Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * nbtree.c
4 : * Implementation of Lehman and Yao's btree management algorithm for
5 : * Postgres.
6 : *
7 : * NOTES
8 : * This file contains only the public interface routines.
9 : *
10 : *
11 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
12 : * Portions Copyright (c) 1994, Regents of the University of California
13 : *
14 : * IDENTIFICATION
15 : * src/backend/access/nbtree/nbtree.c
16 : *
17 : *-------------------------------------------------------------------------
18 : */
19 : #include "postgres.h"
20 :
21 : #include "access/nbtree.h"
22 : #include "access/relscan.h"
23 : #include "access/stratnum.h"
24 : #include "commands/progress.h"
25 : #include "commands/vacuum.h"
26 : #include "nodes/execnodes.h"
27 : #include "pgstat.h"
28 : #include "storage/bulk_write.h"
29 : #include "storage/condition_variable.h"
30 : #include "storage/indexfsm.h"
31 : #include "storage/ipc.h"
32 : #include "storage/lmgr.h"
33 : #include "storage/read_stream.h"
34 : #include "utils/datum.h"
35 : #include "utils/fmgrprotos.h"
36 : #include "utils/index_selfuncs.h"
37 : #include "utils/memutils.h"
38 :
39 :
40 : /*
41 : * BTPARALLEL_NOT_INITIALIZED indicates that the scan has not started.
42 : *
43 : * BTPARALLEL_NEED_PRIMSCAN indicates that some process must now seize the
44 : * scan to advance it via another call to _bt_first.
45 : *
46 : * BTPARALLEL_ADVANCING indicates that some process is advancing the scan to
47 : * a new page; others must wait.
48 : *
49 : * BTPARALLEL_IDLE indicates that no backend is currently advancing the scan
50 : * to a new page; some process can start doing that.
51 : *
52 : * BTPARALLEL_DONE indicates that the scan is complete (including error exit).
53 : */
54 : typedef enum
55 : {
56 : BTPARALLEL_NOT_INITIALIZED,
57 : BTPARALLEL_NEED_PRIMSCAN,
58 : BTPARALLEL_ADVANCING,
59 : BTPARALLEL_IDLE,
60 : BTPARALLEL_DONE,
61 : } BTPS_State;
62 :
63 : /*
64 : * BTParallelScanDescData contains btree specific shared information required
65 : * for parallel scan.
66 : */
67 : typedef struct BTParallelScanDescData
68 : {
69 : BlockNumber btps_nextScanPage; /* next page to be scanned */
70 : BlockNumber btps_lastCurrPage; /* page whose sibling link was copied into
71 : * btps_nextScanPage */
72 : BTPS_State btps_pageStatus; /* indicates whether next page is
73 : * available for scan. see above for
74 : * possible states of parallel scan. */
75 : LWLock btps_lock; /* protects shared parallel state */
76 : ConditionVariable btps_cv; /* used to synchronize parallel scan */
77 :
78 : /*
79 : * btps_arrElems is used when scans need to schedule another primitive
80 : * index scan with one or more SAOP arrays. Holds BTArrayKeyInfo.cur_elem
81 : * offsets for each = scan key associated with a ScalarArrayOp array.
82 : */
83 : int btps_arrElems[FLEXIBLE_ARRAY_MEMBER];
84 :
85 : /*
86 : * Additional space (at the end of the struct) is used when scans need to
87 : * schedule another primitive index scan with one or more skip arrays.
88 : * Holds a flattened datum representation for each = scan key associated
89 : * with a skip array.
90 : */
91 : } BTParallelScanDescData;
92 :
93 : typedef struct BTParallelScanDescData *BTParallelScanDesc;
94 :
95 :
96 : static void _bt_parallel_serialize_arrays(Relation rel, BTParallelScanDesc btscan,
97 : BTScanOpaque so);
98 : static void _bt_parallel_restore_arrays(Relation rel, BTParallelScanDesc btscan,
99 : BTScanOpaque so);
100 : static void btvacuumscan(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
101 : IndexBulkDeleteCallback callback, void *callback_state,
102 : BTCycleId cycleid);
103 : static BlockNumber btvacuumpage(BTVacState *vstate, Buffer buf);
104 : static BTVacuumPosting btreevacuumposting(BTVacState *vstate,
105 : IndexTuple posting,
106 : OffsetNumber updatedoffset,
107 : int *nremaining);
108 :
109 :
110 : /*
111 : * Btree handler function: return IndexAmRoutine with access method parameters
112 : * and callbacks.
113 : */
114 : Datum
115 3389988 : bthandler(PG_FUNCTION_ARGS)
116 : {
117 3389988 : IndexAmRoutine *amroutine = makeNode(IndexAmRoutine);
118 :
119 3389988 : amroutine->amstrategies = BTMaxStrategyNumber;
120 3389988 : amroutine->amsupport = BTNProcs;
121 3389988 : amroutine->amoptsprocnum = BTOPTIONS_PROC;
122 3389988 : amroutine->amcanorder = true;
123 3389988 : amroutine->amcanorderbyop = false;
124 3389988 : amroutine->amcanhash = false;
125 3389988 : amroutine->amconsistentequality = true;
126 3389988 : amroutine->amconsistentordering = true;
127 3389988 : amroutine->amcanbackward = true;
128 3389988 : amroutine->amcanunique = true;
129 3389988 : amroutine->amcanmulticol = true;
130 3389988 : amroutine->amoptionalkey = true;
131 3389988 : amroutine->amsearcharray = true;
132 3389988 : amroutine->amsearchnulls = true;
133 3389988 : amroutine->amstorage = false;
134 3389988 : amroutine->amclusterable = true;
135 3389988 : amroutine->ampredlocks = true;
136 3389988 : amroutine->amcanparallel = true;
137 3389988 : amroutine->amcanbuildparallel = true;
138 3389988 : amroutine->amcaninclude = true;
139 3389988 : amroutine->amusemaintenanceworkmem = false;
140 3389988 : amroutine->amsummarizing = false;
141 3389988 : amroutine->amparallelvacuumoptions =
142 : VACUUM_OPTION_PARALLEL_BULKDEL | VACUUM_OPTION_PARALLEL_COND_CLEANUP;
143 3389988 : amroutine->amkeytype = InvalidOid;
144 :
145 3389988 : amroutine->ambuild = btbuild;
146 3389988 : amroutine->ambuildempty = btbuildempty;
147 3389988 : amroutine->aminsert = btinsert;
148 3389988 : amroutine->aminsertcleanup = NULL;
149 3389988 : amroutine->ambulkdelete = btbulkdelete;
150 3389988 : amroutine->amvacuumcleanup = btvacuumcleanup;
151 3389988 : amroutine->amcanreturn = btcanreturn;
152 3389988 : amroutine->amcostestimate = btcostestimate;
153 3389988 : amroutine->amgettreeheight = btgettreeheight;
154 3389988 : amroutine->amoptions = btoptions;
155 3389988 : amroutine->amproperty = btproperty;
156 3389988 : amroutine->ambuildphasename = btbuildphasename;
157 3389988 : amroutine->amvalidate = btvalidate;
158 3389988 : amroutine->amadjustmembers = btadjustmembers;
159 3389988 : amroutine->ambeginscan = btbeginscan;
160 3389988 : amroutine->amrescan = btrescan;
161 3389988 : amroutine->amgettuple = btgettuple;
162 3389988 : amroutine->amgetbitmap = btgetbitmap;
163 3389988 : amroutine->amendscan = btendscan;
164 3389988 : amroutine->ammarkpos = btmarkpos;
165 3389988 : amroutine->amrestrpos = btrestrpos;
166 3389988 : amroutine->amestimateparallelscan = btestimateparallelscan;
167 3389988 : amroutine->aminitparallelscan = btinitparallelscan;
168 3389988 : amroutine->amparallelrescan = btparallelrescan;
169 3389988 : amroutine->amtranslatestrategy = bttranslatestrategy;
170 3389988 : amroutine->amtranslatecmptype = bttranslatecmptype;
171 :
172 3389988 : PG_RETURN_POINTER(amroutine);
173 : }
174 :
175 : /*
176 : * btbuildempty() -- build an empty btree index in the initialization fork
177 : */
178 : void
179 164 : btbuildempty(Relation index)
180 : {
181 164 : bool allequalimage = _bt_allequalimage(index, false);
182 : BulkWriteState *bulkstate;
183 : BulkWriteBuffer metabuf;
184 :
185 164 : bulkstate = smgr_bulk_start_rel(index, INIT_FORKNUM);
186 :
187 : /* Construct metapage. */
188 164 : metabuf = smgr_bulk_get_buf(bulkstate);
189 164 : _bt_initmetapage((Page) metabuf, P_NONE, 0, allequalimage);
190 164 : smgr_bulk_write(bulkstate, BTREE_METAPAGE, metabuf, true);
191 :
192 164 : smgr_bulk_finish(bulkstate);
193 164 : }
194 :
195 : /*
196 : * btinsert() -- insert an index tuple into a btree.
197 : *
198 : * Descend the tree recursively, find the appropriate location for our
199 : * new tuple, and put it there.
