PostgreSQL Source Code  git master
visibilitymap.c
Go to the documentation of this file.
1 /*-------------------------------------------------------------------------
2  *
3  * visibilitymap.c
4  * bitmap for tracking visibility of heap tuples
5  *
6  * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group
7  * Portions Copyright (c) 1994, Regents of the University of California
8  *
9  *
10  * IDENTIFICATION
11  * src/backend/access/heap/visibilitymap.c
12  *
13  * INTERFACE ROUTINES
14  * visibilitymap_clear - clear bits for one page in the visibility map
15  * visibilitymap_pin - pin a map page for setting a bit
16  * visibilitymap_pin_ok - check whether correct map page is already pinned
17  * visibilitymap_set - set a bit in a previously pinned page
18  * visibilitymap_get_status - get status of bits
19  * visibilitymap_count - count number of bits set in visibility map
20  * visibilitymap_prepare_truncate -
21  * prepare for truncation of the visibility map
22  *
23  * NOTES
24  *
25  * The visibility map is a bitmap with two bits (all-visible and all-frozen)
26  * per heap page. A set all-visible bit means that all tuples on the page are
27  * known visible to all transactions, and therefore the page doesn't need to
28  * be vacuumed. A set all-frozen bit means that all tuples on the page are
29  * completely frozen, and therefore the page doesn't need to be vacuumed even
30  * if whole table scanning vacuum is required (e.g. anti-wraparound vacuum).
31  * The all-frozen bit must be set only when the page is already all-visible.
32  *
33  * The map is conservative in the sense that we make sure that whenever a bit
34  * is set, we know the condition is true, but if a bit is not set, it might or
35  * might not be true.
36  *
37  * Clearing visibility map bits is not separately WAL-logged. The callers
38  * must make sure that whenever a bit is cleared, the bit is cleared on WAL
39  * replay of the updating operation as well.
40  *
41  * When we *set* a visibility map during VACUUM, we must write WAL. This may
42  * seem counterintuitive, since the bit is basically a hint: if it is clear,
43  * it may still be the case that every tuple on the page is visible to all
44  * transactions; we just don't know that for certain. The difficulty is that
45  * there are two bits which are typically set together: the PD_ALL_VISIBLE bit
46  * on the page itself, and the visibility map bit. If a crash occurs after the
47  * visibility map page makes it to disk and before the updated heap page makes
48  * it to disk, redo must set the bit on the heap page. Otherwise, the next
49  * insert, update, or delete on the heap page will fail to realize that the
50  * visibility map bit must be cleared, possibly causing index-only scans to
51  * return wrong answers.
52  *
53  * VACUUM will normally skip pages for which the visibility map bit is set;
54  * such pages can't contain any dead tuples and therefore don't need vacuuming.
55  *
56  * LOCKING
57  *
58  * In heapam.c, whenever a page is modified so that not all tuples on the
59  * page are visible to everyone anymore, the corresponding bit in the
60  * visibility map is cleared. In order to be crash-safe, we need to do this
61  * while still holding a lock on the heap page and in the same critical
62  * section that logs the page modification. However, we don't want to hold
63  * the buffer lock over any I/O that may be required to read in the visibility
64  * map page. To avoid this, we examine the heap page before locking it;
65  * if the page-level PD_ALL_VISIBLE bit is set, we pin the visibility map
66  * bit. Then, we lock the buffer. But this creates a race condition: there
67  * is a possibility that in the time it takes to lock the buffer, the
68  * PD_ALL_VISIBLE bit gets set. If that happens, we have to unlock the
69  * buffer, pin the visibility map page, and relock the buffer. This shouldn't
70  * happen often, because only VACUUM currently sets visibility map bits,
71  * and the race will only occur if VACUUM processes a given page at almost
72  * exactly the same time that someone tries to further modify it.
73  *
74  * To set a bit, you need to hold a lock on the heap page. That prevents
75  * the race condition where VACUUM sees that all tuples on the page are
76  * visible to everyone, but another backend modifies the page before VACUUM
77  * sets the bit in the visibility map.
78  *
79  * When a bit is set, the LSN of the visibility map page is updated to make
80  * sure that the visibility map update doesn't get written to disk before the
81  * WAL record of the changes that made it possible to set the bit is flushed.
