PostgreSQL Source Code  git master
gistutil.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * gistutil.c
4  * utilities routines for the postgres GiST index access method.
5  *
6  *
7  * Portions Copyright (c) 1996-2020, PostgreSQL Global Development Group
8  * Portions Copyright (c) 1994, Regents of the University of California
9  *
10  * IDENTIFICATION
11  * src/backend/access/gist/gistutil.c
12  *-------------------------------------------------------------------------
13  */
14 #include "postgres.h"
15 
16 #include <math.h>
17 
18 #include "access/gist_private.h"
19 #include "access/htup_details.h"
20 #include "access/reloptions.h"
21 #include "catalog/pg_opclass.h"
22 #include "storage/indexfsm.h"
23 #include "storage/lmgr.h"
24 #include "utils/float.h"
25 #include "utils/lsyscache.h"
26 #include "utils/snapmgr.h"
27 #include "utils/syscache.h"
28 
29 /*
30  * Write itup vector to page, has no control of free space.
31  */
32 void
33 gistfillbuffer(Page page, IndexTuple *itup, int len, OffsetNumber off)
34 {
36  int i;
37 
38  if (off == InvalidOffsetNumber)
39  off = (PageIsEmpty(page)) ? FirstOffsetNumber :
41 
42  for (i = 0; i < len; i++)
43  {
44  Size sz = IndexTupleSize(itup[i]);
45 
46  l = PageAddItem(page, (Item) itup[i], sz, off, false, false);
47  if (l == InvalidOffsetNumber)
48  elog(ERROR, "failed to add item to GiST index page, item %d out of %d, size %d bytes",
49  i, len, (int) sz);
50  off++;
51  }
52 }
53 
54 /*
55  * Check space for itup vector on page
56  */
57 bool
58 gistnospace(Page page, IndexTuple *itvec, int len, OffsetNumber todelete, Size freespace)
59 {
60  unsigned int size = freespace,
61  deleted = 0;
62  int i;
63 
64  for (i = 0; i < len; i++)
65  size += IndexTupleSize(itvec[i]) + sizeof(ItemIdData);
66 
67  if (todelete != InvalidOffsetNumber)
68  {
69  IndexTuple itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, todelete));
70 
71  deleted = IndexTupleSize(itup) + sizeof(ItemIdData);
72  }
73 
74  return (PageGetFreeSpace(page) + deleted < size);
75 }
76 
77 bool
78 gistfitpage(IndexTuple *itvec, int len)
79 {
80  int i;
81  Size size = 0;
82 
83  for (i = 0; i < len; i++)
84  size += IndexTupleSize(itvec[i]) + sizeof(ItemIdData);
85 
86  /* TODO: Consider fillfactor */
87  return (size <= GiSTPageSize);
88 }
89 
90 /*
91  * Read buffer into itup vector
92  */
93 IndexTuple *
94 gistextractpage(Page page, int *len /* out */ )
95 {
97  maxoff;
98  IndexTuple *itvec;
99 
100  maxoff = PageGetMaxOffsetNumber(page);
101  *len = maxoff;
102  itvec = palloc(sizeof(IndexTuple) * maxoff);
103  for (i = FirstOffsetNumber; i <= maxoff; i = OffsetNumberNext(i))
104  itvec[i - FirstOffsetNumber] = (IndexTuple) PageGetItem(page, PageGetItemId(page, i));
105 
106  return itvec;
107 }
108 
109 /*
110  * join two vectors into one
111  */
112 IndexTuple *
113 gistjoinvector(IndexTuple *itvec, int *len, IndexTuple *additvec, int addlen)
114 {
115  itvec = (IndexTuple *) repalloc((void *) itvec, sizeof(IndexTuple) * ((*len) + addlen));
116  memmove(&itvec[*len], additvec, sizeof(IndexTuple) * addlen);
117  *len += addlen;
118  return itvec;
119 }
120 
121 /*
122  * make plain IndexTuple vector
123  */
124 
126 gistfillitupvec(IndexTuple *vec, int veclen, int *memlen)
127 {
128  char *ptr,
129  *ret;
130  int i;
131 
132  *memlen = 0;
133 
134  for (i = 0; i < veclen; i++)
135  *memlen += IndexTupleSize(vec[i]);
136 
137  ptr = ret = palloc(*memlen);
138 
139  for (i = 0; i < veclen; i++)
140  {
141  memcpy(ptr, vec[i], IndexTupleSize(vec[i]));
142  ptr += IndexTupleSize(vec[i]);
143  }
144 
145  return (IndexTupleData *) ret;
146 }
147 
148 /*
149  * Make unions of keys in IndexTuple vector (one union datum per index column).
150  * Union Datums are returned into the attr/isnull arrays.
151  * Resulting Datums aren't compressed.
152  */
153 void
154 gistMakeUnionItVec(GISTSTATE *giststate, IndexTuple *itvec, int len,
155  Datum *attr, bool *isnull)
156 {
157  int i;
158  GistEntryVector *evec;
159  int attrsize;
160 
161  evec = (GistEntryVector *) palloc((len + 2) * sizeof(GISTENTRY) + GEVHDRSZ);
162 
163  for (i = 0; i < giststate->nonLeafTupdesc->natts; i++)
164  {
165  int j;
166 
167  /* Collect non-null datums for this column */
168  evec->n = 0;
169  for (j = 0; j < len; j++)
170  {
171  Datum datum;
172  bool IsNull;
173 
174  datum = index_getattr(itvec[j], i + 1, giststate->leafTupdesc,
175  &IsNull);
176  if (IsNull)
177  continue;
178 
179  gistdentryinit(giststate, i,
180  evec->vector + evec->n,
181  datum,
182  NULL, NULL, (OffsetNumber) 0,
183  false, IsNull);
184  evec->n++;
185  }
186 
187  /* If this column was all NULLs, the union is NULL */
188  if (evec->n == 0)
189  {
190  attr[i] = (Datum) 0;
191  isnull[i] = true;
192  }
193  else
194  {
195  if (evec->n == 1)
196  {
197  /* unionFn may expect at least two inputs */
198  evec->n = 2;
199  evec->vector[1] = evec->vector[0];
200  }
201 
202  /* Make union and store in attr array */
203  attr[i] = FunctionCall2Coll(&giststate->unionFn[i],
204  giststate->supportCollation[i],
205  PointerGetDatum(evec),
206  PointerGetDatum(&attrsize));
207 
208  isnull[i] = false;
209  }
210  }
211 }
212 
213 /*
214  * Return an IndexTuple containing the result of applying the "union"
215  * method to the specified IndexTuple vector.