200 : */
201 : bool
202 7213298 : btinsert(Relation rel, Datum *values, bool *isnull,
203 : ItemPointer ht_ctid, Relation heapRel,
204 : IndexUniqueCheck checkUnique,
205 : bool indexUnchanged,
206 : IndexInfo *indexInfo)
207 : {
208 : bool result;
209 : IndexTuple itup;
210 :
211 : /* generate an index tuple */
212 7213298 : itup = index_form_tuple(RelationGetDescr(rel), values, isnull);
213 7213298 : itup->t_tid = *ht_ctid;
214 :
215 7213298 : result = _bt_doinsert(rel, itup, checkUnique, indexUnchanged, heapRel);
216 :
217 7212786 : pfree(itup);
218 :
219 7212786 : return result;
220 : }
221 :
222 : /*
223 : * btgettuple() -- Get the next tuple in the scan.
224 : */
225 : bool
226 34875996 : btgettuple(IndexScanDesc scan, ScanDirection dir)
227 : {
228 34875996 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
229 : bool res;
230 :
231 : /* btree indexes are never lossy */
232 34875996 : scan->xs_recheck = false;
233 :
234 : /* Each loop iteration performs another primitive index scan */
235 : do
236 : {
237 : /*
238 : * If we've already initialized this scan, we can just advance it in
239 : * the appropriate direction. If we haven't done so yet, we call
240 : * _bt_first() to get the first item in the scan.
241 : */
242 34892896 : if (!BTScanPosIsValid(so->currPos))
243 15390880 : res = _bt_first(scan, dir);
244 : else
245 : {
246 : /*
247 : * Check to see if we should kill the previously-fetched tuple.
248 : */
249 19502016 : if (scan->kill_prior_tuple)
250 : {
251 : /*
252 : * Yes, remember it for later. (We'll deal with all such
253 : * tuples at once right before leaving the index page.) The
254 : * test for numKilled overrun is not just paranoia: if the
255 : * caller reverses direction in the indexscan then the same
256 : * item might get entered multiple times. It's not worth
257 : * trying to optimize that, so we don't detect it, but instead
258 : * just forget any excess entries.
259 : */
260 489706 : if (so->killedItems == NULL)
261 170344 : so->killedItems = (int *)
262 170344 : palloc(MaxTIDsPerBTreePage * sizeof(int));
263 489706 : if (so->numKilled < MaxTIDsPerBTreePage)
264 489706 : so->killedItems[so->numKilled++] = so->currPos.itemIndex;
265 : }
266 :
267 : /*
268 : * Now continue the scan.
269 : */
270 19502016 : res = _bt_next(scan, dir);
271 : }
272 :
273 : /* If we have a tuple, return it ... */
274 34892896 : if (res)
275 27907180 : break;
276 : /* ... otherwise see if we need another primitive index scan */
277 6985716 : } while (so->numArrayKeys && _bt_start_prim_scan(scan, dir));
278 :
279 34875996 : return res;
280 : }
281 :
282 : /*
283 : * btgetbitmap() -- gets all matching tuples, and adds them to a bitmap
284 : */
285 : int64
286 13552 : btgetbitmap(IndexScanDesc scan, TIDBitmap *tbm)
287 : {
288 13552 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
289 13552 : int64 ntids = 0;
290 : ItemPointer heapTid;
291 :
292 : /* Each loop iteration performs another primitive index scan */
293 : do
294 : {
295 : /* Fetch the first page & tuple */
296 14612 : if (_bt_first(scan, ForwardScanDirection))
297 : {
298 : /* Save tuple ID, and continue scanning */
299 10484 : heapTid = &scan->xs_heaptid;
300 10484 : tbm_add_tuples(tbm, heapTid, 1, false);
301 10484 : ntids++;
302 :
303 : for (;;)
304 : {
305 : /*
306 : * Advance to next tuple within page. This is the same as the
307 : * easy case in _bt_next().
308 : */
309 2044622 : if (++so->currPos.itemIndex > so->currPos.lastItem)
310 : {
311 : /* let _bt_next do the heavy lifting */
312 16162 : if (!_bt_next(scan, ForwardScanDirection))
313 10484 : break;
314 : }
315 :
316 : /* Save tuple ID, and continue scanning */
317 2034138 : heapTid = &so->currPos.items[so->currPos.itemIndex].heapTid;
318 2034138 : tbm_add_tuples(tbm, heapTid, 1, false);
319 2034138 : ntids++;
320 : }
321 : }
322 : /* Now see if we need another primitive index scan */
323 14612 : } while (so->numArrayKeys && _bt_start_prim_scan(scan, ForwardScanDirection));
324 :
325 13552 : return ntids;
326 : }
327 :
328 : /*
329 : * btbeginscan() -- start a scan on a btree index
330 : */
331 : IndexScanDesc
332 14685860 : btbeginscan(Relation rel, int nkeys, int norderbys)
333 : {
334 : IndexScanDesc scan;
335 : BTScanOpaque so;
336 :
337 : /* no order by operators allowed */
338 : Assert(norderbys == 0);
339 :
340 : /* get the scan */
341 14685860 : scan = RelationGetIndexScan(rel, nkeys, norderbys);
342 :
343 : /* allocate private workspace */
344 14685860 : so = (BTScanOpaque) palloc(sizeof(BTScanOpaqueData));
345 14685860 : BTScanPosInvalidate(so->currPos);
346 14685860 : BTScanPosInvalidate(so->markPos);
347 14685860 : if (scan->numberOfKeys > 0)
348 14672714 : so->keyData = (ScanKey) palloc(scan->numberOfKeys * sizeof(ScanKeyData));
349 : else
350 13146 : so->keyData = NULL;
351 :
352 14685860 : so->skipScan = false;
353 14685860 : so->needPrimScan = false;
354 14685860 : so->scanBehind = false;
355 14685860 : so->oppositeDirCheck = false;
356 14685860 : so->arrayKeys = NULL;
357 14685860 : so->orderProcs = NULL;
358 14685860 : so->arrayContext = NULL;
359 :
360 14685860 : so->killedItems = NULL; /* until needed */
361 14685860 : so->numKilled = 0;
362 :
363 : /*
364 : * We don't know yet whether the scan will be index-only, so we do not
365 : * allocate the tuple workspace arrays until btrescan. However, we set up
366 : * scan->xs_itupdesc whether we'll need it or not, since that's so cheap.
367 : */
368 14685860 : so->currTuples = so->markTuples = NULL;
369 :
370 14685860 : scan->xs_itupdesc = RelationGetDescr(rel);
371 :
372 14685860 : scan->opaque = so;
373 :
374 14685860 : return scan;
375 : }
376 :
377 : /*
378 : * btrescan() -- rescan an index relation
379 : */
380 : void
381 15388814 : btrescan(IndexScanDesc scan, ScanKey scankey, int nscankeys,
382 : ScanKey orderbys, int norderbys)
383 : {
384 15388814 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
385 :
386 : /* we aren't holding any read locks, but gotta drop the pins */
387 15388814 : if (BTScanPosIsValid(so->currPos))
388 : {
389 : /* Before leaving current page, deal with any killed items */
390 93290 : if (so->numKilled > 0)
391 1286 : _bt_killitems(scan);
392 93290 : BTScanPosUnpinIfPinned(so->currPos);
393 93290 : BTScanPosInvalidate(so->currPos);
394 : }
395 :
396 15388814 : so->markItemIndex = -1;
397 15388814 : so->needPrimScan = false;
398 15388814 : so->scanBehind = false;
399 15388814 : so->oppositeDirCheck = false;
400 15388814 : BTScanPosUnpinIfPinned(so->markPos);
401 15388814 : BTScanPosInvalidate(so->markPos);
402 :
403 : /*
404 : * Allocate tuple workspace arrays, if needed for an index-only scan and
405 : * not already done in a previous rescan call. To save on palloc
406 : * overhead, both workspaces are allocated as one palloc block; only this
407 : * function and btendscan know that.
408 : *
409 : * NOTE: this data structure also makes it safe to return data from a
410 : * "name" column, even though btree name_ops uses an underlying storage
411 : * datatype of cstring. The risk there is that "name" is supposed to be
412 : * padded to NAMEDATALEN, but the actual index tuple is probably shorter.
413 : * However, since we only return data out of tuples sitting in the
414 : * currTuples array, a fetch of NAMEDATALEN bytes can at worst pull some
415 : * data out of the markTuples array --- running off the end of memory for
416 : * a SIGSEGV is not possible. Yeah, this is ugly as sin, but it beats
417 : * adding special-case treatment for name_ops elsewhere.