82  * But when a bit is cleared, we don't have to do that because it's always
83  * safe to clear a bit in the map from correctness point of view.
84  *
85  *-------------------------------------------------------------------------
86  */
87 #include "postgres.h"
88 
89 #include "access/heapam_xlog.h"
90 #include "access/visibilitymap.h"
91 #include "access/xloginsert.h"
92 #include "access/xlogutils.h"
93 #include "miscadmin.h"
94 #include "port/pg_bitutils.h"
95 #include "storage/bufmgr.h"
96 #include "storage/lmgr.h"
97 #include "storage/smgr.h"
98 #include "utils/inval.h"
99 
100 
101 /*#define TRACE_VISIBILITYMAP */
102 
103 /*
104  * Size of the bitmap on each visibility map page, in bytes. There's no
105  * extra headers, so the whole page minus the standard page header is
106  * used for the bitmap.
107  */
108 #define MAPSIZE (BLCKSZ - MAXALIGN(SizeOfPageHeaderData))
109 
110 /* Number of heap blocks we can represent in one byte */
111 #define HEAPBLOCKS_PER_BYTE (BITS_PER_BYTE / BITS_PER_HEAPBLOCK)
112 
113 /* Number of heap blocks we can represent in one visibility map page. */
114 #define HEAPBLOCKS_PER_PAGE (MAPSIZE * HEAPBLOCKS_PER_BYTE)
115 
116 /* Mapping from heap block number to the right bit in the visibility map */
117 #define HEAPBLK_TO_MAPBLOCK(x) ((x) / HEAPBLOCKS_PER_PAGE)
118 #define HEAPBLK_TO_MAPBYTE(x) (((x) % HEAPBLOCKS_PER_PAGE) / HEAPBLOCKS_PER_BYTE)
119 #define HEAPBLK_TO_OFFSET(x) (((x) % HEAPBLOCKS_PER_BYTE) * BITS_PER_HEAPBLOCK)
120 
121 /* Masks for counting subsets of bits in the visibility map. */
122 #define VISIBLE_MASK64 UINT64CONST(0x5555555555555555) /* The lower bit of each
123  * bit pair */
124 #define FROZEN_MASK64 UINT64CONST(0xaaaaaaaaaaaaaaaa) /* The upper bit of each
125  * bit pair */
126 
127 /* prototypes for internal routines */
128 static Buffer vm_readbuf(Relation rel, BlockNumber blkno, bool extend);
129 static void vm_extend(Relation rel, BlockNumber vm_nblocks);
130 
131 
132 /*
133  * visibilitymap_clear - clear specified bits for one page in visibility map
134  *
135  * You must pass a buffer containing the correct map page to this function.
136  * Call visibilitymap_pin first to pin the right one. This function doesn't do
137  * any I/O. Returns true if any bits have been cleared and false otherwise.
138  */
139 bool
140 visibilitymap_clear(Relation rel, BlockNumber heapBlk, Buffer vmbuf, uint8 flags)
141 {
142  BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
143  int mapByte = HEAPBLK_TO_MAPBYTE(heapBlk);
144  int mapOffset = HEAPBLK_TO_OFFSET(heapBlk);
145  uint8 mask = flags << mapOffset;
146  char *map;
147  bool cleared = false;
148 
150 
151 #ifdef TRACE_VISIBILITYMAP
152  elog(DEBUG1, "vm_clear %s %d", RelationGetRelationName(rel), heapBlk);
153 #endif
154 
155  if (!BufferIsValid(vmbuf) || BufferGetBlockNumber(vmbuf) != mapBlock)
156  elog(ERROR, "wrong buffer passed to visibilitymap_clear");
157 
159  map = PageGetContents(BufferGetPage(vmbuf));
160 
161  if (map[mapByte] & mask)
162  {
163  map[mapByte] &= ~mask;
164 
165  MarkBufferDirty(vmbuf);
166  cleared = true;
167  }
168 
170 
171  return cleared;
172 }
173 
174 /*
175  * visibilitymap_pin - pin a map page for setting a bit
176  *
177  * Setting a bit in the visibility map is a two-phase operation. First, call
178  * visibilitymap_pin, to pin the visibility map page containing the bit for
179  * the heap page. Because that can require I/O to read the map page, you
180  * shouldn't hold a lock on the heap page while doing that. Then, call
181  * visibilitymap_set to actually set the bit.