216  */
218 gistunion(Relation r, IndexTuple *itvec, int len, GISTSTATE *giststate)
219 {
220  Datum attr[INDEX_MAX_KEYS];
221  bool isnull[INDEX_MAX_KEYS];
222 
223  gistMakeUnionItVec(giststate, itvec, len, attr, isnull);
224 
225  return gistFormTuple(giststate, r, attr, isnull, false);
226 }
227 
228 /*
229  * makes union of two key
230  */
231 void
232 gistMakeUnionKey(GISTSTATE *giststate, int attno,
233  GISTENTRY *entry1, bool isnull1,
234  GISTENTRY *entry2, bool isnull2,
235  Datum *dst, bool *dstisnull)
236 {
237  /* we need a GistEntryVector with room for exactly 2 elements */
238  union
239  {
240  GistEntryVector gev;
241  char padding[2 * sizeof(GISTENTRY) + GEVHDRSZ];
242  } storage;
243  GistEntryVector *evec = &storage.gev;
244  int dstsize;
245 
246  evec->n = 2;
247 
248  if (isnull1 && isnull2)
249  {
250  *dstisnull = true;
251  *dst = (Datum) 0;
252  }
253  else
254  {
255  if (isnull1 == false && isnull2 == false)
256  {
257  evec->vector[0] = *entry1;
258  evec->vector[1] = *entry2;
259  }
260  else if (isnull1 == false)
261  {
262  evec->vector[0] = *entry1;
263  evec->vector[1] = *entry1;
264  }
265  else
266  {
267  evec->vector[0] = *entry2;
268  evec->vector[1] = *entry2;
269  }
270 
271  *dstisnull = false;
272  *dst = FunctionCall2Coll(&giststate->unionFn[attno],
273  giststate->supportCollation[attno],
274  PointerGetDatum(evec),
275  PointerGetDatum(&dstsize));
276  }
277 }
278 
279 bool
280 gistKeyIsEQ(GISTSTATE *giststate, int attno, Datum a, Datum b)
281 {
282  bool result;
283 
284  FunctionCall3Coll(&giststate->equalFn[attno],
285  giststate->supportCollation[attno],
286  a, b,
287  PointerGetDatum(&result));
288  return result;
289 }
290 
291 /*
292  * Decompress all keys in tuple
293  */
294 void
296  OffsetNumber o, GISTENTRY *attdata, bool *isnull)
297 {
298  int i;
299 
300  for (i = 0; i < IndexRelationGetNumberOfKeyAttributes(r); i++)
301  {
302  Datum datum;
303 
304  datum = index_getattr(tuple, i + 1, giststate->leafTupdesc, &isnull[i]);
305  gistdentryinit(giststate, i, &attdata[i],
306  datum, r, p, o,
307  false, isnull[i]);
308  }
309 }
310 
311 /*
312  * Forms union of oldtup and addtup, if union == oldtup then return NULL
313  */
316 {
317  bool neednew = false;
318  GISTENTRY oldentries[INDEX_MAX_KEYS],
319  addentries[INDEX_MAX_KEYS];
320  bool oldisnull[INDEX_MAX_KEYS],
321  addisnull[INDEX_MAX_KEYS];
322  Datum attr[INDEX_MAX_KEYS];
323  bool isnull[INDEX_MAX_KEYS];
324  IndexTuple newtup = NULL;
325  int i;
326 
327  gistDeCompressAtt(giststate, r, oldtup, NULL,
328  (OffsetNumber) 0, oldentries, oldisnull);
329 
330  gistDeCompressAtt(giststate, r, addtup, NULL,
331  (OffsetNumber) 0, addentries, addisnull);
332 
333  for (i = 0; i < IndexRelationGetNumberOfKeyAttributes(r); i++)
334  {
335  gistMakeUnionKey(giststate, i,
336  oldentries + i, oldisnull[i],
337  addentries + i, addisnull[i],
338  attr + i, isnull + i);
339 
340  if (neednew)
341  /* we already need new key, so we can skip check */
342  continue;
343 
344  if (isnull[i])
345  /* union of key may be NULL if and only if both keys are NULL */
346  continue;
347 
348  if (!addisnull[i])
349  {
350  if (oldisnull[i] ||
351  !gistKeyIsEQ(giststate, i, oldentries[i].key, attr[i]))
352  neednew = true;
353  }
354  }
355 
356  if (neednew)
357  {
358  /* need to update key */
359  newtup = gistFormTuple(giststate, r, attr, isnull, false);
360  newtup->t_tid = oldtup->t_tid;
361  }
362 
363  return newtup;
364 }
365 
366 /*
367  * Search an upper index page for the entry with lowest penalty for insertion
368  * of the new index key contained in "it".
369  *
370  * Returns the index of the page entry to insert into.
