381 : 5484392 : bool freeze = false;
7382 : :
7383 : : /* First deal with xmin */
4542 rhaas@postgresql.org 7384 [ + + ]: 5484392 : xid = HeapTupleHeaderGetXmin(tuple);
742 pg@bowt.ie 7385 [ + + ]: 5484392 : if (TransactionIdIsNormal(xid))
7386 : : {
479 7387 [ - + ]: 1892490 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
473 7388 [ + + ]: 1892490 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7389 : 15257 : *NoFreezePageRelfrozenXid = xid;
479 7390 [ + + ]: 1892490 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7391 : 13788 : freeze = true;
7392 : : }
7393 : :
7394 : : /* Now deal with xmax */
742 7395 : 5484392 : xid = InvalidTransactionId;
7396 : 5484392 : multi = InvalidMultiXactId;
7397 [ + + ]: 5484392 : if (tuple->t_infomask & HEAP_XMAX_IS_MULTI)
3790 alvherre@alvh.no-ip. 7398 : 2 : multi = HeapTupleHeaderGetRawXmax(tuple);
7399 : : else
742 pg@bowt.ie 7400 : 5484390 : xid = HeapTupleHeaderGetRawXmax(tuple);
7401 : :
7402 [ + + ]: 5484392 : if (TransactionIdIsNormal(xid))
7403 : : {
479 7404 [ - + ]: 3841035 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
7405 : : /* xmax is a non-permanent XID */
473 7406 [ + + ]: 3841035 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7407 : 3 : *NoFreezePageRelfrozenXid = xid;
479 7408 [ + + ]: 3841035 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
7409 : 4 : freeze = true;
7410 : : }
742 7411 [ + + ]: 1643357 : else if (!MultiXactIdIsValid(multi))
7412 : : {
7413 : : /* xmax is a permanent XID or invalid MultiXactId/XID */
7414 : : }
7415 [ + - + + : 2 : else if (HEAP_LOCKED_UPGRADED(tuple->t_infomask))
- + ]
7416 : : {
7417 : : /* xmax is a pg_upgrade'd MultiXact, which can't have updater XID */
473 pg@bowt.ie 7418 [ # # ]:UBC 0 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7419 : 0 : *NoFreezePageRelminMxid = multi;
7420 : : /* heap_prepare_freeze_tuple always freezes pg_upgrade'd xmax */
479 7421 : 0 : freeze = true;
7422 : : }
7423 : : else
7424 : : {
7425 : : /* xmax is a MultiXactId that may have an updater XID */
7426 : : MultiXactMember *members;
7427 : : int nmembers;
7428 : :
479 pg@bowt.ie 7429 [ - + ]:CBC 2 : Assert(MultiXactIdPrecedesOrEquals(cutoffs->relminmxid, multi));
473 7430 [ + - ]: 2 : if (MultiXactIdPrecedes(multi, *NoFreezePageRelminMxid))
7431 : 2 : *NoFreezePageRelminMxid = multi;
479 7432 [ + - ]: 2 : if (MultiXactIdPrecedes(multi, cutoffs->MultiXactCutoff))
7433 : 2 : freeze = true;
7434 : :
7435 : : /* need to check whether any member of the mxact is old */
742 7436 : 2 : nmembers = GetMultiXactIdMembers(multi, &members, false,
7437 [ + + - + ]: 2 : HEAP_XMAX_IS_LOCKED_ONLY(tuple->t_infomask));
7438 : :
7439 [ + + ]: 5 : for (int i = 0; i < nmembers; i++)
7440 : : {
7441 : 3 : xid = members[i].xid;
479 7442 [ - + ]: 3 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
473 7443 [ - + ]: 3 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
473 pg@bowt.ie 7444 :UBC 0 : *NoFreezePageRelfrozenXid = xid;
479 pg@bowt.ie 7445 [ - + ]:CBC 3 : if (TransactionIdPrecedes(xid, cutoffs->FreezeLimit))
479 pg@bowt.ie 7446 :UBC 0 : freeze = true;
7447 : : }
742 pg@bowt.ie 7448 [ + + ]:CBC 2 : if (nmembers > 0)
7449 : 1 : pfree(members);
7450 : : }
7451 : :
4542 rhaas@postgresql.org 7452 [ - + ]: 5484392 : if (tuple->t_infomask & HEAP_MOVED)
7453 : : {
4542 rhaas@postgresql.org 7454 [ # # ]:UBC 0 : xid = HeapTupleHeaderGetXvac(tuple);
742 pg@bowt.ie 7455 [ # # ]: 0 : if (TransactionIdIsNormal(xid))
7456 : : {
479 7457 [ # # ]: 0 : Assert(TransactionIdPrecedesOrEquals(cutoffs->relfrozenxid, xid));
473 7458 [ # # ]: 0 : if (TransactionIdPrecedes(xid, *NoFreezePageRelfrozenXid))
7459 : 0 : *NoFreezePageRelfrozenXid = xid;
7460 : : /* heap_prepare_freeze_tuple forces xvac freezing */
479 7461 : 0 : freeze = true;
7462 : : }
7463 : : }
7464 : :
479 pg@bowt.ie 7465 :CBC 5484392 : return freeze;
7466 : : }
7467 : :
7468 : : /*
7469 : : * Maintain snapshotConflictHorizon for caller by ratcheting forward its value
7470 : : * using any committed XIDs contained in 'tuple', an obsolescent heap tuple
7471 : : * that caller is in the process of physically removing, e.g. via HOT pruning
7472 : : * or index deletion.
7473 : : *
7474 : : * Caller must initialize its value to InvalidTransactionId, which is
7475 : : * generally interpreted as "definitely no need for a recovery conflict".
7476 : : * Final value must reflect all heap tuples that caller will physically remove
7477 : : * (or remove TID references to) via its ongoing pruning/deletion operation.
7478 : : * ResolveRecoveryConflictWithSnapshot() is passed the final value (taken from
7479 : : * caller's WAL record) by REDO routine when it replays caller's operation.
7480 : : */
7481 : : void
514 7482 : 1515750 : HeapTupleHeaderAdvanceConflictHorizon(HeapTupleHeader tuple,
7483 : : TransactionId *snapshotConflictHorizon)
7484 : : {
5230 simon@2ndQuadrant.co 7485 [ + + ]: 1515750 : TransactionId xmin = HeapTupleHeaderGetXmin(tuple);
4099 alvherre@alvh.no-ip. 7486 [ + + + + : 1515750 : TransactionId xmax = HeapTupleHeaderGetUpdateXid(tuple);
+ - ]
5230 simon@2ndQuadrant.co 7487 [ - + ]: 1515750 : TransactionId xvac = HeapTupleHeaderGetXvac(tuple);
7488 : :
5179 tgl@sss.pgh.pa.us 7489 [ - + ]: 1515750 : if (tuple->t_infomask & HEAP_MOVED)
7490 : : {
514 pg@bowt.ie 7491 [ # # ]:UBC 0 : if (TransactionIdPrecedes(*snapshotConflictHorizon, xvac))
7492 : 0 : *snapshotConflictHorizon = xvac;
7493 : : }
7494 : :
7495 : : /*
7496 : : * Ignore tuples inserted by an aborted transaction or if the tuple was
7497 : : * updated/deleted by the inserting transaction.
7498 : : *
7499 : : * Look for a committed hint bit, or if no xmin bit is set, check clog.
7500 : : */
3766 rhaas@postgresql.org 7501 [ + + ]:CBC 1515750 : if (HeapTupleHeaderXminCommitted(tuple) ||
7502 [ + + + - ]: 89351 : (!HeapTupleHeaderXminInvalid(tuple) && TransactionIdDidCommit(xmin)))
7503 : : {
4875 simon@2ndQuadrant.co 7504 [ + + + + ]: 2730052 : if (xmax != xmin &&
514 pg@bowt.ie 7505 : 1283827 : TransactionIdFollows(xmax, *snapshotConflictHorizon))
7506 : 86824 : *snapshotConflictHorizon = xmax;
7507 : : }
5230 simon@2ndQuadrant.co 7508 : 1515750 : }
7509 : :
7510 : : #ifdef USE_PREFETCH
7511 : : /*
7512 : : * Helper function for heap_index_delete_tuples. Issues prefetch requests for
7513 : : * prefetch_count buffers. The prefetch_state keeps track of all the buffers
7514 : : * we can prefetch, and which have already been prefetched; each call to this
7515 : : * function picks up where the previous call left off.
7516 : : *
7517 : : * Note: we expect the deltids array to be sorted in an order that groups TIDs
7518 : : * by heap block, with all TIDs for each block appearing together in exactly
7519 : : * one group.
7520 : : */
7521 : : static void
1187 pg@bowt.ie 7522 : 16517 : index_delete_prefetch_buffer(Relation rel,
7523 : : IndexDeletePrefetchState *prefetch_state,
7524 : : int prefetch_count)
7525 : : {
1846 andres@anarazel.de 7526 : 16517 : BlockNumber cur_hblkno = prefetch_state->cur_hblkno;
7527 : 16517 : int count = 0;
7528 : : int i;
1187 pg@bowt.ie 7529 : 16517 : int ndeltids = prefetch_state->ndeltids;
7530 : 16517 : TM_IndexDelete *deltids = prefetch_state->deltids;
7531 : :
1846 andres@anarazel.de 7532 : 16517 : for (i = prefetch_state->next_item;
1187 pg@bowt.ie 7533 [ + + + + ]: 598265 : i < ndeltids && count < prefetch_count;
1846 andres@anarazel.de 7534 : 581748 : i++)
7535 : : {
1187 pg@bowt.ie 7536 : 581748 : ItemPointer htid = &deltids[i].tid;
7537 : :
1846 andres@anarazel.de 7538 [ + + + + ]: 1158511 : if (cur_hblkno == InvalidBlockNumber ||
7539 : 576763 : ItemPointerGetBlockNumber(htid) != cur_hblkno)
7540 : : {
7541 : 14973 : cur_hblkno = ItemPointerGetBlockNumber(htid);
7542 : 14973 : PrefetchBuffer(rel, MAIN_FORKNUM, cur_hblkno);
7543 : 14973 : count++;
7544 : : }
7545 : : }
7546 : :
7547 : : /*
7548 : : * Save the prefetch position so that next time we can continue from that
7549 : : * position.
7550 : : */
7551 : 16517 : prefetch_state->next_item = i;
7552 : 16517 : prefetch_state->cur_hblkno = cur_hblkno;
7553 : 16517 : }
7554 : : #endif
7555 : :
7556 : : /*
7557 : : * Helper function for heap_index_delete_tuples. Checks for index corruption
7558 : : * involving an invalid TID in index AM caller's index page.
7559 : : *
7560 : : * This is an ideal place for these checks. The index AM must hold a buffer
7561 : : * lock on the index page containing the TIDs we examine here, so we don't
7562 : : * have to worry about concurrent VACUUMs at all. We can be sure that the
7563 : : * index is corrupt when htid points directly to an LP_UNUSED item or
7564 : : * heap-only tuple, which is not the case during standard index scans.
7565 : : */
7566 : : static inline void
892 pg@bowt.ie 7567 : 485124 : index_delete_check_htid(TM_IndexDeleteOp *delstate,
7568 : : Page page, OffsetNumber maxoff,
7569 : : ItemPointer htid, TM_IndexStatus *istatus)
7570 : : {
7571 : 485124 : OffsetNumber indexpagehoffnum = ItemPointerGetOffsetNumber(htid);
7572 : : ItemId iid;
7573 : :
7574 [ + - + - : 485124 : Assert(OffsetNumberIsValid(istatus->idxoffnum));
- + ]
7575 : :
7576 [ - + ]: 485124 : if (unlikely(indexpagehoffnum > maxoff))
892 pg@bowt.ie 7577 [ # # ]:UBC 0 : ereport(ERROR,
7578 : : (errcode(ERRCODE_INDEX_CORRUPTED),
7579 : : 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\"",
7580 : : ItemPointerGetBlockNumber(htid),
7581 : : indexpagehoffnum,
7582 : : istatus->idxoffnum, delstate->iblknum,
7583 : : RelationGetRelationName(delstate->irel))));
7584 : :
892 pg@bowt.ie 7585 :CBC 485124 : iid = PageGetItemId(page, indexpagehoffnum);
7586 [ - + ]: 485124 : if (unlikely(!ItemIdIsUsed(iid)))
892 pg@bowt.ie 7587 [ # # ]:UBC 0 : ereport(ERROR,
7588 : : (errcode(ERRCODE_INDEX_CORRUPTED),
7589 : : errmsg_internal("heap tid from index tuple (%u,%u) points to unused heap page item at offset %u of block %u in index \"%s\"",
7590 : : ItemPointerGetBlockNumber(htid),
7591 : : indexpagehoffnum,
7592 : : istatus->idxoffnum, delstate->iblknum,
7593 : : RelationGetRelationName(delstate->irel))));
7594 : :
892 pg@bowt.ie 7595 [ + + ]:CBC 485124 : if (ItemIdHasStorage(iid))
7596 : : {
7597 : : HeapTupleHeader htup;
7598 : :
7599 [ - + ]: 282856 : Assert(ItemIdIsNormal(iid));
7600 : 282856 : htup = (HeapTupleHeader) PageGetItem(page, iid);
7601 : :
7602 [ - + ]: 282856 : if (unlikely(HeapTupleHeaderIsHeapOnly(htup)))
892 pg@bowt.ie 7603 [ # # ]:UBC 0 : ereport(ERROR,
7604 : : (errcode(ERRCODE_INDEX_CORRUPTED),
7605 : : errmsg_internal("heap tid from index tuple (%u,%u) points to heap-only tuple at offset %u of block %u in index \"%s\"",
7606 : : ItemPointerGetBlockNumber(htid),
7607 : : indexpagehoffnum,
7608 : : istatus->idxoffnum, delstate->iblknum,
7609 : : RelationGetRelationName(delstate->irel))));
7610 : : }
892 pg@bowt.ie 7611 :CBC 485124 : }
7612 : :
7613 : : /*
7614 : : * heapam implementation of tableam's index_delete_tuples interface.
7615 : : *
7616 : : * This helper function is called by index AMs during index tuple deletion.
7617 : : * See tableam header comments for an explanation of the interface implemented
7618 : : * here and a general theory of operation. Note that each call here is either
7619 : : * a simple index deletion call, or a bottom-up index deletion call.
7620 : : *
7621 : : * It's possible for this to generate a fair amount of I/O, since we may be
7622 : : * deleting hundreds of tuples from a single index block. To amortize that
7623 : : * cost to some degree, this uses prefetching and combines repeat accesses to
7624 : : * the same heap block.
7625 : : */
7626 : : TransactionId
1187 7627 : 4985 : heap_index_delete_tuples(Relation rel, TM_IndexDeleteOp *delstate)
7628 : : {
7629 : : /* Initial assumption is that earlier pruning took care of conflict */
514 7630 : 4985 : TransactionId snapshotConflictHorizon = InvalidTransactionId;
1201 7631 : 4985 : BlockNumber blkno = InvalidBlockNumber;
1846 andres@anarazel.de 7632 : 4985 : Buffer buf = InvalidBuffer;
1201 pg@bowt.ie 7633 : 4985 : Page page = NULL;
7634 : 4985 : OffsetNumber maxoff = InvalidOffsetNumber;
7635 : : TransactionId priorXmax;
7636 : : #ifdef USE_PREFETCH
7637 : : IndexDeletePrefetchState prefetch_state;
7638 : : int prefetch_distance;
7639 : : #endif
7640 : : SnapshotData SnapshotNonVacuumable;
1187 7641 : 4985 : int finalndeltids = 0,
7642 : 4985 : nblocksaccessed = 0;
7643 : :
7644 : : /* State that's only used in bottom-up index deletion case */
7645 : 4985 : int nblocksfavorable = 0;
7646 : 4985 : int curtargetfreespace = delstate->bottomupfreespace,
7647 : 4985 : lastfreespace = 0,
7648 : 4985 : actualfreespace = 0;
7649 : 4985 : bool bottomup_final_block = false;
7650 : :
7651 : 4985 : InitNonVacuumableSnapshot(SnapshotNonVacuumable, GlobalVisTestFor(rel));
7652 : :
7653 : : /* Sort caller's deltids array by TID for further processing */
7654 : 4985 : index_delete_sort(delstate);
7655 : :
7656 : : /*
7657 : : * Bottom-up case: resort deltids array in an order attuned to where the
7658 : : * greatest number of promising TIDs are to be found, and determine how
7659 : : * many blocks from the start of sorted array should be considered
7660 : : * favorable. This will also shrink the deltids array in order to
7661 : : * eliminate completely unfavorable blocks up front.
7662 : : */
7663 [ + + ]: 4985 : if (delstate->bottomup)
7664 : 1729 : nblocksfavorable = bottomup_sort_and_shrink(delstate);
7665 : :
7666 : : #ifdef USE_PREFETCH
7667 : : /* Initialize prefetch state. */
1846 andres@anarazel.de 7668 : 4985 : prefetch_state.cur_hblkno = InvalidBlockNumber;
7669 : 4985 : prefetch_state.next_item = 0;
1187 pg@bowt.ie 7670 : 4985 : prefetch_state.ndeltids = delstate->ndeltids;
7671 : 4985 : prefetch_state.deltids = delstate->deltids;
7672 : :
7673 : : /*
7674 : : * Determine the prefetch distance that we will attempt to maintain.
7675 : : *
7676 : : * Since the caller holds a buffer lock somewhere in rel, we'd better make
7677 : : * sure that isn't a catalog relation before we call code that does
7678 : : * syscache lookups, to avoid risk of deadlock.
7679 : : */
1839 tmunro@postgresql.or 7680 [ + + ]: 4985 : if (IsCatalogRelation(rel))
1490 7681 : 3346 : prefetch_distance = maintenance_io_concurrency;
7682 : : else
7683 : : prefetch_distance =
7684 : 1639 : get_tablespace_maintenance_io_concurrency(rel->rd_rel->reltablespace);
7685 : :
7686 : : /* Cap initial prefetch distance for bottom-up deletion caller */
1187 pg@bowt.ie 7687 [ + + ]: 4985 : if (delstate->bottomup)
7688 : : {
7689 [ - + ]: 1729 : Assert(nblocksfavorable >= 1);
7690 [ - + ]: 1729 : Assert(nblocksfavorable <= BOTTOMUP_MAX_NBLOCKS);
7691 : 1729 : prefetch_distance = Min(prefetch_distance, nblocksfavorable);
7692 : : }
7693 : :
7694 : : /* Start prefetching. */
7695 : 4985 : index_delete_prefetch_buffer(rel, &prefetch_state, prefetch_distance);
7696 : : #endif
7697 : :
7698 : : /* Iterate over deltids, determine which to delete, check their horizon */
7699 [ - + ]: 4985 : Assert(delstate->ndeltids > 0);
7700 [ + + ]: 490109 : for (int i = 0; i < delstate->ndeltids; i++)
7701 : : {
7702 : 486853 : TM_IndexDelete *ideltid = &delstate->deltids[i];
7703 : 486853 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
7704 : 486853 : ItemPointer htid = &ideltid->tid;
7705 : : OffsetNumber offnum;
7706 : :
7707 : : /*
7708 : : * Read buffer, and perform required extra steps each time a new block
7709 : : * is encountered. Avoid refetching if it's the same block as the one
7710 : : * from the last htid.
7711 : : */
1201 7712 [ + + + + ]: 968721 : if (blkno == InvalidBlockNumber ||
7713 : 481868 : ItemPointerGetBlockNumber(htid) != blkno)
7714 : : {
7715 : : /*
7716 : : * Consider giving up early for bottom-up index deletion caller
7717 : : * first. (Only prefetch next-next block afterwards, when it
7718 : : * becomes clear that we're at least going to access the next
7719 : : * block in line.)