418 : */
419 15388814 : if (scan->xs_want_itup && so->currTuples == NULL)
420 : {
421 133320 : so->currTuples = (char *) palloc(BLCKSZ * 2);
422 133320 : so->markTuples = so->currTuples + BLCKSZ;
423 : }
424 :
425 : /*
426 : * Reset the scan keys
427 : */
428 15388814 : if (scankey && scan->numberOfKeys > 0)
429 15375474 : memcpy(scan->keyData, scankey, scan->numberOfKeys * sizeof(ScanKeyData));
430 15388814 : so->numberOfKeys = 0; /* until _bt_preprocess_keys sets it */
431 15388814 : so->numArrayKeys = 0; /* ditto */
432 15388814 : }
433 :
434 : /*
435 : * btendscan() -- close down a scan
436 : */
437 : void
438 14684310 : btendscan(IndexScanDesc scan)
439 : {
440 14684310 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
441 :
442 : /* we aren't holding any read locks, but gotta drop the pins */
443 14684310 : if (BTScanPosIsValid(so->currPos))
444 : {
445 : /* Before leaving current page, deal with any killed items */
446 8310762 : if (so->numKilled > 0)
447 91246 : _bt_killitems(scan);
448 8310762 : BTScanPosUnpinIfPinned(so->currPos);
449 : }
450 :
451 14684310 : so->markItemIndex = -1;
452 14684310 : BTScanPosUnpinIfPinned(so->markPos);
453 :
454 : /* No need to invalidate positions, the RAM is about to be freed. */
455 :
456 : /* Release storage */
457 14684310 : if (so->keyData != NULL)
458 14671194 : pfree(so->keyData);
459 : /* so->arrayKeys and so->orderProcs are in arrayContext */
460 14684310 : if (so->arrayContext != NULL)
461 4750 : MemoryContextDelete(so->arrayContext);
462 14684310 : if (so->killedItems != NULL)
463 170288 : pfree(so->killedItems);
464 14684310 : if (so->currTuples != NULL)
465 133276 : pfree(so->currTuples);
466 : /* so->markTuples should not be pfree'd, see btrescan */
467 14684310 : pfree(so);
468 14684310 : }
469 :
470 : /*
471 : * btmarkpos() -- save current scan position
472 : */
473 : void
474 130074 : btmarkpos(IndexScanDesc scan)
475 : {
476 130074 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
477 :
478 : /* There may be an old mark with a pin (but no lock). */
479 130074 : BTScanPosUnpinIfPinned(so->markPos);
480 :
481 : /*
482 : * Just record the current itemIndex. If we later step to next page
483 : * before releasing the marked position, _bt_steppage makes a full copy of
484 : * the currPos struct in markPos. If (as often happens) the mark is moved
485 : * before we leave the page, we don't have to do that work.
486 : */
487 130074 : if (BTScanPosIsValid(so->currPos))
488 130074 : so->markItemIndex = so->currPos.itemIndex;
489 : else
490 : {
491 0 : BTScanPosInvalidate(so->markPos);
492 0 : so->markItemIndex = -1;
493 : }
494 130074 : }
495 :
496 : /*
497 : * btrestrpos() -- restore scan to last saved position
498 : */
499 : void
500 54018 : btrestrpos(IndexScanDesc scan)
501 : {
502 54018 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
503 :
504 54018 : if (so->markItemIndex >= 0)
505 : {
506 : /*
507 : * The scan has never moved to a new page since the last mark. Just
508 : * restore the itemIndex.
509 : *
510 : * NB: In this case we can't count on anything in so->markPos to be
511 : * accurate.
512 : */
513 53910 : so->currPos.itemIndex = so->markItemIndex;
514 : }
515 : else
516 : {
517 : /*
518 : * The scan moved to a new page after last mark or restore, and we are
519 : * now restoring to the marked page. We aren't holding any read
520 : * locks, but if we're still holding the pin for the current position,
521 : * we must drop it.
522 : */
523 108 : if (BTScanPosIsValid(so->currPos))
524 : {
525 : /* Before leaving current page, deal with any killed items */
526 108 : if (so->numKilled > 0)
527 0 : _bt_killitems(scan);
528 108 : BTScanPosUnpinIfPinned(so->currPos);
529 : }
530 :
531 108 : if (BTScanPosIsValid(so->markPos))
532 : {
533 : /* bump pin on mark buffer for assignment to current buffer */
534 108 : if (BTScanPosIsPinned(so->markPos))
535 0 : IncrBufferRefCount(so->markPos.buf);
536 108 : memcpy(&so->currPos, &so->markPos,
537 : offsetof(BTScanPosData, items[1]) +
538 108 : so->markPos.lastItem * sizeof(BTScanPosItem));
539 108 : if (so->currTuples)
540 0 : memcpy(so->currTuples, so->markTuples,
541 0 : so->markPos.nextTupleOffset);
542 : /* Reset the scan's array keys (see _bt_steppage for why) */
543 108 : if (so->numArrayKeys)
544 : {
545 0 : _bt_start_array_keys(scan, so->currPos.dir);
546 0 : so->needPrimScan = false;
547 : }
548 : }
549 : else
550 0 : BTScanPosInvalidate(so->currPos);
551 : }
552 54018 : }
553 :
554 : /*
555 : * btestimateparallelscan -- estimate storage for BTParallelScanDescData
556 : */
557 : Size
558 64 : btestimateparallelscan(Relation rel, int nkeys, int norderbys)
559 : {
560 64 : int16 nkeyatts = IndexRelationGetNumberOfKeyAttributes(rel);
561 : Size estnbtreeshared,
562 : genericattrspace;
563 :
564 : /*
565 : * Pessimistically assume that every input scan key will be output with
566 : * its own SAOP array
567 : */
568 64 : estnbtreeshared = offsetof(BTParallelScanDescData, btps_arrElems) +
569 : sizeof(int) * nkeys;
570 :
571 : /* Single column indexes cannot possibly use a skip array */
572 64 : if (nkeyatts == 1)
573 46 : return estnbtreeshared;
574 :
575 : /*
576 : * Pessimistically assume that all attributes prior to the least
577 : * significant attribute require a skip array (and an associated key)
578 : */
579 18 : genericattrspace = datumEstimateSpace((Datum) 0, false, true,
580 : sizeof(Datum));
581 36 : for (int attnum = 1; attnum < nkeyatts; attnum++)
582 : {
583 : CompactAttribute *attr;
584 :
585 : /*
586 : * We make the conservative assumption that every index column will
587 : * also require a skip array.
588 : *
589 : * Every skip array must have space to store its scan key's sk_flags.
590 : */
591 18 : estnbtreeshared = add_size(estnbtreeshared, sizeof(int));
592 :
593 : /* Consider space required to store a datum of opclass input type */
594 18 : attr = TupleDescCompactAttr(rel->rd_att, attnum - 1);
595 18 : if (attr->attbyval)
596 : {
597 : /* This index attribute stores pass-by-value datums */
598 18 : Size estfixed = datumEstimateSpace((Datum) 0, false,
599 18 : true, attr->attlen);
600 :
601 18 : estnbtreeshared = add_size(estnbtreeshared, estfixed);
602 18 : continue;
603 : }
604 :
605 : /*
606 : * This index attribute stores pass-by-reference datums.
607 : *
608 : * Assume that serializing this array will use just as much space as a
609 : * pass-by-value datum, in addition to space for the largest possible
610 : * whole index tuple (this is not just a per-datum portion of the
611 : * largest possible tuple because that'd be almost as large anyway).
612 : *
613 : * This is quite conservative, but it's not clear how we could do much
614 : * better. The executor requires an up-front storage request size
615 : * that reliably covers the scan's high watermark memory usage. We
616 : * can't be sure of the real high watermark until the scan is over.
617 : */
618 0 : estnbtreeshared = add_size(estnbtreeshared, genericattrspace);
619 0 : estnbtreeshared = add_size(estnbtreeshared, BTMaxItemSize);
620 : }
621 :
622 18 : return estnbtreeshared;
623 : }
624 :
625 : /*
626 : * _bt_parallel_serialize_arrays() -- Serialize parallel array state.
627 : *
628 : * Caller must have exclusively locked btscan->btps_lock when called.
629 : */
630 : static void
631 36 : _bt_parallel_serialize_arrays(Relation rel, BTParallelScanDesc btscan,
632 : BTScanOpaque so)
633 : {
634 : char *datumshared;
635 :
636 : /* Space for serialized datums begins immediately after btps_arrElems[] */
637 36 : datumshared = ((char *) &btscan->btps_arrElems[so->numArrayKeys]);
638 72 : for (int i = 0; i < so->numArrayKeys; i++)
639 : {
640 36 : BTArrayKeyInfo *array = &so->arrayKeys[i];
641 36 : ScanKey skey = &so->keyData[array->scan_key];
642 :
643 36 : if (array->num_elems != -1)
644 : {
645 : /* Save SAOP array's cur_elem (no need to copy key/datum) */
646 : Assert(!(skey->sk_flags & SK_BT_SKIP));
647 36 : btscan->btps_arrElems[i] = array->cur_elem;
648 36 : continue;
649 : }
650 :
651 : /* Save all mutable state associated with skip array's key */
652 : Assert(skey->sk_flags & SK_BT_SKIP);
653 0 : memcpy(datumshared, &skey->sk_flags, sizeof(int));
654 0 : datumshared += sizeof(int);
655 :
656 0 : if (skey->sk_flags & (SK_BT_MINVAL | SK_BT_MAXVAL))
657 : {
658 : /* No sk_argument datum to serialize */
659 : Assert(skey->sk_argument == 0);
660 0 : continue;
661 : }
662 :
663 0 : datumSerialize(skey->sk_argument, (skey->sk_flags & SK_ISNULL) != 0,
664 0 : array->attbyval, array->attlen, &datumshared);
665 : }
666 36 : }
667 :
668 : /*
669 : * _bt_parallel_restore_arrays() -- Restore serialized parallel array state.
670 : *
671 : * Caller must have exclusively locked btscan->btps_lock when called.