371  */
373 gistchoose(Relation r, Page p, IndexTuple it, /* it has compressed entry */
374  GISTSTATE *giststate)
375 {
376  OffsetNumber result;
377  OffsetNumber maxoff;
378  OffsetNumber i;
379  float best_penalty[INDEX_MAX_KEYS];
380  GISTENTRY entry,
381  identry[INDEX_MAX_KEYS];
382  bool isnull[INDEX_MAX_KEYS];
383  int keep_current_best;
384 
385  Assert(!GistPageIsLeaf(p));
386 
387  gistDeCompressAtt(giststate, r,
388  it, NULL, (OffsetNumber) 0,
389  identry, isnull);
390 
391  /* we'll return FirstOffsetNumber if page is empty (shouldn't happen) */
392  result = FirstOffsetNumber;
393 
394  /*
395  * The index may have multiple columns, and there's a penalty value for
396  * each column. The penalty associated with a column that appears earlier
397  * in the index definition is strictly more important than the penalty of
398  * a column that appears later in the index definition.
399  *
400  * best_penalty[j] is the best penalty we have seen so far for column j,
401  * or -1 when we haven't yet examined column j. Array entries to the
402  * right of the first -1 are undefined.
403  */
404  best_penalty[0] = -1;
405 
406  /*
407  * If we find a tuple that's exactly as good as the currently best one, we
408  * could use either one. When inserting a lot of tuples with the same or
409  * similar keys, it's preferable to descend down the same path when
410  * possible, as that's more cache-friendly. On the other hand, if all
411  * inserts land on the same leaf page after a split, we're never going to
412  * insert anything to the other half of the split, and will end up using
413  * only 50% of the available space. Distributing the inserts evenly would
414  * lead to better space usage, but that hurts cache-locality during
415  * insertion. To get the best of both worlds, when we find a tuple that's
416  * exactly as good as the previous best, choose randomly whether to stick
417  * to the old best, or use the new one. Once we decide to stick to the
418  * old best, we keep sticking to it for any subsequent equally good tuples
419  * we might find. This favors tuples with low offsets, but still allows
420  * some inserts to go to other equally-good subtrees.
421  *
422  * keep_current_best is -1 if we haven't yet had to make a random choice
423  * whether to keep the current best tuple. If we have done so, and
424  * decided to keep it, keep_current_best is 1; if we've decided to
425  * replace, keep_current_best is 0. (This state will be reset to -1 as
426  * soon as we've made the replacement, but sometimes we make the choice in
427  * advance of actually finding a replacement best tuple.)
428  */
429  keep_current_best = -1;
430 
431  /*
432  * Loop over tuples on page.
433  */
434  maxoff = PageGetMaxOffsetNumber(p);
435  Assert(maxoff >= FirstOffsetNumber);
436 
437  for (i = FirstOffsetNumber; i <= maxoff; i = OffsetNumberNext(i))
438  {
439  IndexTuple itup = (IndexTuple) PageGetItem(p, PageGetItemId(p, i));
440  bool zero_penalty;
441  int j;
442 
443  zero_penalty = true;
444 
445  /* Loop over index attributes. */
446  for (j = 0; j < IndexRelationGetNumberOfKeyAttributes(r); j++)
447  {
448  Datum datum;
449  float usize;
450  bool IsNull;
451 
452  /* Compute penalty for this column. */
453  datum = index_getattr(itup, j + 1, giststate->leafTupdesc,
454  &IsNull);
455  gistdentryinit(giststate, j, &entry, datum, r, p, i,
456  false, IsNull);
457  usize = gistpenalty(giststate, j, &entry, IsNull,
458  &identry[j], isnull[j]);
459  if (usize > 0)
460  zero_penalty = false;
461 
462  if (best_penalty[j] < 0 || usize < best_penalty[j])
463  {
464  /*
465  * New best penalty for column. Tentatively select this tuple
466  * as the target, and record the best penalty. Then reset the
467  * next column's penalty to "unknown" (and indirectly, the
468  * same for all the ones to its right). This will force us to
469  * adopt this tuple's penalty values as the best for all the
470  * remaining columns during subsequent loop iterations.
471  */
472  result = i;
473  best_penalty[j] = usize;
474 
476  best_penalty[j + 1] = -1;
477 
478  /* we have new best, so reset keep-it decision */
479  keep_current_best = -1;
480  }
481  else if (best_penalty[j] == usize)
482  {
483  /*
484  * The current tuple is exactly as good for this column as the
485  * best tuple seen so far. The next iteration of this loop
486  * will compare the next column.
487  */
488  }
489  else
490  {
491  /*
492  * The current tuple is worse for this column than the best
493  * tuple seen so far. Skip the remaining columns and move on
494  * to the next tuple, if any.
495  */
496  zero_penalty = false; /* so outer loop won't exit */
497  break;
498  }
499  }
500 
501  /*
502  * If we looped past the last column, and did not update "result",
503  * then this tuple is exactly as good as the prior best tuple.
504  */
505  if (j == IndexRelationGetNumberOfKeyAttributes(r) && result != i)
506  {
507  if (keep_current_best == -1)
508  {
509  /* we didn't make the random choice yet for this old best */
510  keep_current_best = (random() <= (MAX_RANDOM_VALUE / 2)) ? 1 : 0;
511  }
512  if (keep_current_best == 0)
513  {
514  /* we choose to use the new tuple */
515  result = i;
516  /* choose again if there are even more exactly-as-good ones */
517  keep_current_best = -1;
518  }
519  }
520 
521  /*
522  * If we find a tuple with zero penalty for all columns, and we've
523  * decided we don't want to search for another tuple with equal
524  * penalty, there's no need to examine remaining tuples; just break
525  * out of the loop and return it.