182  *
183  * On entry, *vmbuf should be InvalidBuffer or a valid buffer returned by
184  * an earlier call to visibilitymap_pin or visibilitymap_get_status on the same
185  * relation. On return, *vmbuf is a valid buffer with the map page containing
186  * the bit for heapBlk.
187  *
188  * If the page doesn't exist in the map file yet, it is extended.
189  */
190 void
191 visibilitymap_pin(Relation rel, BlockNumber heapBlk, Buffer *vmbuf)
192 {
193  BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
194 
195  /* Reuse the old pinned buffer if possible */
196  if (BufferIsValid(*vmbuf))
197  {
198  if (BufferGetBlockNumber(*vmbuf) == mapBlock)
199  return;
200 
201  ReleaseBuffer(*vmbuf);
202  }
203  *vmbuf = vm_readbuf(rel, mapBlock, true);
204 }
205 
206 /*
207  * visibilitymap_pin_ok - do we already have the correct page pinned?
208  *
209  * On entry, vmbuf should be InvalidBuffer or a valid buffer returned by
210  * an earlier call to visibilitymap_pin or visibilitymap_get_status on the same
211  * relation. The return value indicates whether the buffer covers the
212  * given heapBlk.
213  */
214 bool
216 {
217  BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
218 
219  return BufferIsValid(vmbuf) && BufferGetBlockNumber(vmbuf) == mapBlock;
220 }
221 
222 /*
223  * visibilitymap_set - set bit(s) on a previously pinned page
224  *
225  * recptr is the LSN of the XLOG record we're replaying, if we're in recovery,
226  * or InvalidXLogRecPtr in normal running. The VM page LSN is advanced to the
227  * one provided; in normal running, we generate a new XLOG record and set the
228  * page LSN to that value (though the heap page's LSN may *not* be updated;
229  * see below). cutoff_xid is the largest xmin on the page being marked
230  * all-visible; it is needed for Hot Standby, and can be InvalidTransactionId
231  * if the page contains no tuples. It can also be set to InvalidTransactionId
232  * when a page that is already all-visible is being marked all-frozen.
233  *
234  * Caller is expected to set the heap page's PD_ALL_VISIBLE bit before calling
235  * this function. Except in recovery, caller should also pass the heap
236  * buffer. When checksums are enabled and we're not in recovery, we must add
237  * the heap buffer to the WAL chain to protect it from being torn.
238  *
239  * You must pass a buffer containing the correct map page to this function.
240  * Call visibilitymap_pin first to pin the right one. This function doesn't do
241  * any I/O.
242  */
243 void
244 visibilitymap_set(Relation rel, BlockNumber heapBlk, Buffer heapBuf,
245  XLogRecPtr recptr, Buffer vmBuf, TransactionId cutoff_xid,
246  uint8 flags)
247 {
248  BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
249  uint32 mapByte = HEAPBLK_TO_MAPBYTE(heapBlk);
250  uint8 mapOffset = HEAPBLK_TO_OFFSET(heapBlk);
251  Page page;
252  uint8 *map;
253 
254 #ifdef TRACE_VISIBILITYMAP
255  elog(DEBUG1, "vm_set %s %d", RelationGetRelationName(rel), heapBlk);
256 #endif
257 
259  Assert(InRecovery || BufferIsValid(heapBuf));
261 
262  /* Check that we have the right heap page pinned, if present */
263  if (BufferIsValid(heapBuf) && BufferGetBlockNumber(heapBuf) != heapBlk)
264  elog(ERROR, "wrong heap buffer passed to visibilitymap_set");
265 
266  /* Check that we have the right VM page pinned */
267  if (!BufferIsValid(vmBuf) || BufferGetBlockNumber(vmBuf) != mapBlock)
268  elog(ERROR, "wrong VM buffer passed to visibilitymap_set");
269 
270  page = BufferGetPage(vmBuf);
271  map = (uint8 *) PageGetContents(page);
273 
274  if (flags != (map[mapByte] >> mapOffset & VISIBILITYMAP_VALID_BITS))
275  {
277 
278  map[mapByte] |= (flags << mapOffset);
279  MarkBufferDirty(vmBuf);
280 
281  if (RelationNeedsWAL(rel))
282  {
283  if (XLogRecPtrIsInvalid(recptr))
284  {
285  Assert(!InRecovery);
286  recptr = log_heap_visible(rel->rd_locator, heapBuf, vmBuf,
287  cutoff_xid, flags);
288 
289  /*
290  * If data checksums are enabled (or wal_log_hints=on), we
291  * need to protect the heap page from being torn.