7720 : : *
7721 : : * Sometimes the first block frees so much space for bottom-up
7722 : : * caller that the deletion process can end without accessing any
7723 : : * more blocks. It is usually necessary to access 2 or 3 blocks
7724 : : * per bottom-up deletion operation, though.
7725 : : */
1187 7726 [ + + ]: 13261 : if (delstate->bottomup)
7727 : : {
7728 : : /*
7729 : : * We often allow caller to delete a few additional items
7730 : : * whose entries we reached after the point that space target
7731 : : * from caller was satisfied. The cost of accessing the page
7732 : : * was already paid at that point, so it made sense to finish
7733 : : * it off. When that happened, we finalize everything here
7734 : : * (by finishing off the whole bottom-up deletion operation
7735 : : * without needlessly paying the cost of accessing any more
7736 : : * blocks).
7737 : : */
7738 [ + + ]: 3705 : if (bottomup_final_block)
7739 : 147 : break;
7740 : :
7741 : : /*
7742 : : * Give up when we didn't enable our caller to free any
7743 : : * additional space as a result of processing the page that we
7744 : : * just finished up with. This rule is the main way in which
7745 : : * we keep the cost of bottom-up deletion under control.
7746 : : */
7747 [ + + + + ]: 3558 : if (nblocksaccessed >= 1 && actualfreespace == lastfreespace)
7748 : 1582 : break;
7749 : 1976 : lastfreespace = actualfreespace; /* for next time */
7750 : :
7751 : : /*
7752 : : * Deletion operation (which is bottom-up) will definitely
7753 : : * access the next block in line. Prepare for that now.
7754 : : *
7755 : : * Decay target free space so that we don't hang on for too
7756 : : * long with a marginal case. (Space target is only truly
7757 : : * helpful when it allows us to recognize that we don't need
7758 : : * to access more than 1 or 2 blocks to satisfy caller due to
7759 : : * agreeable workload characteristics.)
7760 : : *
7761 : : * We are a bit more patient when we encounter contiguous
7762 : : * blocks, though: these are treated as favorable blocks. The
7763 : : * decay process is only applied when the next block in line
7764 : : * is not a favorable/contiguous block. This is not an
7765 : : * exception to the general rule; we still insist on finding
7766 : : * at least one deletable item per block accessed. See
7767 : : * bottomup_nblocksfavorable() for full details of the theory
7768 : : * behind favorable blocks and heap block locality in general.
7769 : : *
7770 : : * Note: The first block in line is always treated as a
7771 : : * favorable block, so the earliest possible point that the
7772 : : * decay can be applied is just before we access the second
7773 : : * block in line. The Assert() verifies this for us.
7774 : : */
7775 [ + + - + ]: 1976 : Assert(nblocksaccessed > 0 || nblocksfavorable > 0);
7776 [ + + ]: 1976 : if (nblocksfavorable > 0)
7777 : 1874 : nblocksfavorable--;
7778 : : else
7779 : 102 : curtargetfreespace /= 2;
7780 : : }
7781 : :
7782 : : /* release old buffer */
7783 [ + + ]: 11532 : if (BufferIsValid(buf))
7784 : 6547 : UnlockReleaseBuffer(buf);
7785 : :
7786 : 11532 : blkno = ItemPointerGetBlockNumber(htid);
1201 7787 : 11532 : buf = ReadBuffer(rel, blkno);
1187 7788 : 11532 : nblocksaccessed++;
7789 [ + + - + ]: 11532 : Assert(!delstate->bottomup ||
7790 : : nblocksaccessed <= BOTTOMUP_MAX_NBLOCKS);
7791 : :
7792 : : #ifdef USE_PREFETCH
7793 : :
7794 : : /*
7795 : : * To maintain the prefetch distance, prefetch one more page for
7796 : : * each page we read.
7797 : : */
7798 : 11532 : index_delete_prefetch_buffer(rel, &prefetch_state, 1);
7799 : : #endif
7800 : :
1201 7801 : 11532 : LockBuffer(buf, BUFFER_LOCK_SHARE);
7802 : :
7803 : 11532 : page = BufferGetPage(buf);
7804 : 11532 : maxoff = PageGetMaxOffsetNumber(page);
7805 : : }
7806 : :
7807 : : /*
7808 : : * In passing, detect index corruption involving an index page with a
7809 : : * TID that points to a location in the heap that couldn't possibly be
7810 : : * correct. We only do this with actual TIDs from caller's index page
7811 : : * (not items reached by traversing through a HOT chain).
7812 : : */
892 7813 : 485124 : index_delete_check_htid(delstate, page, maxoff, htid, istatus);
7814 : :
1187 7815 [ + + ]: 485124 : if (istatus->knowndeletable)
7816 [ + - - + ]: 126610 : Assert(!delstate->bottomup && !istatus->promising);
7817 : : else
7818 : : {
7819 : 358514 : ItemPointerData tmp = *htid;
7820 : : HeapTupleData heapTuple;
7821 : :
7822 : : /* Are any tuples from this HOT chain non-vacuumable? */
7823 [ + + ]: 358514 : if (heap_hot_search_buffer(&tmp, rel, buf, &SnapshotNonVacuumable,
7824 : : &heapTuple, NULL, true))
7825 : 215371 : continue; /* can't delete entry */
7826 : :
7827 : : /* Caller will delete, since whole HOT chain is vacuumable */
7828 : 143143 : istatus->knowndeletable = true;
7829 : :
7830 : : /* Maintain index free space info for bottom-up deletion case */
7831 [ + + ]: 143143 : if (delstate->bottomup)
7832 : : {
7833 [ - + ]: 7552 : Assert(istatus->freespace > 0);
7834 : 7552 : actualfreespace += istatus->freespace;
7835 [ + + ]: 7552 : if (actualfreespace >= curtargetfreespace)
7836 : 2358 : bottomup_final_block = true;
7837 : : }
7838 : : }
7839 : :
7840 : : /*
7841 : : * Maintain snapshotConflictHorizon value for deletion operation as a
7842 : : * whole by advancing current value using heap tuple headers. This is
7843 : : * loosely based on the logic for pruning a HOT chain.
7844 : : */
1201 7845 : 269753 : offnum = ItemPointerGetOffsetNumber(htid);
7846 : 269753 : priorXmax = InvalidTransactionId; /* cannot check first XMIN */
7847 : : for (;;)
1846 andres@anarazel.de 7848 : 19355 : {
7849 : : ItemId lp;
7850 : : HeapTupleHeader htup;
7851 : :
7852 : : /* Sanity check (pure paranoia) */
935 pg@bowt.ie 7853 [ - + ]: 289108 : if (offnum < FirstOffsetNumber)
935 pg@bowt.ie 7854 :UBC 0 : break;
7855 : :
7856 : : /*
7857 : : * An offset past the end of page's line pointer array is possible
7858 : : * when the array was truncated
7859 : : */
935 pg@bowt.ie 7860 [ - + ]:CBC 289108 : if (offnum > maxoff)
1201 pg@bowt.ie 7861 :UBC 0 : break;
7862 : :
1201 pg@bowt.ie 7863 :CBC 289108 : lp = PageGetItemId(page, offnum);
7864 [ + + ]: 289108 : if (ItemIdIsRedirected(lp))
7865 : : {
7866 : 8741 : offnum = ItemIdGetRedirect(lp);
7867 : 8741 : continue;
7868 : : }
7869 : :
7870 : : /*
7871 : : * We'll often encounter LP_DEAD line pointers (especially with an
7872 : : * entry marked knowndeletable by our caller up front). No heap
7873 : : * tuple headers get examined for an htid that leads us to an
7874 : : * LP_DEAD item. This is okay because the earlier pruning
7875 : : * operation that made the line pointer LP_DEAD in the first place
7876 : : * must have considered the original tuple header as part of
7877 : : * generating its own snapshotConflictHorizon value.
7878 : : *
7879 : : * Relying on XLOG_HEAP2_PRUNE_VACUUM_SCAN records like this is
7880 : : * the same strategy that index vacuuming uses in all cases. Index
7881 : : * VACUUM WAL records don't even have a snapshotConflictHorizon
7882 : : * field of their own for this reason.
7883 : : */
7884 [ + + ]: 280367 : if (!ItemIdIsNormal(lp))
7885 : 181211 : break;
7886 : :
7887 : 99156 : htup = (HeapTupleHeader) PageGetItem(page, lp);
7888 : :
7889 : : /*
7890 : : * Check the tuple XMIN against prior XMAX, if any
7891 : : */
7892 [ + + - + ]: 109770 : if (TransactionIdIsValid(priorXmax) &&
7893 [ + - ]: 10614 : !TransactionIdEquals(HeapTupleHeaderGetXmin(htup), priorXmax))
1201 pg@bowt.ie 7894 :UBC 0 : break;
7895 : :
514 pg@bowt.ie 7896 :CBC 99156 : HeapTupleHeaderAdvanceConflictHorizon(htup,
7897 : : &snapshotConflictHorizon);
7898 : :
7899 : : /*
7900 : : * If the tuple is not HOT-updated, then we are at the end of this
7901 : : * HOT-chain. No need to visit later tuples from the same update
7902 : : * chain (they get their own index entries) -- just move on to
7903 : : * next htid from index AM caller.
7904 : : */
1201 7905 [ + + + - : 99156 : if (!HeapTupleHeaderIsHotUpdated(htup))
+ + ]
7906 : : break;
7907 : :
7908 : : /* Advance to next HOT chain member */
7909 [ - + ]: 10614 : Assert(ItemPointerGetBlockNumber(&htup->t_ctid) == blkno);
7910 : 10614 : offnum = ItemPointerGetOffsetNumber(&htup->t_ctid);
7911 [ + - - + : 10614 : priorXmax = HeapTupleHeaderGetUpdateXid(htup);
- - ]
7912 : : }
7913 : :
7914 : : /* Enable further/final shrinking of deltids for caller */
1187 7915 : 269753 : finalndeltids = i + 1;
7916 : : }
7917 : :
7918 : 4985 : UnlockReleaseBuffer(buf);
7919 : :
7920 : : /*
7921 : : * Shrink deltids array to exclude non-deletable entries at the end. This
7922 : : * is not just a minor optimization. Final deltids array size might be
7923 : : * zero for a bottom-up caller. Index AM is explicitly allowed to rely on
7924 : : * ndeltids being zero in all cases with zero total deletable entries.
7925 : : */
7926 [ + + - + ]: 4985 : Assert(finalndeltids > 0 || delstate->bottomup);
7927 : 4985 : delstate->ndeltids = finalndeltids;
7928 : :
514 7929 : 4985 : return snapshotConflictHorizon;
7930 : : }
7931 : :
7932 : : /*
7933 : : * Specialized inlineable comparison function for index_delete_sort()
7934 : : */
7935 : : static inline int
1187 7936 : 11736116 : index_delete_sort_cmp(TM_IndexDelete *deltid1, TM_IndexDelete *deltid2)
7937 : : {
7938 : 11736116 : ItemPointer tid1 = &deltid1->tid;
7939 : 11736116 : ItemPointer tid2 = &deltid2->tid;
7940 : :
7941 : : {
7942 : 11736116 : BlockNumber blk1 = ItemPointerGetBlockNumber(tid1);
7943 : 11736116 : BlockNumber blk2 = ItemPointerGetBlockNumber(tid2);
7944 : :
7945 [ + + ]: 11736116 : if (blk1 != blk2)
7946 [ + + ]: 4818212 : return (blk1 < blk2) ? -1 : 1;
7947 : : }
7948 : : {
7949 : 6917904 : OffsetNumber pos1 = ItemPointerGetOffsetNumber(tid1);
7950 : 6917904 : OffsetNumber pos2 = ItemPointerGetOffsetNumber(tid2);
7951 : :
7952 [ + - ]: 6917904 : if (pos1 != pos2)
7953 [ + + ]: 6917904 : return (pos1 < pos2) ? -1 : 1;
7954 : : }
7955 : :
898 pg@bowt.ie 7956 :UBC 0 : Assert(false);
7957 : :
7958 : : return 0;
7959 : : }
7960 : :
7961 : : /*
7962 : : * Sort deltids array from delstate by TID. This prepares it for further
7963 : : * processing by heap_index_delete_tuples().
7964 : : *
7965 : : * This operation becomes a noticeable consumer of CPU cycles with some
7966 : : * workloads, so we go to the trouble of specialization/micro optimization.
7967 : : * We use shellsort for this because it's easy to specialize, compiles to
7968 : : * relatively few instructions, and is adaptive to presorted inputs/subsets
7969 : : * (which are typical here).
7970 : : */
7971 : : static void
1187 pg@bowt.ie 7972 :CBC 4985 : index_delete_sort(TM_IndexDeleteOp *delstate)
7973 : : {
7974 : 4985 : TM_IndexDelete *deltids = delstate->deltids;
7975 : 4985 : int ndeltids = delstate->ndeltids;
7976 : 4985 : int low = 0;
7977 : :
7978 : : /*
7979 : : * Shellsort gap sequence (taken from Sedgewick-Incerpi paper).
7980 : : *
7981 : : * This implementation is fast with array sizes up to ~4500. This covers
7982 : : * all supported BLCKSZ values.
7983 : : */
7984 : 4985 : const int gaps[9] = {1968, 861, 336, 112, 48, 21, 7, 3, 1};
7985 : :
7986 : : /* Think carefully before changing anything here -- keep swaps cheap */
7987 : : StaticAssertDecl(sizeof(TM_IndexDelete) <= 8,
7988 : : "element size exceeds 8 bytes");
7989 : :
7990 [ + + ]: 49850 : for (int g = 0; g < lengthof(gaps); g++)
7991 : : {
7992 [ + + ]: 7011693 : for (int hi = gaps[g], i = low + hi; i < ndeltids; i++)
7993 : : {
7994 : 6966828 : TM_IndexDelete d = deltids[i];
7995 : 6966828 : int j = i;
7996 : :
7997 [ + + + + ]: 12061452 : while (j >= hi && index_delete_sort_cmp(&deltids[j - hi], &d) >= 0)
7998 : : {
7999 : 5094624 : deltids[j] = deltids[j - hi];
8000 : 5094624 : j -= hi;
8001 : : }
8002 : 6966828 : deltids[j] = d;
8003 : : }
8004 : : }
8005 : 4985 : }
8006 : :
8007 : : /*
8008 : : * Returns how many blocks should be considered favorable/contiguous for a
8009 : : * bottom-up index deletion pass. This is a number of heap blocks that starts
8010 : : * from and includes the first block in line.
8011 : : *
8012 : : * There is always at least one favorable block during bottom-up index
8013 : : * deletion. In the worst case (i.e. with totally random heap blocks) the
8014 : : * first block in line (the only favorable block) can be thought of as a
8015 : : * degenerate array of contiguous blocks that consists of a single block.
8016 : : * heap_index_delete_tuples() will expect this.
8017 : : *
8018 : : * Caller passes blockgroups, a description of the final order that deltids
8019 : : * will be sorted in for heap_index_delete_tuples() bottom-up index deletion
8020 : : * processing. Note that deltids need not actually be sorted just yet (caller
8021 : : * only passes deltids to us so that we can interpret blockgroups).
8022 : : *
8023 : : * You might guess that the existence of contiguous blocks cannot matter much,
8024 : : * since in general the main factor that determines which blocks we visit is
8025 : : * the number of promising TIDs, which is a fixed hint from the index AM.
8026 : : * We're not really targeting the general case, though -- the actual goal is
8027 : : * to adapt our behavior to a wide variety of naturally occurring conditions.
8028 : : * The effects of most of the heuristics we apply are only noticeable in the
8029 : : * aggregate, over time and across many _related_ bottom-up index deletion
8030 : : * passes.
8031 : : *
8032 : : * Deeming certain blocks favorable allows heapam to recognize and adapt to
8033 : : * workloads where heap blocks visited during bottom-up index deletion can be
8034 : : * accessed contiguously, in the sense that each newly visited block is the
8035 : : * neighbor of the block that bottom-up deletion just finished processing (or
8036 : : * close enough to it). It will likely be cheaper to access more favorable
8037 : : * blocks sooner rather than later (e.g. in this pass, not across a series of
8038 : : * related bottom-up passes). Either way it is probably only a matter of time
8039 : : * (or a matter of further correlated version churn) before all blocks that
8040 : : * appear together as a single large batch of favorable blocks get accessed by
8041 : : * _some_ bottom-up pass. Large batches of favorable blocks tend to either
8042 : : * appear almost constantly or not even once (it all depends on per-index
8043 : : * workload characteristics).
8044 : : *
8045 : : * Note that the blockgroups sort order applies a power-of-two bucketing
8046 : : * scheme that creates opportunities for contiguous groups of blocks to get
8047 : : * batched together, at least with workloads that are naturally amenable to
8048 : : * being driven by heap block locality. This doesn't just enhance the spatial
8049 : : * locality of bottom-up heap block processing in the obvious way. It also
8050 : : * enables temporal locality of access, since sorting by heap block number
8051 : : * naturally tends to make the bottom-up processing order deterministic.
8052 : : *
8053 : : * Consider the following example to get a sense of how temporal locality
8054 : : * might matter: There is a heap relation with several indexes, each of which
8055 : : * is low to medium cardinality. It is subject to constant non-HOT updates.
8056 : : * The updates are skewed (in one part of the primary key, perhaps). None of
8057 : : * the indexes are logically modified by the UPDATE statements (if they were
8058 : : * then bottom-up index deletion would not be triggered in the first place).
8059 : : * Naturally, each new round of index tuples (for each heap tuple that gets a
8060 : : * heap_update() call) will have the same heap TID in each and every index.
8061 : : * Since these indexes are low cardinality and never get logically modified,
8062 : : * heapam processing during bottom-up deletion passes will access heap blocks
8063 : : * in approximately sequential order. Temporal locality of access occurs due
8064 : : * to bottom-up deletion passes behaving very similarly across each of the
8065 : : * indexes at any given moment. This keeps the number of buffer misses needed
8066 : : * to visit heap blocks to a minimum.
8067 : : */
8068 : : static int
8069 : 1729 : bottomup_nblocksfavorable(IndexDeleteCounts *blockgroups, int nblockgroups,
8070 : : TM_IndexDelete *deltids)
8071 : : {
8072 : 1729 : int64 lastblock = -1;
8073 : 1729 : int nblocksfavorable = 0;
8074 : :
8075 [ - + ]: 1729 : Assert(nblockgroups >= 1);
8076 [ - + ]: 1729 : Assert(nblockgroups <= BOTTOMUP_MAX_NBLOCKS);
8077 : :
8078 : : /*
8079 : : * We tolerate heap blocks that will be accessed only slightly out of
8080 : : * physical order. Small blips occur when a pair of almost-contiguous
8081 : : * blocks happen to fall into different buckets (perhaps due only to a
8082 : : * small difference in npromisingtids that the bucketing scheme didn't
8083 : : * quite manage to ignore). We effectively ignore these blips by applying
8084 : : * a small tolerance. The precise tolerance we use is a little arbitrary,
8085 : : * but it works well enough in practice.