526  */
527  if (zero_penalty)
528  {
529  if (keep_current_best == -1)
530  {
531  /* we didn't make the random choice yet for this old best */
532  keep_current_best = (random() <= (MAX_RANDOM_VALUE / 2)) ? 1 : 0;
533  }
534  if (keep_current_best == 1)
535  break;
536  }
537  }
538 
539  return result;
540 }
541 
542 /*
543  * initialize a GiST entry with a decompressed version of key
544  */
545 void
546 gistdentryinit(GISTSTATE *giststate, int nkey, GISTENTRY *e,
547  Datum k, Relation r, Page pg, OffsetNumber o,
548  bool l, bool isNull)
549 {
550  if (!isNull)
551  {
552  GISTENTRY *dep;
553 
554  gistentryinit(*e, k, r, pg, o, l);
555 
556  /* there may not be a decompress function in opclass */
557  if (!OidIsValid(giststate->decompressFn[nkey].fn_oid))
558  return;
559 
560  dep = (GISTENTRY *)
562  giststate->supportCollation[nkey],
563  PointerGetDatum(e)));
564  /* decompressFn may just return the given pointer */
565  if (dep != e)
566  gistentryinit(*e, dep->key, dep->rel, dep->page, dep->offset,
567  dep->leafkey);
568  }
569  else
570  gistentryinit(*e, (Datum) 0, r, pg, o, l);
571 }
572 
575  Datum attdata[], bool isnull[], bool isleaf)
576 {
577  Datum compatt[INDEX_MAX_KEYS];
578  int i;
579  IndexTuple res;
580 
581  /*
582  * Call the compress method on each attribute.
583  */
584  for (i = 0; i < IndexRelationGetNumberOfKeyAttributes(r); i++)
585  {
586  if (isnull[i])
587  compatt[i] = (Datum) 0;
588  else
589  {
590  GISTENTRY centry;
591  GISTENTRY *cep;
592 
593  gistentryinit(centry, attdata[i], r, NULL, (OffsetNumber) 0,
594  isleaf);
595  /* there may not be a compress function in opclass */
596  if (OidIsValid(giststate->compressFn[i].fn_oid))
597  cep = (GISTENTRY *)
599  giststate->supportCollation[i],
600  PointerGetDatum(&centry)));
601  else
602  cep = &centry;
603  compatt[i] = cep->key;
604  }
605  }
606 
607  if (isleaf)
608  {
609  /*
610  * Emplace each included attribute if any.
611  */
612  for (; i < r->rd_att->natts; i++)
613  {
614  if (isnull[i])
615  compatt[i] = (Datum) 0;
616  else
617  compatt[i] = attdata[i];
618  }
619  }
620 
621  res = index_form_tuple(isleaf ? giststate->leafTupdesc :
622  giststate->nonLeafTupdesc,
623  compatt, isnull);
624 
625  /*
626  * The offset number on tuples on internal pages is unused. For historical
627  * reasons, it is set to 0xffff.
628  */
629  ItemPointerSetOffsetNumber(&(res->t_tid), 0xffff);
630  return res;
631 }
632 
633 /*
634  * initialize a GiST entry with fetched value in key field
635  */
636 static Datum
637 gistFetchAtt(GISTSTATE *giststate, int nkey, Datum k, Relation r)
638 {
639  GISTENTRY fentry;
640  GISTENTRY *fep;
641 
642  gistentryinit(fentry, k, r, NULL, (OffsetNumber) 0, false);
643 
644  fep = (GISTENTRY *)
645  DatumGetPointer(FunctionCall1Coll(&giststate->fetchFn[nkey],
646  giststate->supportCollation[nkey],
647  PointerGetDatum(&fentry)));
648 
649  /* fetchFn set 'key', return it to the caller */
650  return fep->key;
651 }
652 
653 /*
654  * Fetch all keys in tuple.
655  * Returns a new HeapTuple containing the originally-indexed data.
656  */
657 HeapTuple
659 {
660  MemoryContext oldcxt = MemoryContextSwitchTo(giststate->tempCxt);
662  bool isnull[INDEX_MAX_KEYS];
663  int i;
664 
665  for (i = 0; i < IndexRelationGetNumberOfKeyAttributes(r); i++)
666  {
667  Datum datum;
668 
669  datum = index_getattr(tuple, i + 1, giststate->leafTupdesc, &isnull[i]);
670 
671  if (giststate->fetchFn[i].fn_oid != InvalidOid)
672  {
673  if (!isnull[i])
674  fetchatt[i] = gistFetchAtt(giststate, i, datum, r);
675  else
676  fetchatt[i] = (Datum) 0;
677  }
678  else if (giststate->compressFn[i].fn_oid == InvalidOid)
679  {
680  /*
681  * If opclass does not provide compress method that could change
682  * original value, att is necessarily stored in original form.
683  */
684  if (!isnull[i])
685  fetchatt[i] = datum;
686  else
687  fetchatt[i] = (Datum) 0;
688  }
689  else
690  {
691  /*
692  * Index-only scans not supported for this column. Since the
693  * planner chose an index-only scan anyway, it is not interested
694  * in this column, and we can replace it with a NULL.
695  */
696  isnull[i] = true;
697  fetchatt[i] = (Datum) 0;
698  }
699  }
700 
701  /*
702  * Get each included attribute.