672 : */
673 : static void
674 36 : _bt_parallel_restore_arrays(Relation rel, BTParallelScanDesc btscan,
675 : BTScanOpaque so)
676 : {
677 : char *datumshared;
678 :
679 : /* Space for serialized datums begins immediately after btps_arrElems[] */
680 36 : datumshared = ((char *) &btscan->btps_arrElems[so->numArrayKeys]);
681 72 : for (int i = 0; i < so->numArrayKeys; i++)
682 : {
683 36 : BTArrayKeyInfo *array = &so->arrayKeys[i];
684 36 : ScanKey skey = &so->keyData[array->scan_key];
685 : bool isnull;
686 :
687 36 : if (array->num_elems != -1)
688 : {
689 : /* Restore SAOP array using its saved cur_elem */
690 : Assert(!(skey->sk_flags & SK_BT_SKIP));
691 36 : array->cur_elem = btscan->btps_arrElems[i];
692 36 : skey->sk_argument = array->elem_values[array->cur_elem];
693 36 : continue;
694 : }
695 :
696 : /* Restore skip array by restoring its key directly */
697 0 : if (!array->attbyval && skey->sk_argument)
698 0 : pfree(DatumGetPointer(skey->sk_argument));
699 0 : skey->sk_argument = (Datum) 0;
700 0 : memcpy(&skey->sk_flags, datumshared, sizeof(int));
701 0 : datumshared += sizeof(int);
702 :
703 : Assert(skey->sk_flags & SK_BT_SKIP);
704 :
705 0 : if (skey->sk_flags & (SK_BT_MINVAL | SK_BT_MAXVAL))
706 : {
707 : /* No sk_argument datum to restore */
708 0 : continue;
709 : }
710 :
711 0 : skey->sk_argument = datumRestore(&datumshared, &isnull);
712 0 : if (isnull)
713 : {
714 : Assert(skey->sk_argument == 0);
715 : Assert(skey->sk_flags & SK_SEARCHNULL);
716 : Assert(skey->sk_flags & SK_ISNULL);
717 : }
718 : }
719 36 : }
720 :
721 : /*
722 : * btinitparallelscan -- initialize BTParallelScanDesc for parallel btree scan
723 : */
724 : void
725 64 : btinitparallelscan(void *target)
726 : {
727 64 : BTParallelScanDesc bt_target = (BTParallelScanDesc) target;
728 :
729 64 : LWLockInitialize(&bt_target->btps_lock,
730 : LWTRANCHE_PARALLEL_BTREE_SCAN);
731 64 : bt_target->btps_nextScanPage = InvalidBlockNumber;
732 64 : bt_target->btps_lastCurrPage = InvalidBlockNumber;
733 64 : bt_target->btps_pageStatus = BTPARALLEL_NOT_INITIALIZED;
734 64 : ConditionVariableInit(&bt_target->btps_cv);
735 64 : }
736 :
737 : /*
738 : * btparallelrescan() -- reset parallel scan
739 : */
740 : void
741 24 : btparallelrescan(IndexScanDesc scan)
742 : {
743 : BTParallelScanDesc btscan;
744 24 : ParallelIndexScanDesc parallel_scan = scan->parallel_scan;
745 :
746 : Assert(parallel_scan);
747 :
748 24 : btscan = (BTParallelScanDesc) OffsetToPointer(parallel_scan,
749 : parallel_scan->ps_offset_am);
750 :
751 : /*
752 : * In theory, we don't need to acquire the LWLock here, because there
753 : * shouldn't be any other workers running at this point, but we do so for
754 : * consistency.
755 : */
756 24 : LWLockAcquire(&btscan->btps_lock, LW_EXCLUSIVE);
757 24 : btscan->btps_nextScanPage = InvalidBlockNumber;
758 24 : btscan->btps_lastCurrPage = InvalidBlockNumber;
759 24 : btscan->btps_pageStatus = BTPARALLEL_NOT_INITIALIZED;
760 24 : LWLockRelease(&btscan->btps_lock);
761 24 : }
762 :
763 : /*
764 : * _bt_parallel_seize() -- Begin the process of advancing the scan to a new
765 : * page. Other scans must wait until we call _bt_parallel_release()
766 : * or _bt_parallel_done().
767 : *
768 : * The return value is true if we successfully seized the scan and false
769 : * if we did not. The latter case occurs when no pages remain, or when
770 : * another primitive index scan is scheduled that caller's backend cannot
771 : * start just yet (only backends that call from _bt_first are capable of
772 : * starting primitive index scans, which they indicate by passing first=true).
773 : *
774 : * If the return value is true, *next_scan_page returns the next page of the
775 : * scan, and *last_curr_page returns the page that *next_scan_page came from.
776 : * An invalid *next_scan_page means the scan hasn't yet started, or that
777 : * caller needs to start the next primitive index scan (if it's the latter
778 : * case we'll set so.needPrimScan).
779 : *
780 : * Callers should ignore the value of *next_scan_page and *last_curr_page if
781 : * the return value is false.
782 : */
783 : bool
784 1658 : _bt_parallel_seize(IndexScanDesc scan, BlockNumber *next_scan_page,
785 : BlockNumber *last_curr_page, bool first)
786 : {
787 1658 : Relation rel = scan->indexRelation;
788 1658 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
789 1658 : bool exit_loop = false,
790 1658 : status = true,
791 1658 : endscan = false;
792 1658 : ParallelIndexScanDesc parallel_scan = scan->parallel_scan;
793 : BTParallelScanDesc btscan;
794 :
795 1658 : *next_scan_page = InvalidBlockNumber;
796 1658 : *last_curr_page = InvalidBlockNumber;
797 :
798 : /*
799 : * Reset so->currPos, and initialize moreLeft/moreRight such that the next
800 : * call to _bt_readnextpage treats this backend similarly to a serial
801 : * backend that steps from *last_curr_page to *next_scan_page (unless this
802 : * backend's so->currPos is initialized by _bt_readfirstpage before then).
803 : */
804 1658 : BTScanPosInvalidate(so->currPos);
805 1658 : so->currPos.moreLeft = so->currPos.moreRight = true;
806 :
807 1658 : if (first)
808 : {
809 : /*
810 : * Initialize array related state when called from _bt_first, assuming
811 : * that this will be the first primitive index scan for the scan
812 : */
813 446 : so->needPrimScan = false;
814 446 : so->scanBehind = false;
815 446 : so->oppositeDirCheck = false;
816 : }
817 : else
818 : {
819 : /*
820 : * Don't attempt to seize the scan when it requires another primitive
821 : * index scan, since caller's backend cannot start it right now
822 : */
823 1212 : if (so->needPrimScan)
824 0 : return false;
825 : }
826 :
827 1658 : btscan = (BTParallelScanDesc) OffsetToPointer(parallel_scan,
828 : parallel_scan->ps_offset_am);
829 :
830 : while (1)
831 : {
832 1658 : LWLockAcquire(&btscan->btps_lock, LW_EXCLUSIVE);
833 :
834 1658 : if (btscan->btps_pageStatus == BTPARALLEL_DONE)
835 : {
836 : /* We're done with this parallel index scan */
837 314 : status = false;
838 : }
839 1344 : else if (btscan->btps_pageStatus == BTPARALLEL_IDLE &&
840 1220 : btscan->btps_nextScanPage == P_NONE)
841 : {
842 : /* End this parallel index scan */
843 8 : status = false;
844 8 : endscan = true;
845 : }
846 1336 : else if (btscan->btps_pageStatus == BTPARALLEL_NEED_PRIMSCAN)
847 : {
848 : Assert(so->numArrayKeys);
849 :
850 36 : if (first)
851 : {
852 : /* Can start scheduled primitive scan right away, so do so */
853 36 : btscan->btps_pageStatus = BTPARALLEL_ADVANCING;
854 :
855 : /* Restore scan's array keys from serialized values */
856 36 : _bt_parallel_restore_arrays(rel, btscan, so);
857 36 : exit_loop = true;
858 : }
859 : else
860 : {
861 : /*
862 : * Don't attempt to seize the scan when it requires another
863 : * primitive index scan, since caller's backend cannot start
864 : * it right now
865 : */
866 0 : status = false;
867 : }
868 :
869 : /*
870 : * Either way, update backend local state to indicate that a
871 : * pending primitive scan is required
872 : */
873 36 : so->needPrimScan = true;
874 36 : so->scanBehind = false;
875 36 : so->oppositeDirCheck = false;
876 : }
877 1300 : else if (btscan->btps_pageStatus != BTPARALLEL_ADVANCING)
878 : {
879 : /*
880 : * We have successfully seized control of the scan for the purpose
881 : * of advancing it to a new page!
882 : */
883 1300 : btscan->btps_pageStatus = BTPARALLEL_ADVANCING;
884 : Assert(btscan->btps_nextScanPage != P_NONE);
885 1300 : *next_scan_page = btscan->btps_nextScanPage;
886 1300 : *last_curr_page = btscan->btps_lastCurrPage;
887 1300 : exit_loop = true;
888 : }
889 1658 : LWLockRelease(&btscan->btps_lock);
890 1658 : if (exit_loop || !status)
891 : break;
892 0 : ConditionVariableSleep(&btscan->btps_cv, WAIT_EVENT_BTREE_PAGE);
893 : }
894 1658 : ConditionVariableCancelSleep();
895 :
896 : /* When the scan has reached the rightmost (or leftmost) page, end it */
897 1658 : if (endscan)
898 8 : _bt_parallel_done(scan);
899 :
900 1658 : return status;
901 : }
902 :
903 : /*
904 : * _bt_parallel_release() -- Complete the process of advancing the scan to a
905 : * new page. We now have the new value btps_nextScanPage; another backend
906 : * can now begin advancing the scan.