8086 : : */
8087 [ + + ]: 5478 : for (int b = 0; b < nblockgroups; b++)
8088 : : {
8089 : 5256 : IndexDeleteCounts *group = blockgroups + b;
8090 : 5256 : TM_IndexDelete *firstdtid = deltids + group->ifirsttid;
8091 : 5256 : BlockNumber block = ItemPointerGetBlockNumber(&firstdtid->tid);
8092 : :
8093 [ + + ]: 5256 : if (lastblock != -1 &&
8094 [ + + ]: 3527 : ((int64) block < lastblock - BOTTOMUP_TOLERANCE_NBLOCKS ||
8095 [ + + ]: 2997 : (int64) block > lastblock + BOTTOMUP_TOLERANCE_NBLOCKS))
8096 : : break;
8097 : :
8098 : 3749 : nblocksfavorable++;
8099 : 3749 : lastblock = block;
8100 : : }
8101 : :
8102 : : /* Always indicate that there is at least 1 favorable block */
8103 [ - + ]: 1729 : Assert(nblocksfavorable >= 1);
8104 : :
8105 : 1729 : return nblocksfavorable;
8106 : : }
8107 : :
8108 : : /*
8109 : : * qsort comparison function for bottomup_sort_and_shrink()
8110 : : */
8111 : : static int
8112 : 181444 : bottomup_sort_and_shrink_cmp(const void *arg1, const void *arg2)
8113 : : {
8114 : 181444 : const IndexDeleteCounts *group1 = (const IndexDeleteCounts *) arg1;
8115 : 181444 : const IndexDeleteCounts *group2 = (const IndexDeleteCounts *) arg2;
8116 : :
8117 : : /*
8118 : : * Most significant field is npromisingtids (which we invert the order of
8119 : : * so as to sort in desc order).
8120 : : *
8121 : : * Caller should have already normalized npromisingtids fields into
8122 : : * power-of-two values (buckets).
8123 : : */
8124 [ + + ]: 181444 : if (group1->npromisingtids > group2->npromisingtids)
8125 : 7233 : return -1;
8126 [ + + ]: 174211 : if (group1->npromisingtids < group2->npromisingtids)
8127 : 9060 : return 1;
8128 : :
8129 : : /*
8130 : : * Tiebreak: desc ntids sort order.
8131 : : *
8132 : : * We cannot expect power-of-two values for ntids fields. We should
8133 : : * behave as if they were already rounded up for us instead.
8134 : : */
8135 [ + + ]: 165151 : if (group1->ntids != group2->ntids)
8136 : : {
8137 : 117883 : uint32 ntids1 = pg_nextpower2_32((uint32) group1->ntids);
8138 : 117883 : uint32 ntids2 = pg_nextpower2_32((uint32) group2->ntids);
8139 : :
8140 [ + + ]: 117883 : if (ntids1 > ntids2)
8141 : 16017 : return -1;
8142 [ + + ]: 101866 : if (ntids1 < ntids2)
8143 : 20295 : return 1;
8144 : : }
8145 : :
8146 : : /*
8147 : : * Tiebreak: asc offset-into-deltids-for-block (offset to first TID for
8148 : : * block in deltids array) order.
8149 : : *
8150 : : * This is equivalent to sorting in ascending heap block number order
8151 : : * (among otherwise equal subsets of the array). This approach allows us
8152 : : * to avoid accessing the out-of-line TID. (We rely on the assumption
8153 : : * that the deltids array was sorted in ascending heap TID order when
8154 : : * these offsets to the first TID from each heap block group were formed.)
8155 : : */
8156 [ + + ]: 128839 : if (group1->ifirsttid > group2->ifirsttid)
8157 : 63435 : return 1;
8158 [ + - ]: 65404 : if (group1->ifirsttid < group2->ifirsttid)
8159 : 65404 : return -1;
8160 : :
1187 pg@bowt.ie 8161 :UBC 0 : pg_unreachable();
8162 : :
8163 : : return 0;
8164 : : }
8165 : :
8166 : : /*
8167 : : * heap_index_delete_tuples() helper function for bottom-up deletion callers.
8168 : : *
8169 : : * Sorts deltids array in the order needed for useful processing by bottom-up
8170 : : * deletion. The array should already be sorted in TID order when we're
8171 : : * called. The sort process groups heap TIDs from deltids into heap block
8172 : : * groupings. Earlier/more-promising groups/blocks are usually those that are
8173 : : * known to have the most "promising" TIDs.
8174 : : *
8175 : : * Sets new size of deltids array (ndeltids) in state. deltids will only have
8176 : : * TIDs from the BOTTOMUP_MAX_NBLOCKS most promising heap blocks when we
8177 : : * return. This often means that deltids will be shrunk to a small fraction
8178 : : * of its original size (we eliminate many heap blocks from consideration for
8179 : : * caller up front).
8180 : : *
8181 : : * Returns the number of "favorable" blocks. See bottomup_nblocksfavorable()
8182 : : * for a definition and full details.
8183 : : */
8184 : : static int
1187 pg@bowt.ie 8185 :CBC 1729 : bottomup_sort_and_shrink(TM_IndexDeleteOp *delstate)
8186 : : {
8187 : : IndexDeleteCounts *blockgroups;
8188 : : TM_IndexDelete *reordereddeltids;
8189 : 1729 : BlockNumber curblock = InvalidBlockNumber;
8190 : 1729 : int nblockgroups = 0;
8191 : 1729 : int ncopied = 0;
8192 : 1729 : int nblocksfavorable = 0;
8193 : :
8194 [ - + ]: 1729 : Assert(delstate->bottomup);
8195 [ - + ]: 1729 : Assert(delstate->ndeltids > 0);
8196 : :
8197 : : /* Calculate per-heap-block count of TIDs */
8198 : 1729 : blockgroups = palloc(sizeof(IndexDeleteCounts) * delstate->ndeltids);
8199 [ + + ]: 869044 : for (int i = 0; i < delstate->ndeltids; i++)
8200 : : {
8201 : 867315 : TM_IndexDelete *ideltid = &delstate->deltids[i];
8202 : 867315 : TM_IndexStatus *istatus = delstate->status + ideltid->id;
8203 : 867315 : ItemPointer htid = &ideltid->tid;
8204 : 867315 : bool promising = istatus->promising;
8205 : :
8206 [ + + ]: 867315 : if (curblock != ItemPointerGetBlockNumber(htid))
8207 : : {
8208 : : /* New block group */
8209 : 34141 : nblockgroups++;
8210 : :
8211 [ + + - + ]: 34141 : Assert(curblock < ItemPointerGetBlockNumber(htid) ||
8212 : : !BlockNumberIsValid(curblock));
8213 : :
8214 : 34141 : curblock = ItemPointerGetBlockNumber(htid);
8215 : 34141 : blockgroups[nblockgroups - 1].ifirsttid = i;
8216 : 34141 : blockgroups[nblockgroups - 1].ntids = 1;
8217 : 34141 : blockgroups[nblockgroups - 1].npromisingtids = 0;
8218 : : }
8219 : : else
8220 : : {
8221 : 833174 : blockgroups[nblockgroups - 1].ntids++;
8222 : : }
8223 : :
8224 [ + + ]: 867315 : if (promising)
8225 : 106062 : blockgroups[nblockgroups - 1].npromisingtids++;
8226 : : }
8227 : :
8228 : : /*
8229 : : * We're about ready to sort block groups to determine the optimal order
8230 : : * for visiting heap blocks. But before we do, round the number of
8231 : : * promising tuples for each block group up to the next power-of-two,
8232 : : * unless it is very low (less than 4), in which case we round up to 4.
8233 : : * npromisingtids is far too noisy to trust when choosing between a pair
8234 : : * of block groups that both have very low values.
8235 : : *
8236 : : * This scheme divides heap blocks/block groups into buckets. Each bucket
8237 : : * contains blocks that have _approximately_ the same number of promising
8238 : : * TIDs as each other. The goal is to ignore relatively small differences
8239 : : * in the total number of promising entries, so that the whole process can
8240 : : * give a little weight to heapam factors (like heap block locality)
8241 : : * instead. This isn't a trade-off, really -- we have nothing to lose. It
8242 : : * would be foolish to interpret small differences in npromisingtids
8243 : : * values as anything more than noise.
8244 : : *
8245 : : * We tiebreak on nhtids when sorting block group subsets that have the
8246 : : * same npromisingtids, but this has the same issues as npromisingtids,
8247 : : * and so nhtids is subject to the same power-of-two bucketing scheme. The
8248 : : * only reason that we don't fix nhtids in the same way here too is that
8249 : : * we'll need accurate nhtids values after the sort. We handle nhtids
8250 : : * bucketization dynamically instead (in the sort comparator).
8251 : : *
8252 : : * See bottomup_nblocksfavorable() for a full explanation of when and how
8253 : : * heap locality/favorable blocks can significantly influence when and how
8254 : : * heap blocks are accessed.
8255 : : */
8256 [ + + ]: 35870 : for (int b = 0; b < nblockgroups; b++)
8257 : : {
8258 : 34141 : IndexDeleteCounts *group = blockgroups + b;
8259 : :
8260 : : /* Better off falling back on nhtids with low npromisingtids */
8261 [ + + ]: 34141 : if (group->npromisingtids <= 4)
8262 : 29291 : group->npromisingtids = 4;
8263 : : else
8264 : 4850 : group->npromisingtids =
8265 : 4850 : pg_nextpower2_32((uint32) group->npromisingtids);
8266 : : }
8267 : :
8268 : : /* Sort groups and rearrange caller's deltids array */
8269 : 1729 : qsort(blockgroups, nblockgroups, sizeof(IndexDeleteCounts),
8270 : : bottomup_sort_and_shrink_cmp);
8271 : 1729 : reordereddeltids = palloc(delstate->ndeltids * sizeof(TM_IndexDelete));
8272 : :
8273 : 1729 : nblockgroups = Min(BOTTOMUP_MAX_NBLOCKS, nblockgroups);
8274 : : /* Determine number of favorable blocks at the start of final deltids */
8275 : 1729 : nblocksfavorable = bottomup_nblocksfavorable(blockgroups, nblockgroups,
8276 : : delstate->deltids);
8277 : :
8278 [ + + ]: 11493 : for (int b = 0; b < nblockgroups; b++)
8279 : : {
8280 : 9764 : IndexDeleteCounts *group = blockgroups + b;
8281 : 9764 : TM_IndexDelete *firstdtid = delstate->deltids + group->ifirsttid;
8282 : :
8283 : 9764 : memcpy(reordereddeltids + ncopied, firstdtid,
8284 : 9764 : sizeof(TM_IndexDelete) * group->ntids);
8285 : 9764 : ncopied += group->ntids;
8286 : : }
8287 : :
8288 : : /* Copy final grouped and sorted TIDs back into start of caller's array */
8289 : 1729 : memcpy(delstate->deltids, reordereddeltids,
8290 : : sizeof(TM_IndexDelete) * ncopied);
8291 : 1729 : delstate->ndeltids = ncopied;
8292 : :
8293 : 1729 : pfree(reordereddeltids);
8294 : 1729 : pfree(blockgroups);
8295 : :
8296 : 1729 : return nblocksfavorable;
8297 : : }
8298 : :
8299 : : /*
8300 : : * Perform XLogInsert for a heap-visible operation. 'block' is the block
8301 : : * being marked all-visible, and vm_buffer is the buffer containing the
8302 : : * corresponding visibility map block. Both should have already been modified
8303 : : * and dirtied.
8304 : : *
8305 : : * snapshotConflictHorizon comes from the largest xmin on the page being
8306 : : * marked all-visible. REDO routine uses it to generate recovery conflicts.
8307 : : *
8308 : : * If checksums or wal_log_hints are enabled, we may also generate a full-page
8309 : : * image of heap_buffer. Otherwise, we optimize away the FPI (by specifying
8310 : : * REGBUF_NO_IMAGE for the heap buffer), in which case the caller should *not*
8311 : : * update the heap page's LSN.
8312 : : */
8313 : : XLogRecPtr
379 andres@anarazel.de 8314 : 39281 : log_heap_visible(Relation rel, Buffer heap_buffer, Buffer vm_buffer,
8315 : : TransactionId snapshotConflictHorizon, uint8 vmflags)
8316 : : {
8317 : : xl_heap_visible xlrec;
8318 : : XLogRecPtr recptr;
8319 : : uint8 flags;
8320 : :
4041 simon@2ndQuadrant.co 8321 [ - + ]: 39281 : Assert(BufferIsValid(heap_buffer));
8322 [ - + ]: 39281 : Assert(BufferIsValid(vm_buffer));
8323 : :
514 pg@bowt.ie 8324 : 39281 : xlrec.snapshotConflictHorizon = snapshotConflictHorizon;
2966 rhaas@postgresql.org 8325 : 39281 : xlrec.flags = vmflags;
378 andres@anarazel.de 8326 [ + + + - : 39281 : if (RelationIsAccessibleInLogicalDecoding(rel))
- + - - -
- + + - +
- - - - -
- ]
8327 : 139 : xlrec.flags |= VISIBILITYMAP_XLOG_CATALOG_REL;
3433 heikki.linnakangas@i 8328 : 39281 : XLogBeginInsert();
8329 : 39281 : XLogRegisterData((char *) &xlrec, SizeOfHeapVisible);
8330 : :
8331 : 39281 : XLogRegisterBuffer(0, vm_buffer, 0);
8332 : :
8333 : 39281 : flags = REGBUF_STANDARD;
8334 [ + + + + ]: 39281 : if (!XLogHintBitIsNeeded())
8335 : 31134 : flags |= REGBUF_NO_IMAGE;
8336 : 39281 : XLogRegisterBuffer(1, heap_buffer, flags);
8337 : :
8338 : 39281 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_VISIBLE);
8339 : :
4681 rhaas@postgresql.org 8340 : 39281 : return recptr;
8341 : : }
8342 : :
8343 : : /*
8344 : : * Perform XLogInsert for a heap-update operation. Caller must already
8345 : : * have modified the buffer(s) and marked them dirty.
8346 : : */
8347 : : static XLogRecPtr
4099 alvherre@alvh.no-ip. 8348 : 265654 : log_heap_update(Relation reln, Buffer oldbuf,
8349 : : Buffer newbuf, HeapTuple oldtup, HeapTuple newtup,
8350 : : HeapTuple old_key_tuple,
8351 : : bool all_visible_cleared, bool new_all_visible_cleared)
8352 : : {
8353 : : xl_heap_update xlrec;
8354 : : xl_heap_header xlhdr;
8355 : : xl_heap_header xlhdr_idx;
8356 : : uint8 info;
8357 : : uint16 prefix_suffix[2];
3686 heikki.linnakangas@i 8358 : 265654 : uint16 prefixlen = 0,
8359 : 265654 : suffixlen = 0;
8360 : : XLogRecPtr recptr;
2916 kgrittn@postgresql.o 8361 : 265654 : Page page = BufferGetPage(newbuf);
3778 rhaas@postgresql.org 8362 [ + + + - : 265654 : bool need_tuple_data = RelationIsLogicallyLogged(reln);
- + - - -
- + - +
+ ]
8363 : : bool init;
8364 : : int bufflags;
8365 : :
8366 : : /* Caller should not call me on a non-WAL-logged relation */
4871 8367 [ + - + + : 265654 : Assert(RelationNeedsWAL(reln));
+ - - + ]
8368 : :
3433 heikki.linnakangas@i 8369 : 265654 : XLogBeginInsert();
8370 : :
5179 tgl@sss.pgh.pa.us 8371 [ + + ]: 265654 : if (HeapTupleIsHeapOnly(newtup))
6051 8372 : 124964 : info = XLOG_HEAP_HOT_UPDATE;
8373 : : else
8374 : 140690 : info = XLOG_HEAP_UPDATE;
8375 : :
8376 : : /*
8377 : : * If the old and new tuple are on the same page, we only need to log the
8378 : : * parts of the new tuple that were changed. That saves on the amount of
8379 : : * WAL we need to write. Currently, we just count any unchanged bytes in
8380 : : * the beginning and end of the tuple. That's quick to check, and
8381 : : * perfectly covers the common case that only one field is updated.
8382 : : *
8383 : : * We could do this even if the old and new tuple are on different pages,
8384 : : * but only if we don't make a full-page image of the old page, which is
8385 : : * difficult to know in advance. Also, if the old tuple is corrupt for
8386 : : * some reason, it would allow the corruption to propagate the new page,
8387 : : * so it seems best to avoid. Under the general assumption that most
8388 : : * updates tend to create the new tuple version on the same page, there
8389 : : * isn't much to be gained by doing this across pages anyway.
8390 : : *
8391 : : * Skip this if we're taking a full-page image of the new page, as we
8392 : : * don't include the new tuple in the WAL record in that case. Also
8393 : : * disable if wal_level='logical', as logical decoding needs to be able to
8394 : : * read the new tuple in whole from the WAL record alone.
8395 : : */
3686 heikki.linnakangas@i 8396 [ + + + + ]: 265654 : if (oldbuf == newbuf && !need_tuple_data &&
8397 [ + + ]: 123702 : !XLogCheckBufferNeedsBackup(newbuf))
8398 : : {
8399 : 123320 : char *oldp = (char *) oldtup->t_data + oldtup->t_data->t_hoff;
8400 : 123320 : char *newp = (char *) newtup->t_data + newtup->t_data->t_hoff;
8401 : 123320 : int oldlen = oldtup->t_len - oldtup->t_data->t_hoff;
8402 : 123320 : int newlen = newtup->t_len - newtup->t_data->t_hoff;
8403 : :
8404 : : /* Check for common prefix between old and new tuple */
8405 [ + + ]: 9182180 : for (prefixlen = 0; prefixlen < Min(oldlen, newlen); prefixlen++)
8406 : : {
8407 [ + + ]: 9161273 : if (newp[prefixlen] != oldp[prefixlen])
8408 : 102413 : break;
8409 : : }
8410 : :
8411 : : /*
8412 : : * Storing the length of the prefix takes 2 bytes, so we need to save
8413 : : * at least 3 bytes or there's no point.