703  */
704  for (; i < r->rd_att->natts; i++)
705  {
706  fetchatt[i] = index_getattr(tuple, i + 1, giststate->leafTupdesc,
707  &isnull[i]);
708  }
709  MemoryContextSwitchTo(oldcxt);
710 
711  return heap_form_tuple(giststate->fetchTupdesc, fetchatt, isnull);
712 }
713 
714 float
715 gistpenalty(GISTSTATE *giststate, int attno,
716  GISTENTRY *orig, bool isNullOrig,
717  GISTENTRY *add, bool isNullAdd)
718 {
719  float penalty = 0.0;
720 
721  if (giststate->penaltyFn[attno].fn_strict == false ||
722  (isNullOrig == false && isNullAdd == false))
723  {
724  FunctionCall3Coll(&giststate->penaltyFn[attno],
725  giststate->supportCollation[attno],
726  PointerGetDatum(orig),
727  PointerGetDatum(add),
728  PointerGetDatum(&penalty));
729  /* disallow negative or NaN penalty */
730  if (isnan(penalty) || penalty < 0.0)
731  penalty = 0.0;
732  }
733  else if (isNullOrig && isNullAdd)
734  penalty = 0.0;
735  else
736  {
737  /* try to prevent mixing null and non-null values */
738  penalty = get_float4_infinity();
739  }
740 
741  return penalty;
742 }
743 
744 /*
745  * Initialize a new index page
746  */
747 void
749 {
750  GISTPageOpaque opaque;
751  Page page;
752  Size pageSize;
753 
754  pageSize = BufferGetPageSize(b);
755  page = BufferGetPage(b);
756  PageInit(page, pageSize, sizeof(GISTPageOpaqueData));
757 
758  opaque = GistPageGetOpaque(page);
759  /* page was already zeroed by PageInit, so this is not needed: */
760  /* memset(&(opaque->nsn), 0, sizeof(GistNSN)); */
761  opaque->rightlink = InvalidBlockNumber;
762  opaque->flags = f;
763  opaque->gist_page_id = GIST_PAGE_ID;
764 }
765 
766 /*
767  * Verify that a freshly-read page looks sane.
768  */
769 void
771 {
772  Page page = BufferGetPage(buf);
773 
774  /*
775  * ReadBuffer verifies that every newly-read page passes
776  * PageHeaderIsValid, which means it either contains a reasonably sane
777  * page header or is all-zero. We have to defend against the all-zero
778  * case, however.
779  */
780  if (PageIsNew(page))
781  ereport(ERROR,
782  (errcode(ERRCODE_INDEX_CORRUPTED),
783  errmsg("index \"%s\" contains unexpected zero page at block %u",
785  BufferGetBlockNumber(buf)),
786  errhint("Please REINDEX it.")));
787 
788  /*
789  * Additionally check that the special area looks sane.
790  */
791  if (PageGetSpecialSize(page) != MAXALIGN(sizeof(GISTPageOpaqueData)))
792  ereport(ERROR,
793  (errcode(ERRCODE_INDEX_CORRUPTED),
794  errmsg("index \"%s\" contains corrupted page at block %u",
796  BufferGetBlockNumber(buf)),
797  errhint("Please REINDEX it.")));
798 }
799 
800 
801 /*
802  * Allocate a new page (either by recycling, or by extending the index file)
803  *
804  * The returned buffer is already pinned and exclusive-locked
805  *
806  * Caller is responsible for initializing the page by calling GISTInitBuffer
807  */
808 Buffer
810 {
811  Buffer buffer;
812  bool needLock;
813 
814  /* First, try to get a page from FSM */
815  for (;;)
816  {
818 
819  if (blkno == InvalidBlockNumber)
820  break; /* nothing left in FSM */
821 
822  buffer = ReadBuffer(r, blkno);
823 
824  /*
825  * We have to guard against the possibility that someone else already
826  * recycled this page; the buffer may be locked if so.
827  */
828  if (ConditionalLockBuffer(buffer))
829  {
830  Page page = BufferGetPage(buffer);
831 
832  /*
833  * If the page was never initialized, it's OK to use.
834  */
835  if (PageIsNew(page))
836  return buffer;
837 
838  gistcheckpage(r, buffer);
839 
840  /*
841  * Otherwise, recycle it if deleted, and too old to have any
842  * processes interested in it.
843  */
844  if (gistPageRecyclable(page))
845  {
846  /*
847  * If we are generating WAL for Hot Standby then create a WAL
848  * record that will allow us to conflict with queries running
849  * on standby, in case they have snapshots older than the
850  * page's deleteXid.
851  */
853  gistXLogPageReuse(r, blkno, GistPageGetDeleteXid(page));
854 
855  return buffer;
856  }
857 
858  LockBuffer(buffer, GIST_UNLOCK);
859  }
860 
861  /* Can't use it, so release buffer and try again */
862  ReleaseBuffer(buffer);
863  }
864 
865  /* Must extend the file */
866  needLock = !RELATION_IS_LOCAL(r);
867 
868  if (needLock)
870 
871  buffer = ReadBuffer(r, P_NEW);
872  LockBuffer(buffer, GIST_EXCLUSIVE);
873 
874  if (needLock)
876 
877  return buffer;
878 }
879 
880 /* Can this page be recycled yet? */
881 bool
883 {
884  if (PageIsNew(page))
885  return true;
886  if (GistPageIsDeleted(page))
887  {
888  /*
889  * The page was deleted, but when? If it was just deleted, a scan
890  * might have seen the downlink to it, and will read the page later.
891  * As long as that can happen, we must keep the deleted page around as
892  * a tombstone.
893  *
894  * Compare the deletion XID with RecentGlobalXmin. If deleteXid <
895  * RecentGlobalXmin, then no scan that's still in progress could have
896  * seen its downlink, and we can recycle it.
897  */
898  FullTransactionId deletexid_full = GistPageGetDeleteXid(page);
899  FullTransactionId recentxmin_full = GetFullRecentGlobalXmin();
900 
901  if (FullTransactionIdPrecedes(deletexid_full, recentxmin_full))
902  return true;
903  }
904  return false;
905 }
906 
907 bytea *
908 gistoptions(Datum reloptions, bool validate)
909 {
910  static const relopt_parse_elt tab[] = {
911  {"fillfactor", RELOPT_TYPE_INT, offsetof(GiSTOptions, fillfactor)},
912  {"buffering", RELOPT_TYPE_ENUM, offsetof(GiSTOptions, buffering_mode)}
913  };
914 
915  return (bytea *) build_reloptions(reloptions, validate,
917  sizeof(GiSTOptions),
918  tab, lengthof(tab));
919 }
920 
921 /*
922  * gistproperty() -- Check boolean properties of indexes.