907 : *
908 : * Callers whose scan uses array keys must save their curr_page argument so
909 : * that it can be passed to _bt_parallel_primscan_schedule, should caller
910 : * determine that another primitive index scan is required.
911 : *
912 : * If caller's next_scan_page is P_NONE, the scan has reached the index's
913 : * rightmost/leftmost page. This is treated as reaching the end of the scan
914 : * within _bt_parallel_seize.
915 : *
916 : * Note: unlike the serial case, parallel scans don't need to remember both
917 : * sibling links. next_scan_page is whichever link is next given the scan's
918 : * direction. That's all we'll ever need, since the direction of a parallel
919 : * scan can never change.
920 : */
921 : void
922 1336 : _bt_parallel_release(IndexScanDesc scan, BlockNumber next_scan_page,
923 : BlockNumber curr_page)
924 : {
925 1336 : ParallelIndexScanDesc parallel_scan = scan->parallel_scan;
926 : BTParallelScanDesc btscan;
927 :
928 : Assert(BlockNumberIsValid(next_scan_page));
929 :
930 1336 : btscan = (BTParallelScanDesc) OffsetToPointer(parallel_scan,
931 : parallel_scan->ps_offset_am);
932 :
933 1336 : LWLockAcquire(&btscan->btps_lock, LW_EXCLUSIVE);
934 1336 : btscan->btps_nextScanPage = next_scan_page;
935 1336 : btscan->btps_lastCurrPage = curr_page;
936 1336 : btscan->btps_pageStatus = BTPARALLEL_IDLE;
937 1336 : LWLockRelease(&btscan->btps_lock);
938 1336 : ConditionVariableSignal(&btscan->btps_cv);
939 1336 : }
940 :
941 : /*
942 : * _bt_parallel_done() -- Mark the parallel scan as complete.
943 : *
944 : * When there are no pages left to scan, this function should be called to
945 : * notify other workers. Otherwise, they might wait forever for the scan to
946 : * advance to the next page.
947 : */
948 : void
949 7000044 : _bt_parallel_done(IndexScanDesc scan)
950 : {
951 7000044 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
952 7000044 : ParallelIndexScanDesc parallel_scan = scan->parallel_scan;
953 : BTParallelScanDesc btscan;
954 7000044 : bool status_changed = false;
955 :
956 : Assert(!BTScanPosIsValid(so->currPos));
957 :
958 : /* Do nothing, for non-parallel scans */
959 7000044 : if (parallel_scan == NULL)
960 6999888 : return;
961 :
962 : /*
963 : * Should not mark parallel scan done when there's still a pending
964 : * primitive index scan
965 : */
966 156 : if (so->needPrimScan)
967 36 : return;
968 :
969 120 : btscan = (BTParallelScanDesc) OffsetToPointer(parallel_scan,
970 : parallel_scan->ps_offset_am);
971 :
972 : /*
973 : * Mark the parallel scan as done, unless some other process did so
974 : * already
975 : */
976 120 : LWLockAcquire(&btscan->btps_lock, LW_EXCLUSIVE);
977 : Assert(btscan->btps_pageStatus != BTPARALLEL_NEED_PRIMSCAN);
978 120 : if (btscan->btps_pageStatus != BTPARALLEL_DONE)
979 : {
980 88 : btscan->btps_pageStatus = BTPARALLEL_DONE;
981 88 : status_changed = true;
982 : }
983 120 : LWLockRelease(&btscan->btps_lock);
984 :
985 : /* wake up all the workers associated with this parallel scan */
986 120 : if (status_changed)
987 88 : ConditionVariableBroadcast(&btscan->btps_cv);
988 : }
989 :
990 : /*
991 : * _bt_parallel_primscan_schedule() -- Schedule another primitive index scan.
992 : *
993 : * Caller passes the curr_page most recently passed to _bt_parallel_release
994 : * by its backend. Caller successfully schedules the next primitive index scan
995 : * if the shared parallel state hasn't been seized since caller's backend last
996 : * advanced the scan.
997 : */
998 : void
999 36 : _bt_parallel_primscan_schedule(IndexScanDesc scan, BlockNumber curr_page)
1000 : {
1001 36 : Relation rel = scan->indexRelation;
1002 36 : BTScanOpaque so = (BTScanOpaque) scan->opaque;
1003 36 : ParallelIndexScanDesc parallel_scan = scan->parallel_scan;
1004 : BTParallelScanDesc btscan;
1005 :
1006 : Assert(so->numArrayKeys);
1007 :
1008 36 : btscan = (BTParallelScanDesc) OffsetToPointer(parallel_scan,
1009 : parallel_scan->ps_offset_am);
1010 :
1011 36 : LWLockAcquire(&btscan->btps_lock, LW_EXCLUSIVE);
1012 36 : if (btscan->btps_lastCurrPage == curr_page &&
1013 36 : btscan->btps_pageStatus == BTPARALLEL_IDLE)
1014 : {
1015 36 : btscan->btps_nextScanPage = InvalidBlockNumber;
1016 36 : btscan->btps_lastCurrPage = InvalidBlockNumber;
1017 36 : btscan->btps_pageStatus = BTPARALLEL_NEED_PRIMSCAN;
1018 :
1019 : /* Serialize scan's current array keys */
1020 36 : _bt_parallel_serialize_arrays(rel, btscan, so);
1021 : }
1022 36 : LWLockRelease(&btscan->btps_lock);
1023 36 : }
1024 :
1025 : /*
1026 : * Bulk deletion of all index entries pointing to a set of heap tuples.
1027 : * The set of target tuples is specified via a callback routine that tells
1028 : * whether any given heap tuple (identified by ItemPointer) is being deleted.
1029 : *
1030 : * Result: a palloc'd struct containing statistical info for VACUUM displays.
1031 : */
1032 : IndexBulkDeleteResult *
1033 2774 : btbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
1034 : IndexBulkDeleteCallback callback, void *callback_state)
1035 : {
1036 2774 : Relation rel = info->index;
1037 : BTCycleId cycleid;
1038 :
1039 : /* allocate stats if first time through, else re-use existing struct */
1040 2774 : if (stats == NULL)
1041 2774 : stats = (IndexBulkDeleteResult *) palloc0(sizeof(IndexBulkDeleteResult));
1042 :
1043 : /* Establish the vacuum cycle ID to use for this scan */
1044 : /* The ENSURE stuff ensures we clean up shared memory on failure */
1045 2774 : PG_ENSURE_ERROR_CLEANUP(_bt_end_vacuum_callback, PointerGetDatum(rel));
1046 : {
1047 2774 : cycleid = _bt_start_vacuum(rel);
1048 :
1049 2774 : btvacuumscan(info, stats, callback, callback_state, cycleid);
1050 : }
1051 2774 : PG_END_ENSURE_ERROR_CLEANUP(_bt_end_vacuum_callback, PointerGetDatum(rel));
1052 2774 : _bt_end_vacuum(rel);
1053 :
1054 2774 : return stats;
1055 : }
1056 :
1057 : /*
1058 : * Post-VACUUM cleanup.
1059 : *
1060 : * Result: a palloc'd struct containing statistical info for VACUUM displays.
1061 : */
1062 : IndexBulkDeleteResult *
1063 172792 : btvacuumcleanup(IndexVacuumInfo *info, IndexBulkDeleteResult *stats)
1064 : {
1065 : BlockNumber num_delpages;
1066 :
1067 : /* No-op in ANALYZE ONLY mode */
1068 172792 : if (info->analyze_only)
1069 17320 : return stats;
1070 :
1071 : /*
1072 : * If btbulkdelete was called, we need not do anything (we just maintain
1073 : * the information used within _bt_vacuum_needs_cleanup() by calling
1074 : * _bt_set_cleanup_info() below).
1075 : *
1076 : * If btbulkdelete was _not_ called, then we have a choice to make: we
1077 : * must decide whether or not a btvacuumscan() call is needed now (i.e.
1078 : * whether the ongoing VACUUM operation can entirely avoid a physical scan
1079 : * of the index). A call to _bt_vacuum_needs_cleanup() decides it for us
1080 : * now.
1081 : */
1082 155472 : if (stats == NULL)
1083 : {
1084 : /* Check if VACUUM operation can entirely avoid btvacuumscan() call */
1085 153310 : if (!_bt_vacuum_needs_cleanup(info->index))
1086 153294 : return NULL;
1087 :
1088 : /*
1089 : * Since we aren't going to actually delete any leaf items, there's no
1090 : * need to go through all the vacuum-cycle-ID pushups here.
1091 : *
1092 : * Posting list tuples are a source of inaccuracy for cleanup-only
1093 : * scans. btvacuumscan() will assume that the number of index tuples
1094 : * from each page can be used as num_index_tuples, even though
1095 : * num_index_tuples is supposed to represent the number of TIDs in the
1096 : * index. This naive approach can underestimate the number of tuples
1097 : * in the index significantly.
1098 : *
1099 : * We handle the problem by making num_index_tuples an estimate in
1100 : * cleanup-only case.
1101 : */
1102 16 : stats = (IndexBulkDeleteResult *) palloc0(sizeof(IndexBulkDeleteResult));
1103 16 : btvacuumscan(info, stats, NULL, NULL, 0);
1104 16 : stats->estimated_count = true;
1105 : }
1106 :
1107 : /*
1108 : * Maintain num_delpages value in metapage for _bt_vacuum_needs_cleanup().