292  *
293  * If not, then we must *not* update the heap page's LSN. In
294  * this case, the FPI for the heap page was omitted from the
295  * WAL record inserted above, so it would be incorrect to
296  * update the heap page's LSN.
297  */
298  if (XLogHintBitIsNeeded())
299  {
300  Page heapPage = BufferGetPage(heapBuf);
301 
302  /* caller is expected to set PD_ALL_VISIBLE first */
303  Assert(PageIsAllVisible(heapPage));
304  PageSetLSN(heapPage, recptr);
305  }
306  }
307  PageSetLSN(page, recptr);
308  }
309 
311  }
312 
314 }
315 
316 /*
317  * visibilitymap_get_status - get status of bits
318  *
319  * Are all tuples on heapBlk visible to all or are marked frozen, according
320  * to the visibility map?
321  *
322  * On entry, *vmbuf should be InvalidBuffer or a valid buffer returned by an
323  * earlier call to visibilitymap_pin or visibilitymap_get_status on the same
324  * relation. On return, *vmbuf is a valid buffer with the map page containing
325  * the bit for heapBlk, or InvalidBuffer. The caller is responsible for
326  * releasing *vmbuf after it's done testing and setting bits.
327  *
328  * NOTE: This function is typically called without a lock on the heap page,
329  * so somebody else could change the bit just after we look at it. In fact,
330  * since we don't lock the visibility map page either, it's even possible that
331  * someone else could have changed the bit just before we look at it, but yet
332  * we might see the old value. It is the caller's responsibility to deal with
333  * all concurrency issues!
334  */
335 uint8
337 {
338  BlockNumber mapBlock = HEAPBLK_TO_MAPBLOCK(heapBlk);
339  uint32 mapByte = HEAPBLK_TO_MAPBYTE(heapBlk);
340  uint8 mapOffset = HEAPBLK_TO_OFFSET(heapBlk);
341  char *map;
342  uint8 result;
343 
344 #ifdef TRACE_VISIBILITYMAP
345  elog(DEBUG1, "vm_get_status %s %d", RelationGetRelationName(rel), heapBlk);
346 #endif
347 
348  /* Reuse the old pinned buffer if possible */
349  if (BufferIsValid(*vmbuf))
350  {
351  if (BufferGetBlockNumber(*vmbuf) != mapBlock)
352  {
353  ReleaseBuffer(*vmbuf);
354  *vmbuf = InvalidBuffer;
355  }
356  }
357 
358  if (!BufferIsValid(*vmbuf))
359  {
360  *vmbuf = vm_readbuf(rel, mapBlock, false);
361  if (!BufferIsValid(*vmbuf))
362  return false;
363  }
364 
365  map = PageGetContents(BufferGetPage(*vmbuf));
366 
367  /*
368  * A single byte read is atomic. There could be memory-ordering effects
369  * here, but for performance reasons we make it the caller's job to worry
370  * about that.
371  */
372  result = ((map[mapByte] >> mapOffset) & VISIBILITYMAP_VALID_BITS);
373  return result;
374 }
375 
376 /*
377  * visibilitymap_count - count number of bits set in visibility map
378  *
379  * Note: we ignore the possibility of race conditions when the table is being
380  * extended concurrently with the call. New pages added to the table aren't
381  * going to be marked all-visible or all-frozen, so they won't affect the result.
382  */
383 void
384 visibilitymap_count(Relation rel, BlockNumber *all_visible, BlockNumber *all_frozen)
385 {
386  BlockNumber mapBlock;
387  BlockNumber nvisible = 0;
388  BlockNumber nfrozen = 0;
389 
390  /* all_visible must be specified */
391  Assert(all_visible);
392 
393  for (mapBlock = 0;; mapBlock++)
394  {
395  Buffer mapBuffer;
396  uint64 *map;
397  int i;
398 
399  /*
400  * Read till we fall off the end of the map. We assume that any extra
401  * bytes in the last page are zeroed, so we don't bother excluding
402  * them from the count.