923  *
924  * This is optional for most AMs, but is required for GiST because the core
925  * property code doesn't support AMPROP_DISTANCE_ORDERABLE. We also handle
926  * AMPROP_RETURNABLE here to save opening the rel to call gistcanreturn.
927  */
928 bool
929 gistproperty(Oid index_oid, int attno,
930  IndexAMProperty prop, const char *propname,
931  bool *res, bool *isnull)
932 {
933  Oid opclass,
934  opfamily,
935  opcintype;
936  int16 procno;
937 
938  /* Only answer column-level inquiries */
939  if (attno == 0)
940  return false;
941 
942  /*
943  * Currently, GiST distance-ordered scans require that there be a distance
944  * function in the opclass with the default types (i.e. the one loaded
945  * into the relcache entry, see initGISTstate). So we assume that if such
946  * a function exists, then there's a reason for it (rather than grubbing
947  * through all the opfamily's operators to find an ordered one).
948  *
949  * Essentially the same code can test whether we support returning the
950  * column data, since that's true if the opclass provides a fetch proc.
951  */
952 
953  switch (prop)
954  {
956  procno = GIST_DISTANCE_PROC;
957  break;
958  case AMPROP_RETURNABLE:
959  procno = GIST_FETCH_PROC;
960  break;
961  default:
962  return false;
963  }
964 
965  /* First we need to know the column's opclass. */
966  opclass = get_index_column_opclass(index_oid, attno);
967  if (!OidIsValid(opclass))
968  {
969  *isnull = true;
970  return true;
971  }
972 
973  /* Now look up the opclass family and input datatype. */
974  if (!get_opclass_opfamily_and_input_type(opclass, &opfamily, &opcintype))
975  {
976  *isnull = true;
977  return true;
978  }
979 
980  /* And now we can check whether the function is provided. */
981 
983  ObjectIdGetDatum(opfamily),
984  ObjectIdGetDatum(opcintype),
985  ObjectIdGetDatum(opcintype),
986  Int16GetDatum(procno));
987 
988  /*
989  * Special case: even without a fetch function, AMPROP_RETURNABLE is true
990  * if the opclass has no compress function.
991  */
992  if (prop == AMPROP_RETURNABLE && !*res)
993  {
995  ObjectIdGetDatum(opfamily),
996  ObjectIdGetDatum(opcintype),
997  ObjectIdGetDatum(opcintype),
999  }
1000 
1001  *isnull = false;
1002 
1003  return true;
1004 }
1005 
1006 /*
1007  * Some indexes are not WAL-logged, but we need LSNs to detect concurrent page
1008  * splits anyway. This function provides a fake sequence of LSNs for that
1009  * purpose.
1010  */
1011 XLogRecPtr
1013 {
1014  if (rel->rd_rel->relpersistence == RELPERSISTENCE_TEMP)
1015  {
1016  /*
1017  * Temporary relations are only accessible in our session, so a simple
1018  * backend-local counter will do.
1019  */
1020  static XLogRecPtr counter = FirstNormalUnloggedLSN;
1021 
1022  return counter++;
1023  }
1024  else if (rel->rd_rel->relpersistence == RELPERSISTENCE_PERMANENT)
1025  {
1026  /*
1027  * WAL-logging on this relation will start after commit, so its LSNs
1028  * must be distinct numbers smaller than the LSN at the next commit.
1029  * Emit a dummy WAL record if insert-LSN hasn't advanced after the
1030  * last call.
1031  */
1032  static XLogRecPtr lastlsn = InvalidXLogRecPtr;
1033  XLogRecPtr currlsn = GetXLogInsertRecPtr();
1034 
1035  /* Shouldn't be called for WAL-logging relations */
1036  Assert(!RelationNeedsWAL(rel));
1037 
1038  /* No need for an actual record if we already have a distinct LSN */
1039  if (!XLogRecPtrIsInvalid(lastlsn) && lastlsn == currlsn)
1040  currlsn = gistXLogAssignLSN();
1041 
1042  lastlsn = currlsn;
1043  return currlsn;
1044  }
1045  else
1046  {
1047  /*
1048  * Unlogged relations are accessible from other backends, and survive
1049  * (clean) restarts. GetFakeLSNForUnloggedRel() handles that for us.
1050  */
1051  Assert(rel->rd_rel->relpersistence == RELPERSISTENCE_UNLOGGED);
1052  return GetFakeLSNForUnloggedRel();
1053  }
1054 }
Relation rel
Definition: gist.h:152
signed short int16
Definition: c.h:361
static Datum gistFetchAtt(GISTSTATE *giststate, int nkey, Datum k, Relation r)
Definition: gistutil.c:637
XLogRecPtr gistXLogAssignLSN(void)
Definition: gistxlog.c:602
#define InvalidXLogRecPtr
Definition: xlogdefs.h:28
#define GistPageIsDeleted(page)
Definition: gist.h:163
#define PageIsEmpty(page)
Definition: bufpage.h:222
int errhint(const char *fmt,...)