1109 : *
1110 : * num_delpages is the number of deleted pages now in the index that were
1111 : * not safe to place in the FSM to be recycled just yet. num_delpages is
1112 : * greater than 0 only when _bt_pagedel() actually deleted pages during
1113 : * our call to btvacuumscan(). Even then, _bt_pendingfsm_finalize() must
1114 : * have failed to place any newly deleted pages in the FSM just moments
1115 : * ago. (Actually, there are edge cases where recycling of the current
1116 : * VACUUM's newly deleted pages does not even become safe by the time the
1117 : * next VACUUM comes around. See nbtree/README.)
1118 : */
1119 : Assert(stats->pages_deleted >= stats->pages_free);
1120 2178 : num_delpages = stats->pages_deleted - stats->pages_free;
1121 2178 : _bt_set_cleanup_info(info->index, num_delpages);
1122 :
1123 : /*
1124 : * It's quite possible for us to be fooled by concurrent page splits into
1125 : * double-counting some index tuples, so disbelieve any total that exceeds
1126 : * the underlying heap's count ... if we know that accurately. Otherwise
1127 : * this might just make matters worse.
1128 : */
1129 2178 : if (!info->estimated_count)
1130 : {
1131 2132 : if (stats->num_index_tuples > info->num_heap_tuples)
1132 0 : stats->num_index_tuples = info->num_heap_tuples;
1133 : }
1134 :
1135 2178 : return stats;
1136 : }
1137 :
1138 : /*
1139 : * btvacuumscan --- scan the index for VACUUMing purposes
1140 : *
1141 : * This combines the functions of looking for leaf tuples that are deletable
1142 : * according to the vacuum callback, looking for empty pages that can be
1143 : * deleted, and looking for old deleted pages that can be recycled. Both
1144 : * btbulkdelete and btvacuumcleanup invoke this (the latter only if no
1145 : * btbulkdelete call occurred and _bt_vacuum_needs_cleanup returned true).
1146 : *
1147 : * The caller is responsible for initially allocating/zeroing a stats struct
1148 : * and for obtaining a vacuum cycle ID if necessary.
1149 : */
1150 : static void
1151 2790 : btvacuumscan(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
1152 : IndexBulkDeleteCallback callback, void *callback_state,
1153 : BTCycleId cycleid)
1154 : {
1155 2790 : Relation rel = info->index;
1156 : BTVacState vstate;
1157 : BlockNumber num_pages;
1158 : bool needLock;
1159 : BlockRangeReadStreamPrivate p;
1160 2790 : ReadStream *stream = NULL;
1161 :
1162 : /*
1163 : * Reset fields that track information about the entire index now. This
1164 : * avoids double-counting in the case where a single VACUUM command
1165 : * requires multiple scans of the index.
1166 : *
1167 : * Avoid resetting the tuples_removed and pages_newly_deleted fields here,
1168 : * since they track information about the VACUUM command, and so must last
1169 : * across each call to btvacuumscan().
1170 : *
1171 : * (Note that pages_free is treated as state about the whole index, not
1172 : * the current VACUUM. This is appropriate because RecordFreeIndexPage()
1173 : * calls are idempotent, and get repeated for the same deleted pages in
1174 : * some scenarios. The point for us is to track the number of recyclable
1175 : * pages in the index at the end of the VACUUM command.)
1176 : */
1177 2790 : stats->num_pages = 0;
1178 2790 : stats->num_index_tuples = 0;
1179 2790 : stats->pages_deleted = 0;
1180 2790 : stats->pages_free = 0;
1181 :
1182 : /* Set up info to pass down to btvacuumpage */
1183 2790 : vstate.info = info;
1184 2790 : vstate.stats = stats;
1185 2790 : vstate.callback = callback;
1186 2790 : vstate.callback_state = callback_state;
1187 2790 : vstate.cycleid = cycleid;
1188 :
1189 : /* Create a temporary memory context to run _bt_pagedel in */
1190 2790 : vstate.pagedelcontext = AllocSetContextCreate(CurrentMemoryContext,
1191 : "_bt_pagedel",
1192 : ALLOCSET_DEFAULT_SIZES);
1193 :
1194 : /* Initialize vstate fields used by _bt_pendingfsm_finalize */
1195 2790 : vstate.bufsize = 0;
1196 2790 : vstate.maxbufsize = 0;
1197 2790 : vstate.pendingpages = NULL;
1198 2790 : vstate.npendingpages = 0;
1199 : /* Consider applying _bt_pendingfsm_finalize optimization */
1200 2790 : _bt_pendingfsm_init(rel, &vstate, (callback == NULL));
1201 :
1202 : /*
1203 : * The outer loop iterates over all index pages except the metapage, in
1204 : * physical order (we hope the kernel will cooperate in providing
1205 : * read-ahead for speed). It is critical that we visit all leaf pages,
1206 : * including ones added after we start the scan, else we might fail to
1207 : * delete some deletable tuples. Hence, we must repeatedly check the
1208 : * relation length. We must acquire the relation-extension lock while
1209 : * doing so to avoid a race condition: if someone else is extending the
1210 : * relation, there is a window where bufmgr/smgr have created a new
1211 : * all-zero page but it hasn't yet been write-locked by _bt_getbuf(). If
1212 : * we manage to scan such a page here, we'll improperly assume it can be
1213 : * recycled. Taking the lock synchronizes things enough to prevent a
1214 : * problem: either num_pages won't include the new page, or _bt_getbuf
1215 : * already has write lock on the buffer and it will be fully initialized
1216 : * before we can examine it. Also, we need not worry if a page is added
1217 : * immediately after we look; the page splitting code already has
1218 : * write-lock on the left page before it adds a right page, so we must
1219 : * already have processed any tuples due to be moved into such a page.
1220 : *
1221 : * XXX: Now that new pages are locked with RBM_ZERO_AND_LOCK, I don't
1222 : * think the use of the extension lock is still required.
1223 : *
1224 : * We can skip locking for new or temp relations, however, since no one
1225 : * else could be accessing them.
1226 : */
1227 2790 : needLock = !RELATION_IS_LOCAL(rel);
1228 :
1229 2790 : p.current_blocknum = BTREE_METAPAGE + 1;
1230 :
1231 : /*
1232 : * It is safe to use batchmode as block_range_read_stream_cb takes no
1233 : * locks.
1234 : */
1235 2790 : stream = read_stream_begin_relation(READ_STREAM_FULL |
1236 : READ_STREAM_USE_BATCHING,
1237 : info->strategy,
1238 : rel,
1239 : MAIN_FORKNUM,
1240 : block_range_read_stream_cb,
1241 : &p,
1242 : 0);
1243 : for (;;)
1244 : {
1245 : /* Get the current relation length */
1246 5306 : if (needLock)
1247 5302 : LockRelationForExtension(rel, ExclusiveLock);
1248 5306 : num_pages = RelationGetNumberOfBlocks(rel);
1249 5306 : if (needLock)
1250 5302 : UnlockRelationForExtension(rel, ExclusiveLock);
1251 :
1252 5306 : if (info->report_progress)
1253 950 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_TOTAL,
1254 : num_pages);
1255 :
1256 : /* Quit if we've scanned the whole relation */
1257 5306 : if (p.current_blocknum >= num_pages)
1258 2790 : break;
1259 :
1260 2516 : p.last_exclusive = num_pages;
1261 :
1262 : /* Iterate over pages, then loop back to recheck relation length */
1263 : while (true)
1264 23426 : {
1265 : BlockNumber current_block;
1266 : Buffer buf;
1267 :
1268 : /* call vacuum_delay_point while not holding any buffer lock */
1269 25942 : vacuum_delay_point(false);
1270 :
1271 25942 : buf = read_stream_next_buffer(stream, NULL);
1272 :
1273 25942 : if (!BufferIsValid(buf))
1274 2516 : break;
1275 :
1276 23426 : current_block = btvacuumpage(&vstate, buf);
1277 :
1278 23426 : if (info->report_progress)
1279 980 : pgstat_progress_update_param(PROGRESS_SCAN_BLOCKS_DONE,
1280 : current_block);
1281 : }
1282 :
1283 : /*
1284 : * We have to reset the read stream to use it again. After returning
1285 : * InvalidBuffer, the read stream API won't invoke our callback again
1286 : * until the stream has been reset.
1287 : */
1288 2516 : read_stream_reset(stream);
1289 : }
1290 :
1291 2790 : read_stream_end(stream);
1292 :
1293 : /* Set statistics num_pages field to final size of index */
1294 2790 : stats->num_pages = num_pages;
1295 :
1296 2790 : MemoryContextDelete(vstate.pagedelcontext);
1297 :
1298 : /*
1299 : * If there were any calls to _bt_pagedel() during scan of the index then
1300 : * see if any of the resulting pages can be placed in the FSM now. When
1301 : * it's not safe we'll have to leave it up to a future VACUUM operation.
1302 : *
1303 : * Finally, if we placed any pages in the FSM (either just now or during
1304 : * the scan), forcibly update the upper-level FSM pages to ensure that
1305 : * searchers can find them.