8414 : : */
8415 [ + + ]: 123320 : if (prefixlen < 3)
8416 : 22042 : prefixlen = 0;
8417 : :
8418 : : /* Same for suffix */
8419 [ + + ]: 4144937 : for (suffixlen = 0; suffixlen < Min(oldlen, newlen) - prefixlen; suffixlen++)
8420 : : {
8421 [ + + ]: 4123816 : if (newp[newlen - suffixlen - 1] != oldp[oldlen - suffixlen - 1])
8422 : 102199 : break;
8423 : : }
8424 [ + + ]: 123320 : if (suffixlen < 3)
8425 : 30252 : suffixlen = 0;
8426 : : }
8427 : :
8428 : : /* Prepare main WAL data chain */
3778 rhaas@postgresql.org 8429 : 265654 : xlrec.flags = 0;
8430 [ + + ]: 265654 : if (all_visible_cleared)
3264 andres@anarazel.de 8431 : 1175 : xlrec.flags |= XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED;
3778 rhaas@postgresql.org 8432 [ + + ]: 265654 : if (new_all_visible_cleared)
3264 andres@anarazel.de 8433 : 815 : xlrec.flags |= XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED;
3686 heikki.linnakangas@i 8434 [ + + ]: 265654 : if (prefixlen > 0)
3264 andres@anarazel.de 8435 : 101278 : xlrec.flags |= XLH_UPDATE_PREFIX_FROM_OLD;
3686 heikki.linnakangas@i 8436 [ + + ]: 265654 : if (suffixlen > 0)
3264 andres@anarazel.de 8437 : 93068 : xlrec.flags |= XLH_UPDATE_SUFFIX_FROM_OLD;
3433 heikki.linnakangas@i 8438 [ + + ]: 265654 : if (need_tuple_data)
8439 : : {
3264 andres@anarazel.de 8440 : 47008 : xlrec.flags |= XLH_UPDATE_CONTAINS_NEW_TUPLE;
3433 heikki.linnakangas@i 8441 [ + + ]: 47008 : if (old_key_tuple)
8442 : : {
8443 [ + + ]: 133 : if (reln->rd_rel->relreplident == REPLICA_IDENTITY_FULL)
3264 andres@anarazel.de 8444 : 56 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_TUPLE;
8445 : : else
8446 : 77 : xlrec.flags |= XLH_UPDATE_CONTAINS_OLD_KEY;
8447 : : }
8448 : : }
8449 : :
8450 : : /* If new tuple is the single and first tuple on page... */
3686 heikki.linnakangas@i 8451 [ + + + + ]: 268589 : if (ItemPointerGetOffsetNumber(&(newtup->t_self)) == FirstOffsetNumber &&
8452 : 2935 : PageGetMaxOffsetNumber(page) == FirstOffsetNumber)
8453 : : {
8454 : 2753 : info |= XLOG_HEAP_INIT_PAGE;
3433 8455 : 2753 : init = true;
8456 : : }
8457 : : else
8458 : 262901 : init = false;
8459 : :
8460 : : /* Prepare WAL data for the old page */
8461 : 265654 : xlrec.old_offnum = ItemPointerGetOffsetNumber(&oldtup->t_self);
8462 : 265654 : xlrec.old_xmax = HeapTupleHeaderGetRawXmax(oldtup->t_data);
8463 : 531308 : xlrec.old_infobits_set = compute_infobits(oldtup->t_data->t_infomask,
8464 : 265654 : oldtup->t_data->t_infomask2);
8465 : :
8466 : : /* Prepare WAL data for the new page */
8467 : 265654 : xlrec.new_offnum = ItemPointerGetOffsetNumber(&newtup->t_self);
8468 : 265654 : xlrec.new_xmax = HeapTupleHeaderGetRawXmax(newtup->t_data);
8469 : :
8470 : 265654 : bufflags = REGBUF_STANDARD;
8471 [ + + ]: 265654 : if (init)
8472 : 2753 : bufflags |= REGBUF_WILL_INIT;
8473 [ + + ]: 265654 : if (need_tuple_data)
8474 : 47008 : bufflags |= REGBUF_KEEP_DATA;
8475 : :
8476 : 265654 : XLogRegisterBuffer(0, newbuf, bufflags);
8477 [ + + ]: 265654 : if (oldbuf != newbuf)
8478 : 130024 : XLogRegisterBuffer(1, oldbuf, REGBUF_STANDARD);
8479 : :
8480 : 265654 : XLogRegisterData((char *) &xlrec, SizeOfHeapUpdate);
8481 : :
8482 : : /*
8483 : : * Prepare WAL data for the new tuple.
8484 : : */
3686 8485 [ + + + + ]: 265654 : if (prefixlen > 0 || suffixlen > 0)
8486 : : {
8487 [ + + + + ]: 122889 : if (prefixlen > 0 && suffixlen > 0)
8488 : : {
8489 : 71457 : prefix_suffix[0] = prefixlen;
8490 : 71457 : prefix_suffix[1] = suffixlen;
3433 8491 : 71457 : XLogRegisterBufData(0, (char *) &prefix_suffix, sizeof(uint16) * 2);
8492 : : }
3686 8493 [ + + ]: 51432 : else if (prefixlen > 0)
8494 : : {
3433 8495 : 29821 : XLogRegisterBufData(0, (char *) &prefixlen, sizeof(uint16));
8496 : : }
8497 : : else
8498 : : {
8499 : 21611 : XLogRegisterBufData(0, (char *) &suffixlen, sizeof(uint16));
8500 : : }
8501 : : }
8502 : :
8503 : 265654 : xlhdr.t_infomask2 = newtup->t_data->t_infomask2;
8504 : 265654 : xlhdr.t_infomask = newtup->t_data->t_infomask;
8505 : 265654 : xlhdr.t_hoff = newtup->t_data->t_hoff;
3340 tgl@sss.pgh.pa.us 8506 [ - + ]: 265654 : Assert(SizeofHeapTupleHeader + prefixlen + suffixlen <= newtup->t_len);
8507 : :
8508 : : /*
8509 : : * PG73FORMAT: write bitmap [+ padding] [+ oid] + data
8510 : : *
8511 : : * The 'data' doesn't include the common prefix or suffix.
8512 : : */
3433 heikki.linnakangas@i 8513 : 265654 : XLogRegisterBufData(0, (char *) &xlhdr, SizeOfHeapHeader);
3686 8514 [ + + ]: 265654 : if (prefixlen == 0)
8515 : : {
3433 8516 : 164376 : XLogRegisterBufData(0,
3340 tgl@sss.pgh.pa.us 8517 : 164376 : ((char *) newtup->t_data) + SizeofHeapTupleHeader,
2489 8518 : 164376 : newtup->t_len - SizeofHeapTupleHeader - suffixlen);
8519 : : }
8520 : : else
8521 : : {
8522 : : /*
8523 : : * Have to write the null bitmap and data after the common prefix as
8524 : : * two separate rdata entries.
8525 : : */
8526 : : /* bitmap [+ padding] [+ oid] */
3340 8527 [ + - ]: 101278 : if (newtup->t_data->t_hoff - SizeofHeapTupleHeader > 0)
8528 : : {
3433 heikki.linnakangas@i 8529 : 101278 : XLogRegisterBufData(0,
2489 tgl@sss.pgh.pa.us 8530 : 101278 : ((char *) newtup->t_data) + SizeofHeapTupleHeader,
8531 : 101278 : newtup->t_data->t_hoff - SizeofHeapTupleHeader);
8532 : : }
8533 : :
8534 : : /* data after common prefix */
3433 heikki.linnakangas@i 8535 : 101278 : XLogRegisterBufData(0,
2489 tgl@sss.pgh.pa.us 8536 : 101278 : ((char *) newtup->t_data) + newtup->t_data->t_hoff + prefixlen,
8537 : 101278 : newtup->t_len - newtup->t_data->t_hoff - prefixlen - suffixlen);
8538 : : }
8539 : :
8540 : : /* We need to log a tuple identity */
3433 heikki.linnakangas@i 8541 [ + + + + ]: 265654 : if (need_tuple_data && old_key_tuple)
8542 : : {
8543 : : /* don't really need this, but its more comfy to decode */
8544 : 133 : xlhdr_idx.t_infomask2 = old_key_tuple->t_data->t_infomask2;
8545 : 133 : xlhdr_idx.t_infomask = old_key_tuple->t_data->t_infomask;
8546 : 133 : xlhdr_idx.t_hoff = old_key_tuple->t_data->t_hoff;
8547 : :
8548 : 133 : XLogRegisterData((char *) &xlhdr_idx, SizeOfHeapHeader);
8549 : :
8550 : : /* PG73FORMAT: write bitmap [+ padding] [+ oid] + data */
3340 tgl@sss.pgh.pa.us 8551 : 133 : XLogRegisterData((char *) old_key_tuple->t_data + SizeofHeapTupleHeader,
8552 : 133 : old_key_tuple->t_len - SizeofHeapTupleHeader);
8553 : : }
8554 : :
8555 : : /* filtering by origin on a row level is much more efficient */
2670 andres@anarazel.de 8556 : 265654 : XLogSetRecordFlags(XLOG_INCLUDE_ORIGIN);
8557 : :
3433 heikki.linnakangas@i 8558 : 265654 : recptr = XLogInsert(RM_HEAP_ID, info);
8559 : :
6668 neilc@samurai.com 8560 : 265654 : return recptr;
8561 : : }
8562 : :
8563 : : /*
8564 : : * Perform XLogInsert of an XLOG_HEAP2_NEW_CID record
8565 : : *
8566 : : * This is only used in wal_level >= WAL_LEVEL_LOGICAL, and only for catalog
8567 : : * tuples.
8568 : : */
8569 : : static XLogRecPtr
3778 rhaas@postgresql.org 8570 : 21457 : log_heap_new_cid(Relation relation, HeapTuple tup)
8571 : : {
8572 : : xl_heap_new_cid xlrec;
8573 : :
8574 : : XLogRecPtr recptr;
8575 : 21457 : HeapTupleHeader hdr = tup->t_data;
8576 : :
8577 [ - + ]: 21457 : Assert(ItemPointerIsValid(&tup->t_self));
8578 [ - + ]: 21457 : Assert(tup->t_tableOid != InvalidOid);
8579 : :
8580 : 21457 : xlrec.top_xid = GetTopTransactionId();
648 8581 : 21457 : xlrec.target_locator = relation->rd_locator;
3433 heikki.linnakangas@i 8582 : 21457 : xlrec.target_tid = tup->t_self;
8583 : :
8584 : : /*
8585 : : * If the tuple got inserted & deleted in the same TX we definitely have a
8586 : : * combo CID, set cmin and cmax.
8587 : : */
3778 rhaas@postgresql.org 8588 [ + + ]: 21457 : if (hdr->t_infomask & HEAP_COMBOCID)
8589 : : {
8590 [ - + ]: 1928 : Assert(!(hdr->t_infomask & HEAP_XMAX_INVALID));
3766 8591 [ - + ]: 1928 : Assert(!HeapTupleHeaderXminInvalid(hdr));
3778 8592 : 1928 : xlrec.cmin = HeapTupleHeaderGetCmin(hdr);
8593 : 1928 : xlrec.cmax = HeapTupleHeaderGetCmax(hdr);
8594 : 1928 : xlrec.combocid = HeapTupleHeaderGetRawCommandId(hdr);
8595 : : }
8596 : : /* No combo CID, so only cmin or cmax can be set by this TX */
8597 : : else
8598 : : {
8599 : : /*
8600 : : * Tuple inserted.
8601 : : *
8602 : : * We need to check for LOCK ONLY because multixacts might be
8603 : : * transferred to the new tuple in case of FOR KEY SHARE updates in
8604 : : * which case there will be an xmax, although the tuple just got
8605 : : * inserted.
8606 : : */
8607 [ + + ]: 19529 : if (hdr->t_infomask & HEAP_XMAX_INVALID ||
8608 [ + + - + ]: 5782 : HEAP_XMAX_IS_LOCKED_ONLY(hdr->t_infomask))
8609 : : {
8610 : 13748 : xlrec.cmin = HeapTupleHeaderGetRawCommandId(hdr);
8611 : 13748 : xlrec.cmax = InvalidCommandId;
8612 : : }
8613 : : /* Tuple from a different tx updated or deleted. */
8614 : : else
8615 : : {
8616 : 5781 : xlrec.cmin = InvalidCommandId;
8617 : 5781 : xlrec.cmax = HeapTupleHeaderGetRawCommandId(hdr);
8618 : : }
8619 : 19529 : xlrec.combocid = InvalidCommandId;
8620 : : }
8621 : :
8622 : : /*
8623 : : * Note that we don't need to register the buffer here, because this
8624 : : * operation does not modify the page. The insert/update/delete that
8625 : : * called us certainly did, but that's WAL-logged separately.
8626 : : */
3433 heikki.linnakangas@i 8627 : 21457 : XLogBeginInsert();
8628 : 21457 : XLogRegisterData((char *) &xlrec, SizeOfHeapNewCid);
8629 : :
8630 : : /* will be looked at irrespective of origin */
8631 : :
8632 : 21457 : recptr = XLogInsert(RM_HEAP2_ID, XLOG_HEAP2_NEW_CID);
8633 : :
3778 rhaas@postgresql.org 8634 : 21457 : return recptr;
8635 : : }
8636 : :
8637 : : /*
8638 : : * Build a heap tuple representing the configured REPLICA IDENTITY to represent
8639 : : * the old tuple in an UPDATE or DELETE.
8640 : : *
8641 : : * Returns NULL if there's no need to log an identity or if there's no suitable
8642 : : * key defined.
8643 : : *
8644 : : * Pass key_required true if any replica identity columns changed value, or if
8645 : : * any of them have any external data. Delete must always pass true.
8646 : : *
8647 : : * *copy is set to true if the returned tuple is a modified copy rather than
8648 : : * the same tuple that was passed in.
8649 : : */
8650 : : static HeapTuple
790 akapila@postgresql.o 8651 : 1709563 : ExtractReplicaIdentity(Relation relation, HeapTuple tp, bool key_required,
8652 : : bool *copy)
8653 : : {
3778 rhaas@postgresql.org 8654 : 1709563 : TupleDesc desc = RelationGetDescr(relation);
8655 : 1709563 : char replident = relation->rd_rel->relreplident;
8656 : : Bitmapset *idattrs;
8657 : : HeapTuple key_tuple;
8658 : : bool nulls[MaxHeapAttributeNumber];
8659 : : Datum values[MaxHeapAttributeNumber];
8660 : :
8661 : 1709563 : *copy = false;
8662 : :
8663 [ + + + + : 1709563 : if (!RelationIsLogicallyLogged(relation))
- + - - -
- + - +
+ ]
8664 : 1609303 : return NULL;
8665 : :
8666 [ + + ]: 100260 : if (replident == REPLICA_IDENTITY_NOTHING)
8667 : 231 : return NULL;
8668 : :
8669 [ + + ]: 100029 : if (replident == REPLICA_IDENTITY_FULL)
8670 : : {
8671 : : /*
8672 : : * When logging the entire old tuple, it very well could contain
8673 : : * toasted columns. If so, force them to be inlined.
8674 : : */
8675 [ + + ]: 177 : if (HeapTupleHasExternal(tp))
8676 : : {
8677 : 4 : *copy = true;
1686 tgl@sss.pgh.pa.us 8678 : 4 : tp = toast_flatten_tuple(tp, desc);
8679 : : }
3778 rhaas@postgresql.org 8680 : 177 : return tp;
8681 : : }
8682 : :
8683 : : /* if the key isn't required and we're only logging the key, we're done */
790 akapila@postgresql.o 8684 [ + + ]: 99852 : if (!key_required)
3778 rhaas@postgresql.org 8685 : 46875 : return NULL;
8686 : :
8687 : : /* find out the replica identity columns */
1686 tgl@sss.pgh.pa.us 8688 : 52977 : idattrs = RelationGetIndexAttrBitmap(relation,
8689 : : INDEX_ATTR_BITMAP_IDENTITY_KEY);
8690 : :
8691 : : /*
8692 : : * If there's no defined replica identity columns, treat as !key_required.
8693 : : * (This case should not be reachable from heap_update, since that should
8694 : : * calculate key_required accurately. But heap_delete just passes
8695 : : * constant true for key_required, so we can hit this case in deletes.)
8696 : : */
8697 [ + + ]: 52977 : if (bms_is_empty(idattrs))
8698 : 6021 : return NULL;
8699 : :
8700 : : /*
8701 : : * Construct a new tuple containing only the replica identity columns,
8702 : : * with nulls elsewhere. While we're at it, assert that the replica
8703 : : * identity columns aren't null.
8704 : : */
8705 : 46956 : heap_deform_tuple(tp, desc, values, nulls);
8706 : :
8707 [ + + ]: 150861 : for (int i = 0; i < desc->natts; i++)
8708 : : {
8709 [ + + ]: 103905 : if (bms_is_member(i + 1 - FirstLowInvalidHeapAttributeNumber,
8710 : : idattrs))
8711 [ - + ]: 46965 : Assert(!nulls[i]);
8712 : : else
8713 : 56940 : nulls[i] = true;
8714 : : }
8715 : :
3778 rhaas@postgresql.org 8716 : 46956 : key_tuple = heap_form_tuple(desc, values, nulls);
8717 : 46956 : *copy = true;
8718 : :
1686 tgl@sss.pgh.pa.us 8719 : 46956 : bms_free(idattrs);
8720 : :
8721 : : /*
8722 : : * If the tuple, which by here only contains indexed columns, still has
8723 : : * toasted columns, force them to be inlined. This is somewhat unlikely
8724 : : * since there's limits on the size of indexed columns, so we don't
8725 : : * duplicate toast_flatten_tuple()s functionality in the above loop over
8726 : : * the indexed columns, even if it would be more efficient.
8727 : : */
3778 rhaas@postgresql.org 8728 [ + + ]: 46956 : if (HeapTupleHasExternal(key_tuple))
8729 : : {
3631 bruce@momjian.us 8730 : 4 : HeapTuple oldtup = key_tuple;
8731 : :
1686 tgl@sss.pgh.pa.us 8732 : 4 : key_tuple = toast_flatten_tuple(oldtup, desc);
3778 rhaas@postgresql.org 8733 : 4 : heap_freetuple(oldtup);
8734 : : }
8735 : :
8736 : 46956 : return key_tuple;
8737 : : }
8738 : :
8739 : : /*
8740 : : * Replay XLOG_HEAP2_PRUNE_* records.
8741 : : */
8742 : : static void
20 heikki.linnakangas@i 8743 :GNC 13170 : heap_xlog_prune_freeze(XLogReaderState *record)
8744 : : {
3433 heikki.linnakangas@i 8745 :CBC 13170 : XLogRecPtr lsn = record->EndRecPtr;
20 heikki.linnakangas@i 8746 :GNC 13170 : char *maindataptr = XLogRecGetData(record);
8747 : : xl_heap_prune xlrec;
8748 : : Buffer buffer;
8749 : : RelFileLocator rlocator;
8750 : : BlockNumber blkno;
8751 : : XLogRedoAction action;
8752 : :
648 rhaas@postgresql.org 8753 :CBC 13170 : XLogRecGetBlockTag(record, 0, &rlocator, NULL, &blkno);
20 heikki.linnakangas@i 8754 :GNC 13170 : memcpy(&xlrec, maindataptr, SizeOfHeapPrune);
8755 : 13170 : maindataptr += SizeOfHeapPrune;
8756 : :
8757 : : /*
8758 : : * We will take an ordinary exclusive lock or a cleanup lock depending on
8759 : : * whether the XLHP_CLEANUP_LOCK flag is set. With an ordinary exclusive
8760 : : * lock, we better not be doing anything that requires moving existing
8761 : : * tuple data.
8762 : : */
8763 [ + + - + ]: 13170 : Assert((xlrec.flags & XLHP_CLEANUP_LOCK) != 0 ||
8764 : : (xlrec.flags & (XLHP_HAS_REDIRECTIONS | XLHP_HAS_DEAD_ITEMS)) == 0);
8765 : :
8766 : : /*
8767 : : * We are about to remove and/or freeze tuples. In Hot Standby mode,
8768 : : * ensure that there are no queries running for which the removed tuples
8769 : : * are still visible or which still consider the frozen xids as running.
8770 : : * The conflict horizon XID comes after xl_heap_prune.
8771 : : */
8772 [ + + ]: 13170 : if ((xlrec.flags & XLHP_HAS_CONFLICT_HORIZON) != 0)
8773 : : {
8774 : : TransactionId snapshot_conflict_horizon;
8775 : :
8776 : : /* memcpy() because snapshot_conflict_horizon is stored unaligned */
8777 : 8826 : memcpy(&snapshot_conflict_horizon, maindataptr, sizeof(TransactionId));
8778 : 8826 : maindataptr += sizeof(TransactionId);
8779 : :
8780 [ + + ]: 8826 : if (InHotStandby)
8781 : 8636 : ResolveRecoveryConflictWithSnapshot(snapshot_conflict_horizon,
8782 : 8636 : (xlrec.flags & XLHP_IS_CATALOG_REL) != 0,
8783 : : rlocator);
8784 : : }
8785 : :
8786 : : /*
8787 : : * If we have a full-page image, restore it and we're done.