Definition: elog.c:1071
#define GIST_FETCH_PROC
Definition: gist.h:38
IndexAMProperty
Definition: amapi.h:34
FmgrInfo fetchFn[INDEX_MAX_KEYS]
Definition: gist_private.h:94
#define ExclusiveLock
Definition: lockdefs.h:44
Oid supportCollation[INDEX_MAX_KEYS]
Definition: gist_private.h:97
#define PointerGetDatum(X)
Definition: postgres.h:556
long random(void)
Definition: random.c:22
FmgrInfo compressFn[INDEX_MAX_KEYS]
Definition: gist_private.h:88
void gistDeCompressAtt(GISTSTATE *giststate, Relation r, IndexTuple tuple, Page p, OffsetNumber o, GISTENTRY *attdata, bool *isnull)
Definition: gistutil.c:295
FmgrInfo equalFn[INDEX_MAX_KEYS]
Definition: gist_private.h:92
#define RELATION_IS_LOCAL(relation)
Definition: rel.h:583
ItemPointerData t_tid
Definition: itup.h:37
void gistfillbuffer(Page page, IndexTuple *itup, int len, OffsetNumber off)
Definition: gistutil.c:33
static MemoryContext MemoryContextSwitchTo(MemoryContext context)
Definition: palloc.h:109
#define Int16GetDatum(X)
Definition: postgres.h:451
Pointer Item
Definition: item.h:17
int32 n
Definition: gist.h:226
#define SearchSysCacheExists4(cacheId, key1, key2, key3, key4)
Definition: syscache.h:189
int errcode(int sqlerrcode)
Definition: elog.c:610
#define PageAddItem(page, item, size, offsetNumber, overwrite, is_heap)
Definition: bufpage.h:416
#define GIST_UNLOCK
Definition: gist_private.h:44
uint32 BlockNumber
Definition: block.h:31
void ReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:3518
#define P_NEW
Definition: bufmgr.h:91
HeapTuple heap_form_tuple(TupleDesc tupleDescriptor, Datum *values, bool *isnull)
Definition: heaptuple.c:1020
#define lengthof(array)
Definition: c.h:675
Datum FunctionCall2Coll(FmgrInfo *flinfo, Oid collation, Datum arg1, Datum arg2)
Definition: fmgr.c:1152
Form_pg_class rd_rel
Definition: rel.h:109
unsigned int Oid
Definition: postgres_ext.h:31
bytea * gistoptions(Datum reloptions, bool validate)
Definition: gistutil.c:908
#define OidIsValid(objectId)
Definition: c.h:651
#define PageGetMaxOffsetNumber(page)
Definition: bufpage.h:357
IndexTuple gistgetadjusted(Relation r, IndexTuple oldtup, IndexTuple addtup, GISTSTATE *giststate)
Definition: gistutil.c:315
#define fetchatt(A, T)
Definition: tupmacs.h:41
XLogRecPtr gistGetFakeLSN(Relation rel)
Definition: gistutil.c:1012
Size PageGetFreeSpace(Page page)
Definition: bufpage.c:574
bool gistKeyIsEQ(GISTSTATE *giststate, int attno, Datum a, Datum b)
Definition: gistutil.c:280
void * build_reloptions(Datum reloptions, bool validate, relopt_kind kind, Size relopt_struct_size, const relopt_parse_elt *relopt_elems, int num_relopt_elems)
Definition: reloptions.c:1887
uint16 OffsetNumber
Definition: off.h:24
Page page
Definition: gist.h:153
GISTENTRY vector[FLEXIBLE_ARRAY_MEMBER]
Definition: gist.h:227
IndexTuple index_form_tuple(TupleDesc tupleDescriptor, Datum *values, bool *isnull)
Definition: indextuple.c:47
bool get_opclass_opfamily_and_input_type(Oid opclass, Oid *opfamily, Oid *opcintype)
Definition: lsyscache.c:1174
IndexTuple * gistextractpage(Page page, int *len)
Definition: gistutil.c:94
uint16 gist_page_id
Definition: gist.h:73
#define GEVHDRSZ
Definition: gist.h:230
float gistpenalty(GISTSTATE *giststate, int attno, GISTENTRY *orig, bool isNullOrig, GISTENTRY *add, bool isNullAdd)
Definition: gistutil.c:715
void gistdentryinit(GISTSTATE *giststate, int nkey, GISTENTRY *e, Datum k, Relation r, Page pg, OffsetNumber o, bool l, bool isNull)
Definition: gistutil.c:546
struct GISTENTRY GISTENTRY
#define ObjectIdGetDatum(X)
Definition: postgres.h:507
#define ERROR
Definition: elog.h:43
bool fn_strict
Definition: fmgr.h:61
#define GIST_COMPRESS_PROC
Definition: gist.h:32
int fillfactor
Definition: pgbench.c:159
XLogRecPtr GetFakeLSNForUnloggedRel(void)
Definition: xlog.c:4946
XLogRecPtr GetXLogInsertRecPtr(void)
Definition: xlog.c:11494
MemoryContext tempCxt
Definition: gist_private.h:78
void gistXLogPageReuse(Relation rel, BlockNumber blkno, FullTransactionId latestRemovedXid)
Definition: gistxlog.c:620
BlockNumber blkno
Definition: ginvacuum.c:119
Datum key
Definition: gist.h:151
static char * buf
Definition: pg_test_fsync.c:67
#define FirstOffsetNumber
Definition: off.h:27
IndexTupleData * IndexTuple
Definition: itup.h:53
#define MAX_RANDOM_VALUE
bool gistnospace(Page page, IndexTuple *itvec, int len, OffsetNumber todelete, Size freespace)
Definition: gistutil.c:58
#define RelationGetRelationName(relation)
Definition: rel.h:490
FmgrInfo penaltyFn[INDEX_MAX_KEYS]
Definition: gist_private.h:90
unsigned int uint32
Definition: c.h:374
struct ItemIdData ItemIdData
uint16 flags
Definition: gist.h:72
#define IndexRelationGetNumberOfKeyAttributes(relation)
Definition: rel.h:475
HeapTuple gistFetchTuple(GISTSTATE *giststate, Relation r, IndexTuple tuple)
Definition: gistutil.c:658
#define BufferGetPage(buffer)
Definition: bufmgr.h:169
FmgrInfo decompressFn[INDEX_MAX_KEYS]
Definition: gist_private.h:89
bool ConditionalLockBuffer(Buffer buffer)
Definition: bufmgr.c:3783
bool leafkey
Definition: gist.h:155
TupleDesc leafTupdesc
Definition: gist_private.h:80
OffsetNumber gistchoose(Relation r, Page p, IndexTuple it, GISTSTATE *giststate)
Definition: gistutil.c:373
#define GistPageIsLeaf(page)
Definition: gist.h:160
#define XLogRecPtrIsInvalid(r)
Definition: xlogdefs.h:29
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:235
void LockRelationForExtension(Relation relation, LOCKMODE lockmode)
Definition: lmgr.c:402
#define GiSTPageSize
Definition: gist_private.h:474
IndexTuple * gistjoinvector(IndexTuple *itvec, int *len, IndexTuple *additvec, int addlen)
Definition: gistutil.c:113
void UnlockRelationForExtension(Relation relation, LOCKMODE lockmode)
Definition: lmgr.c:452
#define FirstNormalUnloggedLSN
Definition: xlogdefs.h:36
uintptr_t Datum
Definition: postgres.h:367
Datum FunctionCall1Coll(FmgrInfo *flinfo, Oid collation, Datum arg1)
Definition: fmgr.c:1132
#define BufferGetPageSize(buffer)
Definition: bufmgr.h:156
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3757
#define XLogStandbyInfoActive()
Definition: xlog.h:205
TupleDesc rd_att
Definition: rel.h:110
#define InvalidOffsetNumber
Definition: off.h:26
#define InvalidOid
Definition: postgres_ext.h:36
TupleDesc nonLeafTupdesc
Definition: gist_private.h:81
Oid fn_oid
Definition: fmgr.h:59
#define ereport(elevel,...)