1306 : */
1307 2790 : _bt_pendingfsm_finalize(rel, &vstate);
1308 2790 : if (stats->pages_free > 0)
1309 30 : IndexFreeSpaceMapVacuum(rel);
1310 2790 : }
1311 :
1312 : /*
1313 : * btvacuumpage --- VACUUM one page
1314 : *
1315 : * This processes a single page for btvacuumscan(). In some cases we must
1316 : * backtrack to re-examine and VACUUM pages that were on buf's page during
1317 : * a previous call here. This is how we handle page splits (that happened
1318 : * after our cycleid was acquired) whose right half page happened to reuse
1319 : * a block that we might have processed at some point before it was
1320 : * recycled (i.e. before the page split).
1321 : *
1322 : * Returns BlockNumber of a scanned page (not backtracked).
1323 : */
1324 : static BlockNumber
1325 23426 : btvacuumpage(BTVacState *vstate, Buffer buf)
1326 : {
1327 23426 : IndexVacuumInfo *info = vstate->info;
1328 23426 : IndexBulkDeleteResult *stats = vstate->stats;
1329 23426 : IndexBulkDeleteCallback callback = vstate->callback;
1330 23426 : void *callback_state = vstate->callback_state;
1331 23426 : Relation rel = info->index;
1332 23426 : Relation heaprel = info->heaprel;
1333 : bool attempt_pagedel;
1334 : BlockNumber blkno,
1335 : backtrack_to;
1336 23426 : BlockNumber scanblkno = BufferGetBlockNumber(buf);
1337 : Page page;
1338 : BTPageOpaque opaque;
1339 :
1340 23426 : blkno = scanblkno;
1341 :
1342 23426 : backtrack:
1343 :
1344 23426 : attempt_pagedel = false;
1345 23426 : backtrack_to = P_NONE;
1346 :
1347 23426 : _bt_lockbuf(rel, buf, BT_READ);
1348 23426 : page = BufferGetPage(buf);
1349 23426 : opaque = NULL;
1350 23426 : if (!PageIsNew(page))
1351 : {
1352 23426 : _bt_checkpage(rel, buf);
1353 23426 : opaque = BTPageGetOpaque(page);
1354 : }
1355 :
1356 : Assert(blkno <= scanblkno);
1357 23426 : if (blkno != scanblkno)
1358 : {
1359 : /*
1360 : * We're backtracking.
1361 : *
1362 : * We followed a right link to a sibling leaf page (a page that
1363 : * happens to be from a block located before scanblkno). The only
1364 : * case we want to do anything with is a live leaf page having the
1365 : * current vacuum cycle ID.
1366 : *
1367 : * The page had better be in a state that's consistent with what we
1368 : * expect. Check for conditions that imply corruption in passing. It
1369 : * can't be half-dead because only an interrupted VACUUM process can
1370 : * leave pages in that state, so we'd definitely have dealt with it
1371 : * back when the page was the scanblkno page (half-dead pages are
1372 : * always marked fully deleted by _bt_pagedel(), barring corruption).
1373 : */
1374 0 : if (!opaque || !P_ISLEAF(opaque) || P_ISHALFDEAD(opaque))
1375 : {
1376 : Assert(false);
1377 0 : ereport(LOG,
1378 : (errcode(ERRCODE_INDEX_CORRUPTED),
1379 : errmsg_internal("right sibling %u of scanblkno %u unexpectedly in an inconsistent state in index \"%s\"",
1380 : blkno, scanblkno, RelationGetRelationName(rel))));
1381 0 : _bt_relbuf(rel, buf);
1382 0 : return scanblkno;
1383 : }
1384 :
1385 : /*
1386 : * We may have already processed the page in an earlier call, when the
1387 : * page was scanblkno. This happens when the leaf page split occurred
1388 : * after the scan began, but before the right sibling page became the
1389 : * scanblkno.
1390 : *
1391 : * Page may also have been deleted by current btvacuumpage() call,
1392 : * since _bt_pagedel() sometimes deletes the right sibling page of
1393 : * scanblkno in passing (it does so after we decided where to
1394 : * backtrack to). We don't need to process this page as a deleted
1395 : * page a second time now (in fact, it would be wrong to count it as a
1396 : * deleted page in the bulk delete statistics a second time).
1397 : */
1398 0 : if (opaque->btpo_cycleid != vstate->cycleid || P_ISDELETED(opaque))
1399 : {
1400 : /* Done with current scanblkno (and all lower split pages) */
1401 0 : _bt_relbuf(rel, buf);
1402 0 : return scanblkno;
1403 : }
1404 : }
1405 :
1406 23426 : if (!opaque || BTPageIsRecyclable(page, heaprel))
1407 : {
1408 : /* Okay to recycle this page (which could be leaf or internal) */
1409 1160 : RecordFreeIndexPage(rel, blkno);
1410 1160 : stats->pages_deleted++;
1411 1160 : stats->pages_free++;
1412 : }
1413 22266 : else if (P_ISDELETED(opaque))
1414 : {
1415 : /*
1416 : * Already deleted page (which could be leaf or internal). Can't
1417 : * recycle yet.
1418 : */
1419 208 : stats->pages_deleted++;
1420 : }
1421 22058 : else if (P_ISHALFDEAD(opaque))
1422 : {
1423 : /* Half-dead leaf page (from interrupted VACUUM) -- finish deleting */
1424 0 : attempt_pagedel = true;
1425 :
1426 : /*
1427 : * _bt_pagedel() will increment both pages_newly_deleted and
1428 : * pages_deleted stats in all cases (barring corruption)
1429 : */
1430 : }
1431 22058 : else if (P_ISLEAF(opaque))
1432 : {
1433 : OffsetNumber deletable[MaxIndexTuplesPerPage];
1434 : int ndeletable;
1435 : BTVacuumPosting updatable[MaxIndexTuplesPerPage];
1436 : int nupdatable;
1437 : OffsetNumber offnum,
1438 : minoff,
1439 : maxoff;
1440 : int nhtidsdead,
1441 : nhtidslive;
1442 :
1443 : /*
1444 : * Trade in the initial read lock for a full cleanup lock on this
1445 : * page. We must get such a lock on every leaf page over the course
1446 : * of the vacuum scan, whether or not it actually contains any
1447 : * deletable tuples --- see nbtree/README.
1448 : */
1449 20522 : _bt_upgradelockbufcleanup(rel, buf);
1450 :
1451 : /*
1452 : * Check whether we need to backtrack to earlier pages. What we are
1453 : * concerned about is a page split that happened since we started the
1454 : * vacuum scan. If the split moved tuples on the right half of the
1455 : * split (i.e. the tuples that sort high) to a block that we already
1456 : * passed over, then we might have missed the tuples. We need to
1457 : * backtrack now. (Must do this before possibly clearing btpo_cycleid
1458 : * or deleting scanblkno page below!)
1459 : */
1460 20522 : if (vstate->cycleid != 0 &&
1461 20318 : opaque->btpo_cycleid == vstate->cycleid &&
1462 0 : !(opaque->btpo_flags & BTP_SPLIT_END) &&
1463 0 : !P_RIGHTMOST(opaque) &&
1464 0 : opaque->btpo_next < scanblkno)
1465 0 : backtrack_to = opaque->btpo_next;
1466 :
1467 20522 : ndeletable = 0;
1468 20522 : nupdatable = 0;
1469 20522 : minoff = P_FIRSTDATAKEY(opaque);
1470 20522 : maxoff = PageGetMaxOffsetNumber(page);
1471 20522 : nhtidsdead = 0;
1472 20522 : nhtidslive = 0;
1473 20522 : if (callback)
1474 : {
1475 : /* btbulkdelete callback tells us what to delete (or update) */
1476 4008568 : for (offnum = minoff;
1477 : offnum <= maxoff;
1478 3988250 : offnum = OffsetNumberNext(offnum))
1479 : {
1480 : IndexTuple itup;
1481 :
1482 3988250 : itup = (IndexTuple) PageGetItem(page,
1483 3988250 : PageGetItemId(page, offnum));
1484 :
1485 : Assert(!BTreeTupleIsPivot(itup));
1486 3988250 : if (!BTreeTupleIsPosting(itup))
1487 : {
1488 : /* Regular tuple, standard table TID representation */
1489 3872114 : if (callback(&itup->t_tid, callback_state))
1490 : {
1491 1392800 : deletable[ndeletable++] = offnum;
1492 1392800 : nhtidsdead++;
1493 : }
1494 : else
1495 2479314 : nhtidslive++;
1496 : }
1497 : else
1498 : {
1499 : BTVacuumPosting vacposting;
1500 : int nremaining;
1501 :
1502 : /* Posting list tuple */
1503 116136 : vacposting = btreevacuumposting(vstate, itup, offnum,
1504 : &nremaining);
1505 116136 : if (vacposting == NULL)
1506 : {
1507 : /*
1508 : * All table TIDs from the posting tuple remain, so no
1509 : * delete or update required
1510 : */
1511 : Assert(nremaining == BTreeTupleGetNPosting(itup));
1512 : }
1513 70930 : else if (nremaining > 0)
1514 : {
1515 :
1516 : /*
1517 : * Store metadata about posting list tuple in
1518 : * updatable array for entire page. Existing tuple
1519 : * will be updated during the later call to
1520 : * _bt_delitems_vacuum().
1521 : */
1522 : Assert(nremaining < BTreeTupleGetNPosting(itup));
1523 36266 : updatable[nupdatable++] = vacposting;
1524 36266 : nhtidsdead += BTreeTupleGetNPosting(itup) - nremaining;
1525 : }
1526 : else
1527 : {
1528 : /*
1529 : * All table TIDs from the posting list must be
1530 : * deleted. We'll delete the index tuple completely
1531 : * (no update required).