8788 : : */
8789 : 13170 : action = XLogReadBufferForRedoExtended(record, 0, RBM_NORMAL,
8790 : 13170 : (xlrec.flags & XLHP_CLEANUP_LOCK) != 0,
8791 : : &buffer);
3532 heikki.linnakangas@i 8792 [ + + ]:CBC 13170 : if (action == BLK_NEEDS_REDO)
8793 : : {
2916 kgrittn@postgresql.o 8794 : 11958 : Page page = (Page) BufferGetPage(buffer);
8795 : : OffsetNumber *redirected;
8796 : : OffsetNumber *nowdead;
8797 : : OffsetNumber *nowunused;
8798 : : int nredirected;
8799 : : int ndead;
8800 : : int nunused;
8801 : : int nplans;
8802 : : Size datalen;
8803 : : xlhp_freeze_plan *plans;
8804 : : OffsetNumber *frz_offsets;
20 heikki.linnakangas@i 8805 :GNC 11958 : char *dataptr = XLogRecGetBlockData(record, 0, &datalen);
8806 : :
8807 : 11958 : heap_xlog_deserialize_prune_and_freeze(dataptr, xlrec.flags,
8808 : : &nplans, &plans, &frz_offsets,
8809 : : &nredirected, &redirected,
8810 : : &ndead, &nowdead,
8811 : : &nunused, &nowunused);
8812 : :
8813 : : /*
8814 : : * Update all line pointers per the record, and repair fragmentation
8815 : : * if needed.
8816 : : */
8817 [ + + + + : 11958 : if (nredirected > 0 || ndead > 0 || nunused > 0)
+ + ]
8818 : 11721 : heap_page_prune_execute(buffer,
8819 : 11721 : (xlrec.flags & XLHP_CLEANUP_LOCK) == 0,
8820 : : redirected, nredirected,
8821 : : nowdead, ndead,
8822 : : nowunused, nunused);
8823 : :
8824 : : /* Freeze tuples */
8825 [ + + ]: 12551 : for (int p = 0; p < nplans; p++)
8826 : : {
8827 : : HeapTupleFreeze frz;
8828 : :
8829 : : /*
8830 : : * Convert freeze plan representation from WAL record into
8831 : : * per-tuple format used by heap_execute_freeze_tuple
8832 : : */
8833 : 593 : frz.xmax = plans[p].xmax;
8834 : 593 : frz.t_infomask2 = plans[p].t_infomask2;
8835 : 593 : frz.t_infomask = plans[p].t_infomask;
8836 : 593 : frz.frzflags = plans[p].frzflags;
8837 : 593 : frz.offset = InvalidOffsetNumber; /* unused, but be tidy */
8838 : :
8839 [ + + ]: 20660 : for (int i = 0; i < plans[p].ntuples; i++)
8840 : : {
8841 : 20067 : OffsetNumber offset = *(frz_offsets++);
8842 : : ItemId lp;
8843 : : HeapTupleHeader tuple;
8844 : :
8845 : 20067 : lp = PageGetItemId(page, offset);
8846 : 20067 : tuple = (HeapTupleHeader) PageGetItem(page, lp);
8847 : 20067 : heap_execute_freeze_tuple(tuple, &frz);
8848 : : }
8849 : : }
8850 : :
8851 : : /* There should be no more data */
8852 [ - + ]: 11958 : Assert((char *) frz_offsets == dataptr + datalen);
8853 : :
8854 : : /*
8855 : : * Note: we don't worry about updating the page's prunability hints.
8856 : : * At worst this will cause an extra prune cycle to occur soon.
8857 : : */
8858 : :
1104 pg@bowt.ie 8859 :CBC 11958 : PageSetLSN(page, lsn);
8860 : 11958 : MarkBufferDirty(buffer);
8861 : : }
8862 : :
8863 : : /*
8864 : : * If we released any space or line pointers, update the free space map.
8865 : : *
8866 : : * Do this regardless of a full-page image being applied, since the FSM
8867 : : * data is not in the page anyway.
8868 : : */
8869 [ + - ]: 13170 : if (BufferIsValid(buffer))
8870 : : {
20 heikki.linnakangas@i 8871 [ + + ]:GNC 13170 : if (xlrec.flags & (XLHP_HAS_REDIRECTIONS |
8872 : : XLHP_HAS_DEAD_ITEMS |
8873 : : XLHP_HAS_NOW_UNUSED_ITEMS))
8874 : : {
8875 : 12932 : Size freespace = PageGetHeapFreeSpace(BufferGetPage(buffer));
8876 : :
8877 : 12932 : UnlockReleaseBuffer(buffer);
8878 : :
8879 : 12932 : XLogRecordPageWithFreeSpace(rlocator, blkno, freespace);
8880 : : }
8881 : : else
8882 : 238 : UnlockReleaseBuffer(buffer);
8883 : : }
1104 pg@bowt.ie 8884 :CBC 13170 : }
8885 : :
8886 : : /*
8887 : : * Replay XLOG_HEAP2_VISIBLE record.
8888 : : *
8889 : : * The critical integrity requirement here is that we must never end up with
8890 : : * a situation where the visibility map bit is set, and the page-level
8891 : : * PD_ALL_VISIBLE bit is clear. If that were to occur, then a subsequent
8892 : : * page modification would fail to clear the visibility map bit.
8893 : : */
8894 : : static void
3433 heikki.linnakangas@i 8895 : 7594 : heap_xlog_visible(XLogReaderState *record)
8896 : : {
8897 : 7594 : XLogRecPtr lsn = record->EndRecPtr;
4681 rhaas@postgresql.org 8898 : 7594 : xl_heap_visible *xlrec = (xl_heap_visible *) XLogRecGetData(record);
3433 heikki.linnakangas@i 8899 : 7594 : Buffer vmbuffer = InvalidBuffer;
8900 : : Buffer buffer;
8901 : : Page page;
8902 : : RelFileLocator rlocator;
8903 : : BlockNumber blkno;
8904 : : XLogRedoAction action;
8905 : :
378 andres@anarazel.de 8906 [ - + ]: 7594 : Assert((xlrec->flags & VISIBILITYMAP_XLOG_VALID_BITS) == xlrec->flags);
8907 : :
648 rhaas@postgresql.org 8908 : 7594 : XLogRecGetBlockTag(record, 1, &rlocator, NULL, &blkno);
8909 : :
8910 : : /*
8911 : : * If there are any Hot Standby transactions running that have an xmin
8912 : : * horizon old enough that this page isn't all-visible for them, they
8913 : : * might incorrectly decide that an index-only scan can skip a heap fetch.
8914 : : *
8915 : : * NB: It might be better to throw some kind of "soft" conflict here that
8916 : : * forces any index-only scan that is in flight to perform heap fetches,
8917 : : * rather than killing the transaction outright.
8918 : : */
4336 simon@2ndQuadrant.co 8919 [ + + ]: 7594 : if (InHotStandby)
514 pg@bowt.ie 8920 : 7420 : ResolveRecoveryConflictWithSnapshot(xlrec->snapshotConflictHorizon,
373 andres@anarazel.de 8921 : 7420 : xlrec->flags & VISIBILITYMAP_XLOG_CATALOG_REL,
8922 : : rlocator);
8923 : :
8924 : : /*
8925 : : * Read the heap page, if it still exists. If the heap file has dropped or
8926 : : * truncated later in recovery, we don't need to update the page, but we'd
8927 : : * better still update the visibility map.
8928 : : */
3433 heikki.linnakangas@i 8929 : 7594 : action = XLogReadBufferForRedo(record, 1, &buffer);
3532 8930 [ + + ]: 7594 : if (action == BLK_NEEDS_REDO)
8931 : : {
8932 : : /*
8933 : : * We don't bump the LSN of the heap page when setting the visibility
8934 : : * map bit (unless checksums or wal_hint_bits is enabled, in which
8935 : : * case we must). This exposes us to torn page hazards, but since
8936 : : * we're not inspecting the existing page contents in any way, we
8937 : : * don't care.
8938 : : */
2916 kgrittn@postgresql.o 8939 : 5744 : page = BufferGetPage(buffer);
8940 : :
2959 rhaas@postgresql.org 8941 : 5744 : PageSetAllVisible(page);
8942 : :
521 jdavis@postgresql.or 8943 [ + - + + ]: 5744 : if (XLogHintBitIsNeeded())
8944 : 5654 : PageSetLSN(page, lsn);
8945 : :
3532 heikki.linnakangas@i 8946 : 5744 : MarkBufferDirty(buffer);
8947 : : }
8948 : : else if (action == BLK_RESTORED)
8949 : : {
8950 : : /*
8951 : : * If heap block was backed up, we already restored it and there's
8952 : : * nothing more to do. (This can only happen with checksums or
8953 : : * wal_log_hints enabled.)
8954 : : */
8955 : : }
8956 : :
8957 [ + + ]: 7594 : if (BufferIsValid(buffer))
8958 : : {
2069 alvherre@alvh.no-ip. 8959 : 7549 : Size space = PageGetFreeSpace(BufferGetPage(buffer));
8960 : :
3532 heikki.linnakangas@i 8961 : 7549 : UnlockReleaseBuffer(buffer);
8962 : :
8963 : : /*
8964 : : * Since FSM is not WAL-logged and only updated heuristically, it
8965 : : * easily becomes stale in standbys. If the standby is later promoted
8966 : : * and runs VACUUM, it will skip updating individual free space
8967 : : * figures for pages that became all-visible (or all-frozen, depending
8968 : : * on the vacuum mode,) which is troublesome when FreeSpaceMapVacuum
8969 : : * propagates too optimistic free space values to upper FSM layers;
8970 : : * later inserters try to use such pages only to find out that they
8971 : : * are unusable. This can cause long stalls when there are many such
8972 : : * pages.
8973 : : *
8974 : : * Forestall those problems by updating FSM's idea about a page that
8975 : : * is becoming all-visible or all-frozen.
8976 : : *
8977 : : * Do this regardless of a full-page image being applied, since the
8978 : : * FSM data is not in the page anyway.
8979 : : */
2069 alvherre@alvh.no-ip. 8980 [ + - ]: 7549 : if (xlrec->flags & VISIBILITYMAP_VALID_BITS)
648 rhaas@postgresql.org 8981 : 7549 : XLogRecordPageWithFreeSpace(rlocator, blkno, space);
8982 : : }
8983 : :
8984 : : /*
8985 : : * Even if we skipped the heap page update due to the LSN interlock, it's
8986 : : * still safe to update the visibility map. Any WAL record that clears
8987 : : * the visibility map bit does so before checking the page LSN, so any
8988 : : * bits that need to be cleared will still be cleared.
8989 : : */
3433 heikki.linnakangas@i 8990 [ + + ]: 7594 : if (XLogReadBufferForRedoExtended(record, 0, RBM_ZERO_ON_ERROR, false,
8991 : : &vmbuffer) == BLK_NEEDS_REDO)
8992 : : {
2916 kgrittn@postgresql.o 8993 : 7325 : Page vmpage = BufferGetPage(vmbuffer);
8994 : : Relation reln;
8995 : : uint8 vmbits;
8996 : :
8997 : : /* initialize the page if it was read as zeros */
3433 heikki.linnakangas@i 8998 [ - + ]: 7325 : if (PageIsNew(vmpage))
3433 heikki.linnakangas@i 8999 :UBC 0 : PageInit(vmpage, BLCKSZ, 0);
9000 : :
9001 : : /* remove VISIBILITYMAP_XLOG_* */
378 andres@anarazel.de 9002 :CBC 7325 : vmbits = xlrec->flags & VISIBILITYMAP_VALID_BITS;
9003 : :
9004 : : /*
9005 : : * XLogReadBufferForRedoExtended locked the buffer. But
9006 : : * visibilitymap_set will handle locking itself.
9007 : : */
3433 heikki.linnakangas@i 9008 : 7325 : LockBuffer(vmbuffer, BUFFER_LOCK_UNLOCK);
9009 : :
648 rhaas@postgresql.org 9010 : 7325 : reln = CreateFakeRelcacheEntry(rlocator);
3532 heikki.linnakangas@i 9011 : 7325 : visibilitymap_pin(reln, blkno, &vmbuffer);
9012 : :
520 jdavis@postgresql.or 9013 : 7325 : visibilitymap_set(reln, blkno, InvalidBuffer, lsn, vmbuffer,
9014 : : xlrec->snapshotConflictHorizon, vmbits);
9015 : :
4681 rhaas@postgresql.org 9016 : 7325 : ReleaseBuffer(vmbuffer);
9017 : 7325 : FreeFakeRelcacheEntry(reln);
9018 : : }
3433 heikki.linnakangas@i 9019 [ + - ]: 269 : else if (BufferIsValid(vmbuffer))
9020 : 269 : UnlockReleaseBuffer(vmbuffer);
4681 rhaas@postgresql.org 9021 : 7594 : }
9022 : :
9023 : : /*
9024 : : * Given an "infobits" field from an XLog record, set the correct bits in the
9025 : : * given infomask and infomask2 for the tuple touched by the record.
9026 : : *
9027 : : * (This is the reverse of compute_infobits).
9028 : : */
9029 : : static void
4099 alvherre@alvh.no-ip. 9030 : 481432 : fix_infomask_from_infobits(uint8 infobits, uint16 *infomask, uint16 *infomask2)
9031 : : {
9032 : 481432 : *infomask &= ~(HEAP_XMAX_IS_MULTI | HEAP_XMAX_LOCK_ONLY |
9033 : : HEAP_XMAX_KEYSHR_LOCK | HEAP_XMAX_EXCL_LOCK);
9034 : 481432 : *infomask2 &= ~HEAP_KEYS_UPDATED;
9035 : :
9036 [ + + ]: 481432 : if (infobits & XLHL_XMAX_IS_MULTI)
9037 : 2 : *infomask |= HEAP_XMAX_IS_MULTI;
9038 [ + + ]: 481432 : if (infobits & XLHL_XMAX_LOCK_ONLY)
9039 : 54763 : *infomask |= HEAP_XMAX_LOCK_ONLY;
9040 [ + + ]: 481432 : if (infobits & XLHL_XMAX_EXCL_LOCK)
9041 : 54381 : *infomask |= HEAP_XMAX_EXCL_LOCK;
9042 : : /* note HEAP_XMAX_SHR_LOCK isn't considered here */
9043 [ + + ]: 481432 : if (infobits & XLHL_XMAX_KEYSHR_LOCK)
9044 : 393 : *infomask |= HEAP_XMAX_KEYSHR_LOCK;
9045 : :
9046 [ + + ]: 481432 : if (infobits & XLHL_KEYS_UPDATED)
9047 : 336013 : *infomask2 |= HEAP_KEYS_UPDATED;
9048 : 481432 : }
9049 : :
9050 : : static void
3433 heikki.linnakangas@i 9051 : 334896 : heap_xlog_delete(XLogReaderState *record)
9052 : : {
9053 : 334896 : XLogRecPtr lsn = record->EndRecPtr;
8424 bruce@momjian.us 9054 : 334896 : xl_heap_delete *xlrec = (xl_heap_delete *) XLogRecGetData(record);
9055 : : Buffer buffer;
9056 : : Page page;
9057 : 334896 : ItemId lp = NULL;
9058 : : HeapTupleHeader htup;
9059 : : BlockNumber blkno;
9060 : : RelFileLocator target_locator;
9061 : : ItemPointerData target_tid;
9062 : :
648 rhaas@postgresql.org 9063 : 334896 : XLogRecGetBlockTag(record, 0, &target_locator, NULL, &blkno);
3433 heikki.linnakangas@i 9064 : 334896 : ItemPointerSetBlockNumber(&target_tid, blkno);
9065 : 334896 : ItemPointerSetOffsetNumber(&target_tid, xlrec->offnum);
9066 : :
9067 : : /*
9068 : : * The visibility map may need to be fixed even if the heap page is
9069 : : * already up-to-date.
9070 : : */
3264 andres@anarazel.de 9071 [ + + ]: 334896 : if (xlrec->flags & XLH_DELETE_ALL_VISIBLE_CLEARED)
9072 : : {
648 rhaas@postgresql.org 9073 : 21 : Relation reln = CreateFakeRelcacheEntry(target_locator);
4681 9074 : 21 : Buffer vmbuffer = InvalidBuffer;
9075 : :
9076 : 21 : visibilitymap_pin(reln, blkno, &vmbuffer);
2827 andres@anarazel.de 9077 : 21 : visibilitymap_clear(reln, blkno, vmbuffer, VISIBILITYMAP_VALID_BITS);
4681 rhaas@postgresql.org 9078 : 21 : ReleaseBuffer(vmbuffer);
5611 heikki.linnakangas@i 9079 : 21 : FreeFakeRelcacheEntry(reln);
9080 : : }
9081 : :
3433 9082 [ + + ]: 334896 : if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
9083 : : {
2916 kgrittn@postgresql.o 9084 : 334751 : page = BufferGetPage(buffer);
9085 : :
3433 heikki.linnakangas@i 9086 [ + - ]: 334751 : if (PageGetMaxOffsetNumber(page) >= xlrec->offnum)
9087 : 334751 : lp = PageGetItemId(page, xlrec->offnum);
9088 : :
9089 [ + - - + ]: 334751 : if (PageGetMaxOffsetNumber(page) < xlrec->offnum || !ItemIdIsNormal(lp))
3069 andres@anarazel.de 9090 [ # # ]:UBC 0 : elog(PANIC, "invalid lp");
9091 : :
3532 heikki.linnakangas@i 9092 :CBC 334751 : htup = (HeapTupleHeader) PageGetItem(page, lp);
9093 : :
9094 : 334751 : htup->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
9095 : 334751 : htup->t_infomask2 &= ~HEAP_KEYS_UPDATED;
9096 : 334751 : HeapTupleHeaderClearHotUpdated(htup);
9097 : 334751 : fix_infomask_from_infobits(xlrec->infobits_set,
9098 : : &htup->t_infomask, &htup->t_infomask2);
3264 andres@anarazel.de 9099 [ + - ]: 334751 : if (!(xlrec->flags & XLH_DELETE_IS_SUPER))
9100 : 334751 : HeapTupleHeaderSetXmax(htup, xlrec->xmax);
9101 : : else
3264 andres@anarazel.de 9102 :UBC 0 : HeapTupleHeaderSetXmin(htup, InvalidTransactionId);
3532 heikki.linnakangas@i 9103 [ - + ]:CBC 334751 : HeapTupleHeaderSetCmax(htup, FirstCommandId, false);
9104 : :
9105 : : /* Mark the page as a candidate for pruning */
3433 9106 [ - + + + : 334751 : PageSetPrunable(page, XLogRecGetXid(record));
+ + ]
9107 : :
3264 andres@anarazel.de 9108 [ + + ]: 334751 : if (xlrec->flags & XLH_DELETE_ALL_VISIBLE_CLEARED)
3532 heikki.linnakangas@i 9109 : 17 : PageClearAllVisible(page);
9110 : :
9111 : : /* Make sure t_ctid is set correctly */
2199 andres@anarazel.de 9112 [ + + ]: 334751 : if (xlrec->flags & XLH_DELETE_IS_PARTITION_MOVE)
9113 : 132 : HeapTupleHeaderSetMovedPartitions(htup);
9114 : : else
9115 : 334619 : htup->t_ctid = target_tid;
3532 heikki.linnakangas@i 9116 : 334751 : PageSetLSN(page, lsn);
9117 : 334751 : MarkBufferDirty(buffer);
9118 : : }
9119 [ + - ]: 334896 : if (BufferIsValid(buffer))
9120 : 334896 : UnlockReleaseBuffer(buffer);
8685 vadim4o@yahoo.com 9121 : 334896 : }
9122 : :
9123 : : static void
3433 heikki.linnakangas@i 9124 : 1423598 : heap_xlog_insert(XLogReaderState *record)
9125 : : {
9126 : 1423598 : XLogRecPtr lsn = record->EndRecPtr;
8424 bruce@momjian.us 9127 : 1423598 : xl_heap_insert *xlrec = (xl_heap_insert *) XLogRecGetData(record);
9128 : : Buffer buffer;
9129 : : Page page;
9130 : : union
9131 : : {
9132 : : HeapTupleHeaderData hdr;
9133 : : char data[MaxHeapTupleSize];
9134 : : } tbuf;
9135 : : HeapTupleHeader htup;
9136 : : xl_heap_header xlhdr;
9137 : : uint32 newlen;
3532 heikki.linnakangas@i 9138 : 1423598 : Size freespace = 0;
9139 : : RelFileLocator target_locator;
9140 : : BlockNumber blkno;
9141 : : ItemPointerData target_tid;
9142 : : XLogRedoAction action;
9143 : :
648 rhaas@postgresql.org 9144 : 1423598 : XLogRecGetBlockTag(record, 0, &target_locator, NULL, &blkno);
3433 heikki.linnakangas@i 9145 : 1423598 : ItemPointerSetBlockNumber(&target_tid, blkno);
9146 : 1423598 : ItemPointerSetOffsetNumber(&target_tid, xlrec->offnum);
9147 : :
9148 : : /*
9149 : : * The visibility map may need to be fixed even if the heap page is
9150 : : * already up-to-date.