Definition: elog.h:144
Datum FunctionCall3Coll(FmgrInfo *flinfo, Oid collation, Datum arg1, Datum arg2, Datum arg3)
Definition: fmgr.c:1174
BlockNumber GetFreeIndexPage(Relation rel)
Definition: indexfsm.c:38
bool gistfitpage(IndexTuple *itvec, int len)
Definition: gistutil.c:78
#define GistPageGetOpaque(page)
Definition: gist.h:158
void gistMakeUnionItVec(GISTSTATE *giststate, IndexTuple *itvec, int len, Datum *attr, bool *isnull)
Definition: gistutil.c:154
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:745
#define gistentryinit(e, k, r, pg, o, l)
Definition: gist.h:235
TupleDesc fetchTupdesc
Definition: gist_private.h:83
void gistcheckpage(Relation rel, Buffer buf)
Definition: gistutil.c:770
Buffer ReadBuffer(Relation reln, BlockNumber blockNum)
Definition: bufmgr.c:607
void gistMakeUnionKey(GISTSTATE *giststate, int attno, GISTENTRY *entry1, bool isnull1, GISTENTRY *entry2, bool isnull2, Datum *dst, bool *dstisnull)
Definition: gistutil.c:232
#define INDEX_MAX_KEYS
#define OffsetNumberNext(offsetNumber)
Definition: off.h:52
size_t Size
Definition: c.h:473
IndexTupleData * gistfillitupvec(IndexTuple *vec, int veclen, int *memlen)
Definition: gistutil.c:126
#define InvalidBlockNumber
Definition: block.h:33
#define GIST_DISTANCE_PROC
Definition: gist.h:37
#define MAXALIGN(LEN)
Definition: c.h:698
#define index_getattr(tup, attnum, tupleDesc, isnull)
Definition: itup.h:100
#define RelationNeedsWAL(relation)
Definition: rel.h:562
#define PageGetSpecialSize(page)
Definition: bufpage.h:300
void * repalloc(void *pointer, Size size)
Definition: mcxt.c:1069
#define DatumGetPointer(X)
Definition: postgres.h:549
#define ItemPointerSetOffsetNumber(pointer, offsetNumber)
Definition: itemptr.h:148
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:2668
e
Definition: preproc-init.c:82
static FullTransactionId GistPageGetDeleteXid(Page page)
Definition: gist.h:206
IndexTuple gistunion(Relation r, IndexTuple *itvec, int len, GISTSTATE *giststate)
Definition: gistutil.c:218
bool gistPageRecyclable(Page page)
Definition: gistutil.c:882
Oid get_index_column_opclass(Oid index_oid, int attno)
Definition: lsyscache.c:3287
#define PageIsNew(page)
Definition: bufpage.h:229
void * palloc(Size size)
Definition: mcxt.c:949
int errmsg(const char *fmt,...)
Definition: elog.c:824
#define elog(elevel,...)
Definition: elog.h:214
int i
void GISTInitBuffer(Buffer b, uint32 f)
Definition: gistutil.c:748
IndexTuple gistFormTuple(GISTSTATE *giststate, Relation r, Datum attdata[], bool isnull[], bool isleaf)
Definition: gistutil.c:574
BlockNumber rightlink
Definition: gist.h:71
Definition: c.h:562
#define FullTransactionIdPrecedes(a, b)
Definition: transam.h:51
FmgrInfo unionFn[INDEX_MAX_KEYS]
Definition: gist_private.h:87
#define GIST_EXCLUSIVE
Definition: gist_private.h:43
OffsetNumber offset
Definition: gist.h:154
bool gistproperty(Oid index_oid, int attno, IndexAMProperty prop, const char *propname, bool *res, bool *isnull)
Definition: gistutil.c:929
int Buffer
Definition: buf.h:23
#define GIST_PAGE_ID
Definition: gist.h:102
Buffer gistNewBuffer(Relation r)
Definition: gistutil.c:809
FullTransactionId GetFullRecentGlobalXmin(void)
Definition: snapmgr.c:963
#define offsetof(type, field)
Definition: c.h:668
#define PageGetItem(page, itemId)
Definition: bufpage.h:340
Pointer Page
Definition: bufpage.h:78
#define IndexTupleSize(itup)
Definition: itup.h:71
static float4 get_float4_infinity(void)
Definition: float.h:73
void PageInit(Page page, Size pageSize, Size specialSize)
Definition: bufpage.c:42