1532 : */
1533 : Assert(nremaining == 0);
1534 34664 : deletable[ndeletable++] = offnum;
1535 34664 : nhtidsdead += BTreeTupleGetNPosting(itup);
1536 34664 : pfree(vacposting);
1537 : }
1538 :
1539 116136 : nhtidslive += nremaining;
1540 : }
1541 : }
1542 : }
1543 :
1544 : /*
1545 : * Apply any needed deletes or updates. We issue just one
1546 : * _bt_delitems_vacuum() call per page, so as to minimize WAL traffic.
1547 : */
1548 20522 : if (ndeletable > 0 || nupdatable > 0)
1549 : {
1550 : Assert(nhtidsdead >= ndeletable + nupdatable);
1551 13180 : _bt_delitems_vacuum(rel, buf, deletable, ndeletable, updatable,
1552 : nupdatable);
1553 :
1554 13180 : stats->tuples_removed += nhtidsdead;
1555 : /* must recompute maxoff */
1556 13180 : maxoff = PageGetMaxOffsetNumber(page);
1557 :
1558 : /* can't leak memory here */
1559 49446 : for (int i = 0; i < nupdatable; i++)
1560 36266 : pfree(updatable[i]);
1561 : }
1562 : else
1563 : {
1564 : /*
1565 : * If the leaf page has been split during this vacuum cycle, it
1566 : * seems worth expending a write to clear btpo_cycleid even if we
1567 : * don't have any deletions to do. (If we do, _bt_delitems_vacuum
1568 : * takes care of this.) This ensures we won't process the page
1569 : * again.
1570 : *
1571 : * We treat this like a hint-bit update because there's no need to
1572 : * WAL-log it.
1573 : */
1574 : Assert(nhtidsdead == 0);
1575 7342 : if (vstate->cycleid != 0 &&
1576 7138 : opaque->btpo_cycleid == vstate->cycleid)
1577 : {
1578 0 : opaque->btpo_cycleid = 0;
1579 0 : MarkBufferDirtyHint(buf, true);
1580 : }
1581 : }
1582 :
1583 : /*
1584 : * If the leaf page is now empty, try to delete it; else count the
1585 : * live tuples (live table TIDs in posting lists are counted as
1586 : * separate live tuples). We don't delete when backtracking, though,
1587 : * since that would require teaching _bt_pagedel() about backtracking
1588 : * (doesn't seem worth adding more complexity to deal with that).
1589 : *
1590 : * We don't count the number of live TIDs during cleanup-only calls to
1591 : * btvacuumscan (i.e. when callback is not set). We count the number
1592 : * of index tuples directly instead. This avoids the expense of
1593 : * directly examining all of the tuples on each page. VACUUM will
1594 : * treat num_index_tuples as an estimate in cleanup-only case, so it
1595 : * doesn't matter that this underestimates num_index_tuples
1596 : * significantly in some cases.
1597 : */
1598 20522 : if (minoff > maxoff)
1599 5602 : attempt_pagedel = (blkno == scanblkno);
1600 14920 : else if (callback)
1601 14724 : stats->num_index_tuples += nhtidslive;
1602 : else
1603 196 : stats->num_index_tuples += maxoff - minoff + 1;
1604 :
1605 : Assert(!attempt_pagedel || nhtidslive == 0);
1606 : }
1607 :
1608 23426 : if (attempt_pagedel)
1609 : {
1610 : MemoryContext oldcontext;
1611 :
1612 : /* Run pagedel in a temp context to avoid memory leakage */
1613 5602 : MemoryContextReset(vstate->pagedelcontext);
1614 5602 : oldcontext = MemoryContextSwitchTo(vstate->pagedelcontext);
1615 :
1616 : /*
1617 : * _bt_pagedel maintains the bulk delete stats on our behalf;
1618 : * pages_newly_deleted and pages_deleted are likely to be incremented
1619 : * during call
1620 : */
1621 : Assert(blkno == scanblkno);
1622 5602 : _bt_pagedel(rel, buf, vstate);
1623 :
1624 5602 : MemoryContextSwitchTo(oldcontext);
1625 : /* pagedel released buffer, so we shouldn't */
1626 : }
1627 : else
1628 17824 : _bt_relbuf(rel, buf);
1629 :
1630 23426 : if (backtrack_to != P_NONE)
1631 : {
1632 0 : blkno = backtrack_to;
1633 :
1634 : /* check for vacuum delay while not holding any buffer lock */
1635 0 : vacuum_delay_point(false);
1636 :
1637 : /*
1638 : * We can't use _bt_getbuf() here because it always applies
1639 : * _bt_checkpage(), which will barf on an all-zero page. We want to
1640 : * recycle all-zero pages, not fail. Also, we want to use a
1641 : * nondefault buffer access strategy.
1642 : */
1643 0 : buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_NORMAL,
1644 : info->strategy);
1645 0 : goto backtrack;
1646 : }
1647 :
1648 23426 : return scanblkno;
1649 : }
1650 :
1651 : /*
1652 : * btreevacuumposting --- determine TIDs still needed in posting list
1653 : *
1654 : * Returns metadata describing how to build replacement tuple without the TIDs
1655 : * that VACUUM needs to delete. Returned value is NULL in the common case
1656 : * where no changes are needed to caller's posting list tuple (we avoid
1657 : * allocating memory here as an optimization).
1658 : *
1659 : * The number of TIDs that should remain in the posting list tuple is set for
1660 : * caller in *nremaining.
1661 : */
1662 : static BTVacuumPosting
1663 116136 : btreevacuumposting(BTVacState *vstate, IndexTuple posting,
1664 : OffsetNumber updatedoffset, int *nremaining)
1665 : {
1666 116136 : int live = 0;
1667 116136 : int nitem = BTreeTupleGetNPosting(posting);
1668 116136 : ItemPointer items = BTreeTupleGetPosting(posting);
1669 116136 : BTVacuumPosting vacposting = NULL;
1670 :
1671 667870 : for (int i = 0; i < nitem; i++)
1672 : {
1673 551734 : if (!vstate->callback(items + i, vstate->callback_state))
1674 : {
1675 : /* Live table TID */
1676 311820 : live++;
1677 : }
1678 239914 : else if (vacposting == NULL)
1679 : {
1680 : /*
1681 : * First dead table TID encountered.
1682 : *
1683 : * It's now clear that we need to delete one or more dead table
1684 : * TIDs, so start maintaining metadata describing how to update
1685 : * existing posting list tuple.
1686 : */
1687 70930 : vacposting = palloc(offsetof(BTVacuumPostingData, deletetids) +
1688 : nitem * sizeof(uint16));
1689 :
1690 70930 : vacposting->itup = posting;
1691 70930 : vacposting->updatedoffset = updatedoffset;
1692 70930 : vacposting->ndeletedtids = 0;
1693 70930 : vacposting->deletetids[vacposting->ndeletedtids++] = i;
1694 : }
1695 : else
1696 : {
1697 : /* Second or subsequent dead table TID */
1698 168984 : vacposting->deletetids[vacposting->ndeletedtids++] = i;
1699 : }
1700 : }
1701 :
1702 116136 : *nremaining = live;
1703 116136 : return vacposting;
1704 : }
1705 :
1706 : /*
1707 : * btcanreturn() -- Check whether btree indexes support index-only scans.
1708 : *
1709 : * btrees always do, so this is trivial.
1710 : */
1711 : bool
1712 1128926 : btcanreturn(Relation index, int attno)
1713 : {
1714 1128926 : return true;
1715 : }
1716 :
1717 : /*
1718 : * btgettreeheight() -- Compute tree height for use by btcostestimate().
1719 : */
1720 : int
1721 720490 : btgettreeheight(Relation rel)
1722 : {
1723 720490 : return _bt_getrootheight(rel);
1724 : }
1725 :
1726 : CompareType
1727 0 : bttranslatestrategy(StrategyNumber strategy, Oid opfamily)
1728 : {
1729 0 : switch (strategy)
1730 : {
1731 0 : case BTLessStrategyNumber:
1732 0 : return COMPARE_LT;
1733 0 : case BTLessEqualStrategyNumber:
1734 0 : return COMPARE_LE;
1735 0 : case BTEqualStrategyNumber:
1736 0 : return COMPARE_EQ;
1737 0 : case BTGreaterEqualStrategyNumber:
1738 0 : return COMPARE_GE;
1739 0 : case BTGreaterStrategyNumber:
1740 0 : return COMPARE_GT;
1741 0 : default:
1742 0 : return COMPARE_INVALID;
1743 : }
1744 : }
1745 :
1746 : StrategyNumber
1747 0 : bttranslatecmptype(CompareType cmptype, Oid opfamily)
1748 : {
1749 0 : switch (cmptype)
1750 : {
1751 0 : case COMPARE_LT:
1752 0 : return BTLessStrategyNumber;
1753 0 : case COMPARE_LE:
1754 0 : return BTLessEqualStrategyNumber;
1755 0 : case COMPARE_EQ:
1756 0 : return BTEqualStrategyNumber;
1757 0 : case COMPARE_GE:
1758 0 : return BTGreaterEqualStrategyNumber;
1759 0 : case COMPARE_GT:
1760 0 : return BTGreaterStrategyNumber;
1761 0 : default:
1762 0 : return InvalidStrategy;
1763 : }
1764 : }
|