9151 : : */
3264 andres@anarazel.de 9152 [ + + ]: 1423598 : if (xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED)
9153 : : {
648 rhaas@postgresql.org 9154 : 2307 : Relation reln = CreateFakeRelcacheEntry(target_locator);
4681 9155 : 2307 : Buffer vmbuffer = InvalidBuffer;
9156 : :
9157 : 2307 : visibilitymap_pin(reln, blkno, &vmbuffer);
2827 andres@anarazel.de 9158 : 2307 : visibilitymap_clear(reln, blkno, vmbuffer, VISIBILITYMAP_VALID_BITS);
4681 rhaas@postgresql.org 9159 : 2307 : ReleaseBuffer(vmbuffer);
5611 heikki.linnakangas@i 9160 : 2307 : FreeFakeRelcacheEntry(reln);
9161 : : }
9162 : :
9163 : : /*
9164 : : * If we inserted the first and only tuple on the page, re-initialize the
9165 : : * page from scratch.
9166 : : */
3433 9167 [ + + ]: 1423598 : if (XLogRecGetInfo(record) & XLOG_HEAP_INIT_PAGE)
9168 : : {
9169 : 18114 : buffer = XLogInitBufferForRedo(record, 0);
2916 kgrittn@postgresql.o 9170 : 18114 : page = BufferGetPage(buffer);
6591 tgl@sss.pgh.pa.us 9171 : 18114 : PageInit(page, BufferGetPageSize(buffer), 0);
3532 heikki.linnakangas@i 9172 : 18114 : action = BLK_NEEDS_REDO;
9173 : : }
9174 : : else
3433 9175 : 1405484 : action = XLogReadBufferForRedo(record, 0, &buffer);
3532 9176 [ + + ]: 1423598 : if (action == BLK_NEEDS_REDO)
9177 : : {
9178 : : Size datalen;
9179 : : char *data;
9180 : :
2916 kgrittn@postgresql.o 9181 : 1422816 : page = BufferGetPage(buffer);
9182 : :
3433 heikki.linnakangas@i 9183 [ - + ]: 1422816 : if (PageGetMaxOffsetNumber(page) + 1 < xlrec->offnum)
3069 andres@anarazel.de 9184 [ # # ]:UBC 0 : elog(PANIC, "invalid max offset number");
9185 : :
3433 heikki.linnakangas@i 9186 :CBC 1422816 : data = XLogRecGetBlockData(record, 0, &datalen);
9187 : :
9188 : 1422816 : newlen = datalen - SizeOfHeapHeader;
9189 [ + - - + ]: 1422816 : Assert(datalen > SizeOfHeapHeader && newlen <= MaxHeapTupleSize);
9190 : 1422816 : memcpy((char *) &xlhdr, data, SizeOfHeapHeader);
9191 : 1422816 : data += SizeOfHeapHeader;
9192 : :
3532 9193 : 1422816 : htup = &tbuf.hdr;
3340 tgl@sss.pgh.pa.us 9194 [ + - - + : 1422816 : MemSet((char *) htup, 0, SizeofHeapTupleHeader);
- - - - -
- ]
9195 : : /* PG73FORMAT: get bitmap [+ padding] [+ oid] + data */
9196 : 1422816 : memcpy((char *) htup + SizeofHeapTupleHeader,
9197 : : data,
9198 : : newlen);
9199 : 1422816 : newlen += SizeofHeapTupleHeader;
3532 heikki.linnakangas@i 9200 : 1422816 : htup->t_infomask2 = xlhdr.t_infomask2;
9201 : 1422816 : htup->t_infomask = xlhdr.t_infomask;
9202 : 1422816 : htup->t_hoff = xlhdr.t_hoff;
3433 9203 : 1422816 : HeapTupleHeaderSetXmin(htup, XLogRecGetXid(record));
3532 9204 [ - + ]: 1422816 : HeapTupleHeaderSetCmin(htup, FirstCommandId);
3433 9205 : 1422816 : htup->t_ctid = target_tid;
9206 : :
9207 [ - + ]: 1422816 : if (PageAddItem(page, (Item) htup, newlen, xlrec->offnum,
9208 : : true, true) == InvalidOffsetNumber)
3069 andres@anarazel.de 9209 [ # # ]:UBC 0 : elog(PANIC, "failed to add tuple");
9210 : :
3532 heikki.linnakangas@i 9211 :CBC 1422816 : freespace = PageGetHeapFreeSpace(page); /* needed to update FSM below */
9212 : :
9213 : 1422816 : PageSetLSN(page, lsn);
9214 : :
3264 andres@anarazel.de 9215 [ + + ]: 1422816 : if (xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED)
3532 heikki.linnakangas@i 9216 : 1992 : PageClearAllVisible(page);
9217 : :
9218 : : /* XLH_INSERT_ALL_FROZEN_SET implies that all tuples are visible */
1183 tomas.vondra@postgre 9219 [ - + ]: 1422816 : if (xlrec->flags & XLH_INSERT_ALL_FROZEN_SET)
1183 tomas.vondra@postgre 9220 :UBC 0 : PageSetAllVisible(page);
9221 : :
3532 heikki.linnakangas@i 9222 :CBC 1422816 : MarkBufferDirty(buffer);
9223 : : }
9224 [ + - ]: 1423598 : if (BufferIsValid(buffer))
9225 : 1423598 : UnlockReleaseBuffer(buffer);
9226 : :
9227 : : /*
9228 : : * If the page is running low on free space, update the FSM as well.
9229 : : * Arbitrarily, our definition of "low" is less than 20%. We can't do much
9230 : : * better than that without knowing the fill-factor for the table.
9231 : : *
9232 : : * XXX: Don't do this if the page was restored from full page image. We
9233 : : * don't bother to update the FSM in that case, it doesn't need to be
9234 : : * totally accurate anyway.
9235 : : */
9236 [ + + + + ]: 1423598 : if (action == BLK_NEEDS_REDO && freespace < BLCKSZ / 5)
648 rhaas@postgresql.org 9237 : 311033 : XLogRecordPageWithFreeSpace(target_locator, blkno, freespace);
8620 vadim4o@yahoo.com 9238 : 1423598 : }
9239 : :
9240 : : /*
9241 : : * Handles MULTI_INSERT record type.
9242 : : */
9243 : : static void
3433 heikki.linnakangas@i 9244 : 52736 : heap_xlog_multi_insert(XLogReaderState *record)
9245 : : {
9246 : 52736 : XLogRecPtr lsn = record->EndRecPtr;
9247 : : xl_heap_multi_insert *xlrec;
9248 : : RelFileLocator rlocator;
9249 : : BlockNumber blkno;
9250 : : Buffer buffer;
9251 : : Page page;
9252 : : union
9253 : : {
9254 : : HeapTupleHeaderData hdr;
9255 : : char data[MaxHeapTupleSize];
9256 : : } tbuf;
9257 : : HeapTupleHeader htup;
9258 : : uint32 newlen;
3532 9259 : 52736 : Size freespace = 0;
9260 : : int i;
3433 9261 : 52736 : bool isinit = (XLogRecGetInfo(record) & XLOG_HEAP_INIT_PAGE) != 0;
9262 : : XLogRedoAction action;
9263 : :
9264 : : /*
9265 : : * Insertion doesn't overwrite MVCC data, so no conflict processing is
9266 : : * required.
9267 : : */
9268 : 52736 : xlrec = (xl_heap_multi_insert *) XLogRecGetData(record);
9269 : :
648 rhaas@postgresql.org 9270 : 52736 : XLogRecGetBlockTag(record, 0, &rlocator, NULL, &blkno);
9271 : :
9272 : : /* check that the mutually exclusive flags are not both set */
1068 tgl@sss.pgh.pa.us 9273 [ + + - + ]: 52736 : Assert(!((xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED) &&
9274 : : (xlrec->flags & XLH_INSERT_ALL_FROZEN_SET)));
9275 : :
9276 : : /*
9277 : : * The visibility map may need to be fixed even if the heap page is
9278 : : * already up-to-date.
9279 : : */
3264 andres@anarazel.de 9280 [ + + ]: 52736 : if (xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED)
9281 : : {
648 rhaas@postgresql.org 9282 : 964 : Relation reln = CreateFakeRelcacheEntry(rlocator);
4540 heikki.linnakangas@i 9283 : 964 : Buffer vmbuffer = InvalidBuffer;
9284 : :
9285 : 964 : visibilitymap_pin(reln, blkno, &vmbuffer);
2827 andres@anarazel.de 9286 : 964 : visibilitymap_clear(reln, blkno, vmbuffer, VISIBILITYMAP_VALID_BITS);
4540 heikki.linnakangas@i 9287 : 964 : ReleaseBuffer(vmbuffer);
9288 : 964 : FreeFakeRelcacheEntry(reln);
9289 : : }
9290 : :
9291 [ + + ]: 52736 : if (isinit)
9292 : : {
3433 9293 : 1727 : buffer = XLogInitBufferForRedo(record, 0);
2916 kgrittn@postgresql.o 9294 : 1727 : page = BufferGetPage(buffer);
4540 heikki.linnakangas@i 9295 : 1727 : PageInit(page, BufferGetPageSize(buffer), 0);
3532 9296 : 1727 : action = BLK_NEEDS_REDO;
9297 : : }
9298 : : else
3433 9299 : 51009 : action = XLogReadBufferForRedo(record, 0, &buffer);
3532 9300 [ + + ]: 52736 : if (action == BLK_NEEDS_REDO)
9301 : : {
9302 : : char *tupdata;
9303 : : char *endptr;
9304 : : Size len;
9305 : :
9306 : : /* Tuples are stored as block data */
3433 9307 : 52130 : tupdata = XLogRecGetBlockData(record, 0, &len);
9308 : 52130 : endptr = tupdata + len;
9309 : :
2916 kgrittn@postgresql.o 9310 : 52130 : page = (Page) BufferGetPage(buffer);
9311 : :
3532 heikki.linnakangas@i 9312 [ + + ]: 252665 : for (i = 0; i < xlrec->ntuples; i++)
9313 : : {
9314 : : OffsetNumber offnum;
9315 : : xl_multi_insert_tuple *xlhdr;
9316 : :
9317 : : /*
9318 : : * If we're reinitializing the page, the tuples are stored in
9319 : : * order from FirstOffsetNumber. Otherwise there's an array of
9320 : : * offsets in the WAL record, and the tuples come after that.
9321 : : */
9322 [ + + ]: 200535 : if (isinit)
9323 : 99123 : offnum = FirstOffsetNumber + i;
9324 : : else
9325 : 101412 : offnum = xlrec->offsets[i];
9326 [ - + ]: 200535 : if (PageGetMaxOffsetNumber(page) + 1 < offnum)
3069 andres@anarazel.de 9327 [ # # ]:UBC 0 : elog(PANIC, "invalid max offset number");
9328 : :
3433 heikki.linnakangas@i 9329 :CBC 200535 : xlhdr = (xl_multi_insert_tuple *) SHORTALIGN(tupdata);
9330 : 200535 : tupdata = ((char *) xlhdr) + SizeOfMultiInsertTuple;
9331 : :
3532 9332 : 200535 : newlen = xlhdr->datalen;
9333 [ - + ]: 200535 : Assert(newlen <= MaxHeapTupleSize);
9334 : 200535 : htup = &tbuf.hdr;
3340 tgl@sss.pgh.pa.us 9335 [ + - - + : 200535 : MemSet((char *) htup, 0, SizeofHeapTupleHeader);
- - - - -
- ]
9336 : : /* PG73FORMAT: get bitmap [+ padding] [+ oid] + data */
9337 : 200535 : memcpy((char *) htup + SizeofHeapTupleHeader,
9338 : : (char *) tupdata,
9339 : : newlen);
3433 heikki.linnakangas@i 9340 : 200535 : tupdata += newlen;
9341 : :
3340 tgl@sss.pgh.pa.us 9342 : 200535 : newlen += SizeofHeapTupleHeader;
3532 heikki.linnakangas@i 9343 : 200535 : htup->t_infomask2 = xlhdr->t_infomask2;
9344 : 200535 : htup->t_infomask = xlhdr->t_infomask;
9345 : 200535 : htup->t_hoff = xlhdr->t_hoff;
3433 9346 : 200535 : HeapTupleHeaderSetXmin(htup, XLogRecGetXid(record));
3532 9347 [ - + ]: 200535 : HeapTupleHeaderSetCmin(htup, FirstCommandId);
9348 : 200535 : ItemPointerSetBlockNumber(&htup->t_ctid, blkno);
9349 : 200535 : ItemPointerSetOffsetNumber(&htup->t_ctid, offnum);
9350 : :
9351 : 200535 : offnum = PageAddItem(page, (Item) htup, newlen, offnum, true, true);
9352 [ - + ]: 200535 : if (offnum == InvalidOffsetNumber)
3069 andres@anarazel.de 9353 [ # # ]:UBC 0 : elog(PANIC, "failed to add tuple");
9354 : : }
3433 heikki.linnakangas@i 9355 [ - + ]:CBC 52130 : if (tupdata != endptr)
3069 andres@anarazel.de 9356 [ # # ]:UBC 0 : elog(PANIC, "total tuple length mismatch");
9357 : :
3532 heikki.linnakangas@i 9358 :CBC 52130 : freespace = PageGetHeapFreeSpace(page); /* needed to update FSM below */
9359 : :
9360 : 52130 : PageSetLSN(page, lsn);
9361 : :
3264 andres@anarazel.de 9362 [ + + ]: 52130 : if (xlrec->flags & XLH_INSERT_ALL_VISIBLE_CLEARED)
3532 heikki.linnakangas@i 9363 : 537 : PageClearAllVisible(page);
9364 : :
9365 : : /* XLH_INSERT_ALL_FROZEN_SET implies that all tuples are visible */
1176 tomas.vondra@postgre 9366 [ + + ]: 52130 : if (xlrec->flags & XLH_INSERT_ALL_FROZEN_SET)
9367 : 4 : PageSetAllVisible(page);
9368 : :
3532 heikki.linnakangas@i 9369 : 52130 : MarkBufferDirty(buffer);
9370 : : }
9371 [ + - ]: 52736 : if (BufferIsValid(buffer))
9372 : 52736 : UnlockReleaseBuffer(buffer);
9373 : :
9374 : : /*
9375 : : * If the page is running low on free space, update the FSM as well.
9376 : : * Arbitrarily, our definition of "low" is less than 20%. We can't do much
9377 : : * better than that without knowing the fill-factor for the table.
9378 : : *
9379 : : * XXX: Don't do this if the page was restored from full page image. We
9380 : : * don't bother to update the FSM in that case, it doesn't need to be
9381 : : * totally accurate anyway.
9382 : : */
9383 [ + + + + ]: 52736 : if (action == BLK_NEEDS_REDO && freespace < BLCKSZ / 5)
648 rhaas@postgresql.org 9384 : 14847 : XLogRecordPageWithFreeSpace(rlocator, blkno, freespace);
4540 heikki.linnakangas@i 9385 : 52736 : }
9386 : :
9387 : : /*
9388 : : * Handles UPDATE and HOT_UPDATE
9389 : : */
9390 : : static void
3433 9391 : 91961 : heap_xlog_update(XLogReaderState *record, bool hot_update)
9392 : : {
9393 : 91961 : XLogRecPtr lsn = record->EndRecPtr;
8424 bruce@momjian.us 9394 : 91961 : xl_heap_update *xlrec = (xl_heap_update *) XLogRecGetData(record);
9395 : : RelFileLocator rlocator;
9396 : : BlockNumber oldblk;
9397 : : BlockNumber newblk;
9398 : : ItemPointerData newtid;
9399 : : Buffer obuffer,
9400 : : nbuffer;
9401 : : Page page;
9402 : : OffsetNumber offnum;
9403 : 91961 : ItemId lp = NULL;
9404 : : HeapTupleData oldtup;
9405 : : HeapTupleHeader htup;
3686 heikki.linnakangas@i 9406 : 91961 : uint16 prefixlen = 0,
9407 : 91961 : suffixlen = 0;
9408 : : char *newp;
9409 : : union
9410 : : {
9411 : : HeapTupleHeaderData hdr;
9412 : : char data[MaxHeapTupleSize];
9413 : : } tbuf;
9414 : : xl_heap_header xlhdr;
9415 : : uint32 newlen;
3532 9416 : 91961 : Size freespace = 0;
9417 : : XLogRedoAction oldaction;
9418 : : XLogRedoAction newaction;
9419 : :
9420 : : /* initialize to keep the compiler quiet */
3686 9421 : 91961 : oldtup.t_data = NULL;
9422 : 91961 : oldtup.t_len = 0;
9423 : :
648 rhaas@postgresql.org 9424 : 91961 : XLogRecGetBlockTag(record, 0, &rlocator, NULL, &newblk);
734 tgl@sss.pgh.pa.us 9425 [ + + ]: 91961 : if (XLogRecGetBlockTagExtended(record, 1, NULL, NULL, &oldblk, NULL))
9426 : : {
9427 : : /* HOT updates are never done across pages */
3433 heikki.linnakangas@i 9428 [ - + ]: 53974 : Assert(!hot_update);
9429 : : }
9430 : : else
9431 : 37987 : oldblk = newblk;
9432 : :
9433 : 91961 : ItemPointerSet(&newtid, newblk, xlrec->new_offnum);
9434 : :
9435 : : /*
9436 : : * The visibility map may need to be fixed even if the heap page is
9437 : : * already up-to-date.
9438 : : */
3264 andres@anarazel.de 9439 [ + + ]: 91961 : if (xlrec->flags & XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED)
9440 : : {
648 rhaas@postgresql.org 9441 : 175 : Relation reln = CreateFakeRelcacheEntry(rlocator);
4681 9442 : 175 : Buffer vmbuffer = InvalidBuffer;
9443 : :
3532 heikki.linnakangas@i 9444 : 175 : visibilitymap_pin(reln, oldblk, &vmbuffer);
2827 andres@anarazel.de 9445 : 175 : visibilitymap_clear(reln, oldblk, vmbuffer, VISIBILITYMAP_VALID_BITS);
4681 rhaas@postgresql.org 9446 : 175 : ReleaseBuffer(vmbuffer);
5611 heikki.linnakangas@i 9447 : 175 : FreeFakeRelcacheEntry(reln);
9448 : : }
9449 : :
9450 : : /*
9451 : : * In normal operation, it is important to lock the two pages in
9452 : : * page-number order, to avoid possible deadlocks against other update
9453 : : * operations going the other way. However, during WAL replay there can
9454 : : * be no other update happening, so we don't need to worry about that. But
9455 : : * we *do* need to worry that we don't expose an inconsistent state to Hot
9456 : : * Standby queries --- so the original page can't be unlocked before we've
9457 : : * added the new tuple to the new page.
9458 : : */
9459 : :
9460 : : /* Deal with old tuple version */
3433 9461 : 91961 : oldaction = XLogReadBufferForRedo(record, (oldblk == newblk) ? 0 : 1,
9462 : : &obuffer);
3532 9463 [ + + ]: 91961 : if (oldaction == BLK_NEEDS_REDO)
9464 : : {
2916 kgrittn@postgresql.o 9465 : 91918 : page = BufferGetPage(obuffer);
3433 heikki.linnakangas@i 9466 : 91918 : offnum = xlrec->old_offnum;
3532 9467 [ + - ]: 91918 : if (PageGetMaxOffsetNumber(page) >= offnum)
9468 : 91918 : lp = PageGetItemId(page, offnum);
9469 : :
9470 [ + - - + ]: 91918 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
3069 andres@anarazel.de 9471 [ # # ]:UBC 0 : elog(PANIC, "invalid lp");
9472 : :
3532 heikki.linnakangas@i 9473 :CBC 91918 : htup = (HeapTupleHeader) PageGetItem(page, lp);
9474 : :
9475 : 91918 : oldtup.t_data = htup;
9476 : 91918 : oldtup.t_len = ItemIdGetLength(lp);
9477 : :
9478 : 91918 : htup->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
9479 : 91918 : htup->t_infomask2 &= ~HEAP_KEYS_UPDATED;
9480 [ + + ]: 91918 : if (hot_update)
9481 : 35355 : HeapTupleHeaderSetHotUpdated(htup);
9482 : : else
9483 : 56563 : HeapTupleHeaderClearHotUpdated(htup);
9484 : 91918 : fix_infomask_from_infobits(xlrec->old_infobits_set, &htup->t_infomask,
9485 : : &htup->t_infomask2);
9486 : 91918 : HeapTupleHeaderSetXmax(htup, xlrec->old_xmax);
9487 [ - + ]: 91918 : HeapTupleHeaderSetCmax(htup, FirstCommandId, false);
9488 : : /* Set forward chain link in t_ctid */
3433 9489 : 91918 : htup->t_ctid = newtid;
9490 : :
9491 : : /* Mark the page as a candidate for pruning */
9492 [ - + + + : 91918 : PageSetPrunable(page, XLogRecGetXid(record));
+ + ]
9493 : :
3264 andres@anarazel.de 9494 [ + + ]: 91918 : if (xlrec->flags & XLH_UPDATE_OLD_ALL_VISIBLE_CLEARED)
3532 heikki.linnakangas@i 9495 : 173 : PageClearAllVisible(page);
9496 : :
9497 : 91918 : PageSetLSN(page, lsn);
9498 : 91918 : MarkBufferDirty(obuffer);
9499 : : }
9500 : :
9501 : : /*
9502 : : * Read the page the new tuple goes into, if different from old.
9503 : : */
9504 [ + + ]: 91961 : if (oldblk == newblk)
9505 : : {
4171 tgl@sss.pgh.pa.us 9506 : 37987 : nbuffer = obuffer;
3532 heikki.linnakangas@i 9507 : 37987 : newaction = oldaction;
9508 : : }
3433 9509 [ + + ]: 53974 : else if (XLogRecGetInfo(record) & XLOG_HEAP_INIT_PAGE)
9510 : : {
9511 : 572 : nbuffer = XLogInitBufferForRedo(record, 0);
2916 kgrittn@postgresql.o 9512 : 572 : page = (Page) BufferGetPage(nbuffer);
3532 heikki.linnakangas@i 9513 : 572 : PageInit(page, BufferGetPageSize(nbuffer), 0);
9514 : 572 : newaction = BLK_NEEDS_REDO;
9515 : : }
9516 : : else
3433 9517 : 53402 : newaction = XLogReadBufferForRedo(record, 0, &nbuffer);
9518 : :
9519 : : /*
9520 : : * The visibility map may need to be fixed even if the heap page is
9521 : : * already up-to-date.
9522 : : */
3264 andres@anarazel.de 9523 [ + + ]: 91961 : if (xlrec->flags & XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED)
9524 : : {
648 rhaas@postgresql.org 9525 : 147 : Relation reln = CreateFakeRelcacheEntry(rlocator);
4681 9526 : 147 : Buffer vmbuffer = InvalidBuffer;
9527 : :
3532 heikki.linnakangas@i 9528 : 147 : visibilitymap_pin(reln, newblk, &vmbuffer);
2827 andres@anarazel.de 9529 : 147 : visibilitymap_clear(reln, newblk, vmbuffer, VISIBILITYMAP_VALID_BITS);
4681 rhaas@postgresql.org 9530 : 147 : ReleaseBuffer(vmbuffer);
5611 heikki.linnakangas@i 9531 : 147 : FreeFakeRelcacheEntry(reln);
9532 : : }
9533 : :
9534 : : /* Deal with new tuple */
3532 9535 [ + + ]: 91961 : if (newaction == BLK_NEEDS_REDO)
9536 : : {
9537 : : char *recdata;
9538 : : char *recdata_end;
9539 : : Size datalen;
9540 : : Size tuplen;
9541 : :
3433 9542 : 91852 : recdata = XLogRecGetBlockData(record, 0, &datalen);
9543 : 91852 : recdata_end = recdata + datalen;
9544 : :
2916 kgrittn@postgresql.o 9545 : 91852 : page = BufferGetPage(nbuffer);
9546 : :
3433 heikki.linnakangas@i 9547 : 91852 : offnum = xlrec->new_offnum;
3532 9548 [ - + ]: 91852 : if (PageGetMaxOffsetNumber(page) + 1 < offnum)
3069 andres@anarazel.de 9549 [ # # ]:UBC 0 : elog(PANIC, "invalid max offset number");
9550 : :
3264 andres@anarazel.de 9551 [ + + ]:CBC 91852 : if (xlrec->flags & XLH_UPDATE_PREFIX_FROM_OLD)
9552 : : {
3532 heikki.linnakangas@i 9553 [ - + ]: 14373 : Assert(newblk == oldblk);
9554 : 14373 : memcpy(&prefixlen, recdata, sizeof(uint16));
9555 : 14373 : recdata += sizeof(uint16);
9556 : : }
3264 andres@anarazel.de 9557 [ + + ]: 91852 : if (xlrec->flags & XLH_UPDATE_SUFFIX_FROM_OLD)
9558 : : {
3532 heikki.linnakangas@i 9559 [ - + ]: 32853 : Assert(newblk == oldblk);
9560 : 32853 : memcpy(&suffixlen, recdata, sizeof(uint16));
9561 : 32853 : recdata += sizeof(uint16);
9562 : : }
9563 : :
3433 9564 : 91852 : memcpy((char *) &xlhdr, recdata, SizeOfHeapHeader);
9565 : 91852 : recdata += SizeOfHeapHeader;
9566 : :
9567 : 91852 : tuplen = recdata_end - recdata;
9568 [ - + ]: 91852 : Assert(tuplen <= MaxHeapTupleSize);
9569 : :
3532 9570 : 91852 : htup = &tbuf.hdr;
3340 tgl@sss.pgh.pa.us 9571 [ + - - + : 91852 : MemSet((char *) htup, 0, SizeofHeapTupleHeader);
- - - - -
- ]
9572 : :
9573 : : /*
9574 : : * Reconstruct the new tuple using the prefix and/or suffix from the
9575 : : * old tuple, and the data stored in the WAL record.
9576 : : */
9577 : 91852 : newp = (char *) htup + SizeofHeapTupleHeader;
3532 heikki.linnakangas@i 9578 [ + + ]: 91852 : if (prefixlen > 0)
9579 : : {
9580 : : int len;
9581 : :
9582 : : /* copy bitmap [+ padding] [+ oid] from WAL record */
3340 tgl@sss.pgh.pa.us 9583 : 14373 : len = xlhdr.t_hoff - SizeofHeapTupleHeader;
3532 heikki.linnakangas@i 9584 : 14373 : memcpy(newp, recdata, len);
9585 : 14373 : recdata += len;
9586 : 14373 : newp += len;
9587 : :
9588 : : /* copy prefix from old tuple */
9589 : 14373 : memcpy(newp, (char *) oldtup.t_data + oldtup.t_data->t_hoff, prefixlen);
9590 : 14373 : newp += prefixlen;
9591 : :
9592 : : /* copy new tuple data from WAL record */
3340 tgl@sss.pgh.pa.us 9593 : 14373 : len = tuplen - (xlhdr.t_hoff - SizeofHeapTupleHeader);
3532 heikki.linnakangas@i 9594 : 14373 : memcpy(newp, recdata, len);
9595 : 14373 : recdata += len;
9596 : 14373 : newp += len;
9597 : : }
9598 : : else
9599 : : {
9600 : : /*
9601 : : * copy bitmap [+ padding] [+ oid] + data from record, all in one
9602 : : * go
9603 : : */
3433 9604 : 77479 : memcpy(newp, recdata, tuplen);
9605 : 77479 : recdata += tuplen;
9606 : 77479 : newp += tuplen;
9607 : : }
9608 [ - + ]: 91852 : Assert(recdata == recdata_end);
9609 : :
9610 : : /* copy suffix from old tuple */
3532 9611 [ + + ]: 91852 : if (suffixlen > 0)
9612 : 32853 : memcpy(newp, (char *) oldtup.t_data + oldtup.t_len - suffixlen, suffixlen);
9613 : :
3340 tgl@sss.pgh.pa.us 9614 : 91852 : newlen = SizeofHeapTupleHeader + tuplen + prefixlen + suffixlen;
3433 heikki.linnakangas@i 9615 : 91852 : htup->t_infomask2 = xlhdr.t_infomask2;
9616 : 91852 : htup->t_infomask = xlhdr.t_infomask;
9617 : 91852 : htup->t_hoff = xlhdr.t_hoff;
9618 : :
9619 : 91852 : HeapTupleHeaderSetXmin(htup, XLogRecGetXid(record));
3532 9620 [ - + ]: 91852 : HeapTupleHeaderSetCmin(htup, FirstCommandId);
9621 : 91852 : HeapTupleHeaderSetXmax(htup, xlrec->new_xmax);
9622 : : /* Make sure there is no forward chain link in t_ctid */
3433 9623 : 91852 : htup->t_ctid = newtid;
9624 : :
3532 9625 : 91852 : offnum = PageAddItem(page, (Item) htup, newlen, offnum, true, true);
9626 [ - + ]: 91852 : if (offnum == InvalidOffsetNumber)
3069 andres@anarazel.de 9627 [ # # ]:UBC 0 : elog(PANIC, "failed to add tuple");
9628 : :
3264 andres@anarazel.de 9629 [ + + ]:CBC 91852 : if (xlrec->flags & XLH_UPDATE_NEW_ALL_VISIBLE_CLEARED)
3532 heikki.linnakangas@i 9630 : 113 : PageClearAllVisible(page);
9631 : :
9632 : 91852 : freespace = PageGetHeapFreeSpace(page); /* needed to update FSM below */
9633 : :
9634 : 91852 : PageSetLSN(page, lsn);
9635 : 91852 : MarkBufferDirty(nbuffer);
9636 : : }
9637 : :
9638 [ + - + + ]: 91961 : if (BufferIsValid(nbuffer) && nbuffer != obuffer)
9639 : 53974 : UnlockReleaseBuffer(nbuffer);
9640 [ + - ]: 91961 : if (BufferIsValid(obuffer))
4171 tgl@sss.pgh.pa.us 9641 : 91961 : UnlockReleaseBuffer(obuffer);
9642 : :
9643 : : /*
9644 : : * If the new page is running low on free space, update the FSM as well.
9645 : : * Arbitrarily, our definition of "low" is less than 20%. We can't do much
9646 : : * better than that without knowing the fill-factor for the table.
9647 : : *
9648 : : * However, don't update the FSM on HOT updates, because after crash
9649 : : * recovery, either the old or the new tuple will certainly be dead and
9650 : : * prunable. After pruning, the page will have roughly as much free space
9651 : : * as it did before the update, assuming the new tuple is about the same
9652 : : * size as the old one.
9653 : : *
9654 : : * XXX: Don't do this if the page was restored from full page image. We
9655 : : * don't bother to update the FSM in that case, it doesn't need to be
9656 : : * totally accurate anyway.
9657 : : */
3532 heikki.linnakangas@i 9658 [ + + + + : 91961 : if (newaction == BLK_NEEDS_REDO && !hot_update && freespace < BLCKSZ / 5)
+ + ]
648 rhaas@postgresql.org 9659 : 11497 : XLogRecordPageWithFreeSpace(rlocator, newblk, freespace);
8577 vadim4o@yahoo.com 9660 : 91961 : }
9661 : :
9662 : : static void
3264 andres@anarazel.de 9663 : 61 : heap_xlog_confirm(XLogReaderState *record)
9664 : : {
9665 : 61 : XLogRecPtr lsn = record->EndRecPtr;
9666 : 61 : xl_heap_confirm *xlrec = (xl_heap_confirm *) XLogRecGetData(record);
9667 : : Buffer buffer;
9668 : : Page page;
9669 : : OffsetNumber offnum;
9670 : 61 : ItemId lp = NULL;
9671 : : HeapTupleHeader htup;
9672 : :
9673 [ + - ]: 61 : if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
9674 : : {
2916 kgrittn@postgresql.o 9675 : 61 : page = BufferGetPage(buffer);
9676 : :
3264 andres@anarazel.de 9677 : 61 : offnum = xlrec->offnum;
9678 [ + - ]: 61 : if (PageGetMaxOffsetNumber(page) >= offnum)
9679 : 61 : lp = PageGetItemId(page, offnum);
9680 : :
9681 [ + - - + ]: 61 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
3069 andres@anarazel.de 9682 [ # # ]:UBC 0 : elog(PANIC, "invalid lp");
9683 : :
3264 andres@anarazel.de 9684 :CBC 61 : htup = (HeapTupleHeader) PageGetItem(page, lp);
9685 : :
9686 : : /*
9687 : : * Confirm tuple as actually inserted
9688 : : */
9689 : 61 : ItemPointerSet(&htup->t_ctid, BufferGetBlockNumber(buffer), offnum);
9690 : :
9691 : 61 : PageSetLSN(page, lsn);
9692 : 61 : MarkBufferDirty(buffer);
9693 : : }
9694 [ + - ]: 61 : if (BufferIsValid(buffer))
9695 : 61 : UnlockReleaseBuffer(buffer);
9696 : 61 : }
9697 : :
9698 : : static void
3433 heikki.linnakangas@i 9699 : 54845 : heap_xlog_lock(XLogReaderState *record)
9700 : : {
9701 : 54845 : XLogRecPtr lsn = record->EndRecPtr;
6926 tgl@sss.pgh.pa.us 9702 : 54845 : xl_heap_lock *xlrec = (xl_heap_lock *) XLogRecGetData(record);
9703 : : Buffer buffer;
9704 : : Page page;
9705 : : OffsetNumber offnum;
9706 : 54845 : ItemId lp = NULL;
9707 : : HeapTupleHeader htup;
9708 : :
9709 : : /*
9710 : : * The visibility map may need to be fixed even if the heap page is
9711 : : * already up-to-date.
9712 : : */
2827 andres@anarazel.de 9713 [ + + ]: 54845 : if (xlrec->flags & XLH_LOCK_ALL_FROZEN_CLEARED)
9714 : : {
9715 : : RelFileLocator rlocator;
9716 : 34 : Buffer vmbuffer = InvalidBuffer;
9717 : : BlockNumber block;
9718 : : Relation reln;
9719 : :
648 rhaas@postgresql.org 9720 : 34 : XLogRecGetBlockTag(record, 0, &rlocator, NULL, &block);
9721 : 34 : reln = CreateFakeRelcacheEntry(rlocator);
9722 : :
2827 andres@anarazel.de 9723 : 34 : visibilitymap_pin(reln, block, &vmbuffer);
9724 : 34 : visibilitymap_clear(reln, block, vmbuffer, VISIBILITYMAP_ALL_FROZEN);
9725 : :
9726 : 34 : ReleaseBuffer(vmbuffer);
9727 : 34 : FreeFakeRelcacheEntry(reln);
9728 : : }
9729 : :
3433 heikki.linnakangas@i 9730 [ + + ]: 54845 : if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
9731 : : {
2916 kgrittn@postgresql.o 9732 : 54763 : page = (Page) BufferGetPage(buffer);
9733 : :
3433 heikki.linnakangas@i 9734 : 54763 : offnum = xlrec->offnum;
3532 9735 [ + - ]: 54763 : if (PageGetMaxOffsetNumber(page) >= offnum)
9736 : 54763 : lp = PageGetItemId(page, offnum);
9737 : :
9738 [ + - - + ]: 54763 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
3069 andres@anarazel.de 9739 [ # # ]:UBC 0 : elog(PANIC, "invalid lp");
9740 : :
3532 heikki.linnakangas@i 9741 :CBC 54763 : htup = (HeapTupleHeader) PageGetItem(page, lp);
9742 : :
2830 andres@anarazel.de 9743 : 54763 : htup->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
9744 : 54763 : htup->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3532 heikki.linnakangas@i 9745 : 54763 : fix_infomask_from_infobits(xlrec->infobits_set, &htup->t_infomask,
9746 : : &htup->t_infomask2);
9747 : :
9748 : : /*
9749 : : * Clear relevant update flags, but only if the modified infomask says
9750 : : * there's no update.
9751 : : */
9752 [ - + - - ]: 54763 : if (HEAP_XMAX_IS_LOCKED_ONLY(htup->t_infomask))
9753 : : {
9754 : 54763 : HeapTupleHeaderClearHotUpdated(htup);
9755 : : /* Make sure there is no forward chain link in t_ctid */
3433 9756 : 54763 : ItemPointerSet(&htup->t_ctid,
9757 : : BufferGetBlockNumber(buffer),
9758 : : offnum);
9759 : : }
369 pg@bowt.ie 9760 : 54763 : HeapTupleHeaderSetXmax(htup, xlrec->xmax);
3532 heikki.linnakangas@i 9761 [ - + ]: 54763 : HeapTupleHeaderSetCmax(htup, FirstCommandId, false);
9762 : 54763 : PageSetLSN(page, lsn);
9763 : 54763 : MarkBufferDirty(buffer);
9764 : : }
9765 [ + - ]: 54845 : if (BufferIsValid(buffer))
9766 : 54845 : UnlockReleaseBuffer(buffer);
6926 tgl@sss.pgh.pa.us 9767 : 54845 : }
9768 : :
9769 : : static void
3433 heikki.linnakangas@i 9770 :UBC 0 : heap_xlog_lock_updated(XLogReaderState *record)
9771 : : {
9772 : 0 : XLogRecPtr lsn = record->EndRecPtr;
9773 : : xl_heap_lock_updated *xlrec;
9774 : : Buffer buffer;
9775 : : Page page;
9776 : : OffsetNumber offnum;
4099 alvherre@alvh.no-ip. 9777 : 0 : ItemId lp = NULL;
9778 : : HeapTupleHeader htup;
9779 : :
3433 heikki.linnakangas@i 9780 : 0 : xlrec = (xl_heap_lock_updated *) XLogRecGetData(record);
9781 : :
9782 : : /*
9783 : : * The visibility map may need to be fixed even if the heap page is
9784 : : * already up-to-date.
9785 : : */
2827 andres@anarazel.de 9786 [ # # ]: 0 : if (xlrec->flags & XLH_LOCK_ALL_FROZEN_CLEARED)
9787 : : {
9788 : : RelFileLocator rlocator;
9789 : 0 : Buffer vmbuffer = InvalidBuffer;
9790 : : BlockNumber block;
9791 : : Relation reln;
9792 : :
648 rhaas@postgresql.org 9793 : 0 : XLogRecGetBlockTag(record, 0, &rlocator, NULL, &block);
9794 : 0 : reln = CreateFakeRelcacheEntry(rlocator);
9795 : :
2827 andres@anarazel.de 9796 : 0 : visibilitymap_pin(reln, block, &vmbuffer);
9797 : 0 : visibilitymap_clear(reln, block, vmbuffer, VISIBILITYMAP_ALL_FROZEN);
9798 : :
9799 : 0 : ReleaseBuffer(vmbuffer);
9800 : 0 : FreeFakeRelcacheEntry(reln);
9801 : : }
9802 : :
3433 heikki.linnakangas@i 9803 [ # # ]: 0 : if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
9804 : : {
2916 kgrittn@postgresql.o 9805 : 0 : page = BufferGetPage(buffer);
9806 : :
3433 heikki.linnakangas@i 9807 : 0 : offnum = xlrec->offnum;
3532 9808 [ # # ]: 0 : if (PageGetMaxOffsetNumber(page) >= offnum)
9809 : 0 : lp = PageGetItemId(page, offnum);
9810 : :
9811 [ # # # # ]: 0 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
3069 andres@anarazel.de 9812 [ # # ]: 0 : elog(PANIC, "invalid lp");
9813 : :
3532 heikki.linnakangas@i 9814 : 0 : htup = (HeapTupleHeader) PageGetItem(page, lp);
9815 : :
2830 andres@anarazel.de 9816 : 0 : htup->t_infomask &= ~(HEAP_XMAX_BITS | HEAP_MOVED);
9817 : 0 : htup->t_infomask2 &= ~HEAP_KEYS_UPDATED;
3532 heikki.linnakangas@i 9818 : 0 : fix_infomask_from_infobits(xlrec->infobits_set, &htup->t_infomask,
9819 : : &htup->t_infomask2);
9820 : 0 : HeapTupleHeaderSetXmax(htup, xlrec->xmax);
9821 : :
9822 : 0 : PageSetLSN(page, lsn);
9823 : 0 : MarkBufferDirty(buffer);
9824 : : }
9825 [ # # ]: 0 : if (BufferIsValid(buffer))
9826 : 0 : UnlockReleaseBuffer(buffer);
4099 alvherre@alvh.no-ip. 9827 : 0 : }
9828 : :
9829 : : static void
3433 heikki.linnakangas@i 9830 :CBC 7205 : heap_xlog_inplace(XLogReaderState *record)
9831 : : {
9832 : 7205 : XLogRecPtr lsn = record->EndRecPtr;
6549 tgl@sss.pgh.pa.us 9833 : 7205 : xl_heap_inplace *xlrec = (xl_heap_inplace *) XLogRecGetData(record);
9834 : : Buffer buffer;
9835 : : Page page;
9836 : : OffsetNumber offnum;
9837 : 7205 : ItemId lp = NULL;
9838 : : HeapTupleHeader htup;
9839 : : uint32 oldlen;
9840 : : Size newlen;
9841 : :
3433 heikki.linnakangas@i 9842 [ + + ]: 7205 : if (XLogReadBufferForRedo(record, 0, &buffer) == BLK_NEEDS_REDO)
9843 : : {
9844 : 7144 : char *newtup = XLogRecGetBlockData(record, 0, &newlen);
9845 : :
2916 kgrittn@postgresql.o 9846 : 7144 : page = BufferGetPage(buffer);
9847 : :
3433 heikki.linnakangas@i 9848 : 7144 : offnum = xlrec->offnum;
3532 9849 [ + - ]: 7144 : if (PageGetMaxOffsetNumber(page) >= offnum)
9850 : 7144 : lp = PageGetItemId(page, offnum);
9851 : :
9852 [ + - - + ]: 7144 : if (PageGetMaxOffsetNumber(page) < offnum || !ItemIdIsNormal(lp))
3069 andres@anarazel.de 9853 [ # # ]:UBC 0 : elog(PANIC, "invalid lp");
9854 : :
3532 heikki.linnakangas@i 9855 :CBC 7144 : htup = (HeapTupleHeader) PageGetItem(page, lp);
9856 : :
9857 : 7144 : oldlen = ItemIdGetLength(lp) - htup->t_hoff;
9858 [ - + ]: 7144 : if (oldlen != newlen)
3069 andres@anarazel.de 9859 [ # # ]:UBC 0 : elog(PANIC, "wrong tuple length");
9860 : :
3433 heikki.linnakangas@i 9861 :CBC 7144 : memcpy((char *) htup + htup->t_hoff, newtup, newlen);
9862 : :
3532 9863 : 7144 : PageSetLSN(page, lsn);
9864 : 7144 : MarkBufferDirty(buffer);
9865 : : }
9866 [ + - ]: 7205 : if (BufferIsValid(buffer))
9867 : 7205 : UnlockReleaseBuffer(buffer);
6549 tgl@sss.pgh.pa.us 9868 : 7205 : }
9869 : :
9870 : : void
3433 heikki.linnakangas@i 9871 : 1912569 : heap_redo(XLogReaderState *record)
9872 : : {
9873 : 1912569 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
9874 : :
9875 : : /*
9876 : : * These operations don't overwrite MVCC data so no conflict processing is
9877 : : * required. The ones in heap2 rmgr do.
9878 : : */
9879 : :
6051 tgl@sss.pgh.pa.us 9880 [ + + + + : 1912569 : switch (info & XLOG_HEAP_OPMASK)
+ + + +
- ]
9881 : : {
9882 : 1423598 : case XLOG_HEAP_INSERT:
3433 heikki.linnakangas@i 9883 : 1423598 : heap_xlog_insert(record);
6051 tgl@sss.pgh.pa.us 9884 : 1423598 : break;
9885 : 334896 : case XLOG_HEAP_DELETE:
3433 heikki.linnakangas@i 9886 : 334896 : heap_xlog_delete(record);
6051 tgl@sss.pgh.pa.us 9887 : 334896 : break;
9888 : 56572 : case XLOG_HEAP_UPDATE:
3433 heikki.linnakangas@i 9889 : 56572 : heap_xlog_update(record, false);
6051 tgl@sss.pgh.pa.us 9890 : 56572 : break;
2199 peter_e@gmx.net 9891 :GBC 3 : case XLOG_HEAP_TRUNCATE:
9892 : :
9893 : : /*
9894 : : * TRUNCATE is a no-op because the actions are already logged as
9895 : : * SMGR WAL records. TRUNCATE WAL record only exists for logical
9896 : : * decoding.
9897 : : */
9898 : 3 : break;
6051 tgl@sss.pgh.pa.us 9899 :CBC 35389 : case XLOG_HEAP_HOT_UPDATE:
3433 heikki.linnakangas@i 9900 : 35389 : heap_xlog_update(record, true);
6051 tgl@sss.pgh.pa.us 9901 : 35389 : break;
3264 andres@anarazel.de 9902 : 61 : case XLOG_HEAP_CONFIRM:
9903 : 61 : heap_xlog_confirm(record);
9904 : 61 : break;
6051 tgl@sss.pgh.pa.us 9905 : 54845 : case XLOG_HEAP_LOCK:
3433 heikki.linnakangas@i 9906 : 54845 : heap_xlog_lock(record);
6051 tgl@sss.pgh.pa.us 9907 : 54845 : break;
9908 : 7205 : case XLOG_HEAP_INPLACE:
3433 heikki.linnakangas@i 9909 : 7205 : heap_xlog_inplace(record);
6051 tgl@sss.pgh.pa.us 9910 : 7205 : break;
6051 tgl@sss.pgh.pa.us 9911 :UBC 0 : default:
9912 [ # # ]: 0 : elog(PANIC, "heap_redo: unknown op code %u", info);
9913 : : }
8577 vadim4o@yahoo.com 9914 :CBC 1912569 : }
9915 : :
9916 : : void
3433 heikki.linnakangas@i 9917 : 74388 : heap2_redo(XLogReaderState *record)
9918 : : {
9919 : 74388 : uint8 info = XLogRecGetInfo(record) & ~XLR_INFO_MASK;
9920 : :
6051 tgl@sss.pgh.pa.us 9921 [ + + + - : 74388 : switch (info & XLOG_HEAP_OPMASK)
+ - - -
- ]
9922 : : {
20 heikki.linnakangas@i 9923 :GNC 13170 : case XLOG_HEAP2_PRUNE_ON_ACCESS:
9924 : : case XLOG_HEAP2_PRUNE_VACUUM_SCAN:
9925 : : case XLOG_HEAP2_PRUNE_VACUUM_CLEANUP:
9926 : 13170 : heap_xlog_prune_freeze(record);
3772 alvherre@alvh.no-ip. 9927 :CBC 13170 : break;
4681 rhaas@postgresql.org 9928 : 7594 : case XLOG_HEAP2_VISIBLE:
3433 heikki.linnakangas@i 9929 : 7594 : heap_xlog_visible(record);
4681 rhaas@postgresql.org 9930 : 7594 : break;
4540 heikki.linnakangas@i 9931 : 52736 : case XLOG_HEAP2_MULTI_INSERT:
3433 9932 : 52736 : heap_xlog_multi_insert(record);
4540 9933 : 52736 : break;
4099 alvherre@alvh.no-ip. 9934 :UBC 0 : case XLOG_HEAP2_LOCK_UPDATED:
3433 heikki.linnakangas@i 9935 : 0 : heap_xlog_lock_updated(record);
4099 alvherre@alvh.no-ip. 9936 : 0 : break;
3778 rhaas@postgresql.org 9937 :CBC 888 : case XLOG_HEAP2_NEW_CID:
9938 : :
9939 : : /*
9940 : : * Nothing to do on a real replay, only used during logical
9941 : : * decoding.
9942 : : */
9943 : 888 : break;
3695 rhaas@postgresql.org 9944 :UBC 0 : case XLOG_HEAP2_REWRITE:
3433 heikki.linnakangas@i 9945 : 0 : heap_xlog_logical_rewrite(record);
3695 rhaas@postgresql.org 9946 : 0 : break;
6051 tgl@sss.pgh.pa.us 9947 : 0 : default:
9948 [ # # ]: 0 : elog(PANIC, "heap2_redo: unknown op code %u", info);
9949 : : }
6370 tgl@sss.pgh.pa.us 9950 :CBC 74388 : }
9951 : :
9952 : : /*
9953 : : * Mask a heap page before performing consistency checks on it.
9954 : : */
9955 : : void
2622 rhaas@postgresql.org 9956 :UBC 0 : heap_mask(char *pagedata, BlockNumber blkno)
9957 : : {
9958 : 0 : Page page = (Page) pagedata;
9959 : : OffsetNumber off;
9960 : :
2396 9961 : 0 : mask_page_lsn_and_checksum(page);
9962 : :
2622 9963 : 0 : mask_page_hint_bits(page);
9964 : 0 : mask_unused_space(page);
9965 : :
9966 [ # # ]: 0 : for (off = 1; off <= PageGetMaxOffsetNumber(page); off++)
9967 : : {
9968 : 0 : ItemId iid = PageGetItemId(page, off);
9969 : : char *page_item;
9970 : :
9971 : 0 : page_item = (char *) (page + ItemIdGetOffset(iid));
9972 : :
9973 [ # # ]: 0 : if (ItemIdIsNormal(iid))
9974 : : {
9975 : 0 : HeapTupleHeader page_htup = (HeapTupleHeader) page_item;
9976 : :
9977 : : /*
9978 : : * If xmin of a tuple is not yet frozen, we should ignore
9979 : : * differences in hint bits, since they can be set without
9980 : : * emitting WAL.
9981 : : */
9982 [ # # ]: 0 : if (!HeapTupleHeaderXminFrozen(page_htup))
9983 : 0 : page_htup->t_infomask &= ~HEAP_XACT_MASK;
9984 : : else
9985 : : {
9986 : : /* Still we need to mask xmax hint bits. */
9987 : 0 : page_htup->t_infomask &= ~HEAP_XMAX_INVALID;
9988 : 0 : page_htup->t_infomask &= ~HEAP_XMAX_COMMITTED;
9989 : : }
9990 : :
9991 : : /*
9992 : : * During replay, we set Command Id to FirstCommandId. Hence, mask
9993 : : * it. See heap_xlog_insert() for details.
9994 : : */
9995 : 0 : page_htup->t_choice.t_heap.t_field3.t_cid = MASK_MARKER;
9996 : :
9997 : : /*
9998 : : * For a speculative tuple, heap_insert() does not set ctid in the
9999 : : * caller-passed heap tuple itself, leaving the ctid field to
10000 : : * contain a speculative token value - a per-backend monotonically
10001 : : * increasing identifier. Besides, it does not WAL-log ctid under
10002 : : * any circumstances.
10003 : : *
10004 : : * During redo, heap_xlog_insert() sets t_ctid to current block
10005 : : * number and self offset number. It doesn't care about any
10006 : : * speculative insertions on the primary. Hence, we set t_ctid to
10007 : : * current block number and self offset number to ignore any
10008 : : * inconsistency.
10009 : : */
10010 [ # # ]: 0 : if (HeapTupleHeaderIsSpeculative(page_htup))
10011 : 0 : ItemPointerSet(&page_htup->t_ctid, blkno, off);
10012 : :
10013 : : /*
10014 : : * NB: Not ignoring ctid changes due to the tuple having moved
10015 : : * (i.e. HeapTupleHeaderIndicatesMovedPartitions), because that's
10016 : : * important information that needs to be in-sync between primary
10017 : : * and standby, and thus is WAL logged.
10018 : : */
10019 : : }
10020 : :
10021 : : /*
10022 : : * Ignore any padding bytes after the tuple, when the length of the
10023 : : * item is not MAXALIGNed.
10024 : : */
10025 [ # # ]: 0 : if (ItemIdHasStorage(iid))
10026 : : {
10027 : 0 : int len = ItemIdGetLength(iid);
10028 : 0 : int padlen = MAXALIGN(len) - len;
10029 : :
10030 [ # # ]: 0 : if (padlen > 0)
10031 : 0 : memset(page_item + len, MASK_MARKER, padlen);
10032 : : }
10033 : : }
10034 : 0 : }
10035 : :
10036 : : /*
10037 : : * HeapCheckForSerializableConflictOut
10038 : : * We are reading a tuple. If it's not visible, there may be a
10039 : : * rw-conflict out with the inserter. Otherwise, if it is visible to us
10040 : : * but has been deleted, there may be a rw-conflict out with the deleter.
10041 : : *
10042 : : * We will determine the top level xid of the writing transaction with which
10043 : : * we may be in conflict, and ask CheckForSerializableConflictOut() to check
10044 : : * for overlap with our own transaction.
10045 : : *
10046 : : * This function should be called just about anywhere in heapam.c where a
10047 : : * tuple has been read. The caller must hold at least a shared lock on the
10048 : : * buffer, because this function might set hint bits on the tuple. There is
10049 : : * currently no known reason to call this function from an index AM.
10050 : : */
10051 : : void
1538 tmunro@postgresql.or 10052 :CBC 28152682 : HeapCheckForSerializableConflictOut(bool visible, Relation relation,
10053 : : HeapTuple tuple, Buffer buffer,
10054 : : Snapshot snapshot)
10055 : : {
10056 : : TransactionId xid;
10057 : : HTSV_Result htsvResult;
10058 : :
10059 [ + + ]: 28152682 : if (!CheckForSerializableConflictOutNeeded(relation, snapshot))
10060 : 28127335 : return;
10061 : :
10062 : : /*
10063 : : * Check to see whether the tuple has been written to by a concurrent
10064 : : * transaction, either to create it not visible to us, or to delete it
10065 : : * while it is visible to us. The "visible" bool indicates whether the
10066 : : * tuple is visible to us, while HeapTupleSatisfiesVacuum checks what else
10067 : : * is going on with it.
10068 : : *
10069 : : * In the event of a concurrently inserted tuple that also happens to have
10070 : : * been concurrently updated (by a separate transaction), the xmin of the
10071 : : * tuple will be used -- not the updater's xid.
10072 : : */
10073 : 25347 : htsvResult = HeapTupleSatisfiesVacuum(tuple, TransactionXmin, buffer);
10074 [ + + + + : 25347 : switch (htsvResult)
- ]
10075 : : {
10076 : 24545 : case HEAPTUPLE_LIVE:
10077 [ + + ]: 24545 : if (visible)
10078 : 24532 : return;
10079 [ + - ]: 13 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
10080 : 13 : break;
10081 : 352 : case HEAPTUPLE_RECENTLY_DEAD:
10082 : : case HEAPTUPLE_DELETE_IN_PROGRESS:
1403 pg@bowt.ie 10083 [ + + ]: 352 : if (visible)
10084 [ + - - + : 281 : xid = HeapTupleHeaderGetUpdateXid(tuple->t_data);
- - ]
10085 : : else
10086 [ + - ]: 71 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
10087 : :
10088 [ + + ]: 352 : if (TransactionIdPrecedes(xid, TransactionXmin))
10089 : : {
10090 : : /* This is like the HEAPTUPLE_DEAD case */
10091 [ - + ]: 62 : Assert(!visible);
10092 : 62 : return;
10093 : : }
1538 tmunro@postgresql.or 10094 : 290 : break;
10095 : 326 : case HEAPTUPLE_INSERT_IN_PROGRESS:
10096 [ + - ]: 326 : xid = HeapTupleHeaderGetXmin(tuple->t_data);
10097 : 326 : break;
10098 : 124 : case HEAPTUPLE_DEAD:
1403 pg@bowt.ie 10099 [ - + ]: 124 : Assert(!visible);
1538 tmunro@postgresql.or 10100 : 124 : return;
1538 tmunro@postgresql.or 10101 :UBC 0 : default:
10102 : :
10103 : : /*
10104 : : * The only way to get to this default clause is if a new value is
10105 : : * added to the enum type without adding it to this switch
10106 : : * statement. That's a bug, so elog.
10107 : : */
10108 [ # # ]: 0 : elog(ERROR, "unrecognized return value from HeapTupleSatisfiesVacuum: %u", htsvResult);
10109 : :
10110 : : /*
10111 : : * In spite of having all enum values covered and calling elog on
10112 : : * this default, some compilers think this is a code path which
10113 : : * allows xid to be used below without initialization. Silence
10114 : : * that warning.
10115 : : */
10116 : : xid = InvalidTransactionId;
10117 : : }
10118 : :
1538 tmunro@postgresql.or 10119 [ - + ]:CBC 629 : Assert(TransactionIdIsValid(xid));
10120 [ - + ]: 629 : Assert(TransactionIdFollowsOrEquals(xid, TransactionXmin));
10121 : :
10122 : : /*
10123 : : * Find top level xid. Bail out if xid is too early to be a conflict, or
10124 : : * if it's our own xid.
10125 : : */
10126 [ + + ]: 629 : if (TransactionIdEquals(xid, GetTopTransactionIdIfAny()))
10127 : 62 : return;
10128 : 567 : xid = SubTransGetTopmostTransaction(xid);
10129 [ - + ]: 567 : if (TransactionIdPrecedes(xid, TransactionXmin))
1538 tmunro@postgresql.or 10130 :UBC 0 : return;
10131 : :
1538 tmunro@postgresql.or 10132 :CBC 567 : CheckForSerializableConflictOut(relation, xid, snapshot);
10133 : : }
|