403  */
404  mapBuffer = vm_readbuf(rel, mapBlock, false);
405  if (!BufferIsValid(mapBuffer))
406  break;
407 
408  /*
409  * We choose not to lock the page, since the result is going to be
410  * immediately stale anyway if anyone is concurrently setting or
411  * clearing bits, and we only really need an approximate value.
412  */
413  map = (uint64 *) PageGetContents(BufferGetPage(mapBuffer));
414 
415  StaticAssertStmt(MAPSIZE % sizeof(uint64) == 0,
416  "unsupported MAPSIZE");
417  if (all_frozen == NULL)
418  {
419  for (i = 0; i < MAPSIZE / sizeof(uint64); i++)
420  nvisible += pg_popcount64(map[i] & VISIBLE_MASK64);
421  }
422  else
423  {
424  for (i = 0; i < MAPSIZE / sizeof(uint64); i++)
425  {
426  nvisible += pg_popcount64(map[i] & VISIBLE_MASK64);
427  nfrozen += pg_popcount64(map[i] & FROZEN_MASK64);
428  }
429  }
430 
431  ReleaseBuffer(mapBuffer);
432  }
433 
434  *all_visible = nvisible;
435  if (all_frozen)
436  *all_frozen = nfrozen;
437 }
438 
439 /*
440  * visibilitymap_prepare_truncate -
441  * prepare for truncation of the visibility map
442  *
443  * nheapblocks is the new size of the heap.
444  *
445  * Return the number of blocks of new visibility map.
446  * If it's InvalidBlockNumber, there is nothing to truncate;
447  * otherwise the caller is responsible for calling smgrtruncate()
448  * to truncate the visibility map pages.
449  */
452 {
453  BlockNumber newnblocks;
454 
455  /* last remaining block, byte, and bit */
456  BlockNumber truncBlock = HEAPBLK_TO_MAPBLOCK(nheapblocks);
457  uint32 truncByte = HEAPBLK_TO_MAPBYTE(nheapblocks);
458  uint8 truncOffset = HEAPBLK_TO_OFFSET(nheapblocks);
459 
460 #ifdef TRACE_VISIBILITYMAP
461  elog(DEBUG1, "vm_truncate %s %d", RelationGetRelationName(rel), nheapblocks);
462 #endif
463 
464  /*
465  * If no visibility map has been created yet for this relation, there's
466  * nothing to truncate.
467  */
469  return InvalidBlockNumber;
470 
471  /*
472  * Unless the new size is exactly at a visibility map page boundary, the
473  * tail bits in the last remaining map page, representing truncated heap
474  * blocks, need to be cleared. This is not only tidy, but also necessary
475  * because we don't get a chance to clear the bits if the heap is extended
476  * again.
477  */
478  if (truncByte != 0 || truncOffset != 0)
479  {
480  Buffer mapBuffer;
481  Page page;
482  char *map;
483 
484  newnblocks = truncBlock + 1;
485 
486  mapBuffer = vm_readbuf(rel, truncBlock, false);
487  if (!BufferIsValid(mapBuffer))
488  {
489  /* nothing to do, the file was already smaller */
490  return InvalidBlockNumber;
491  }
492 
493  page = BufferGetPage(mapBuffer);
494  map = PageGetContents(page);
495 
496  LockBuffer(mapBuffer, BUFFER_LOCK_EXCLUSIVE);
497 
498  /* NO EREPORT(ERROR) from here till changes are logged */
500 
501  /* Clear out the unwanted bytes. */
502  MemSet(&map[truncByte + 1], 0, MAPSIZE - (truncByte + 1));
503 
504  /*----
505  * Mask out the unwanted bits of the last remaining byte.
506  *
507  * ((1 << 0) - 1) = 00000000
508  * ((1 << 1) - 1) = 00000001
509  * ...
510  * ((1 << 6) - 1) = 00111111
511  * ((1 << 7) - 1) = 01111111
512  *----
513  */
514  map[truncByte] &= (1 << truncOffset) - 1;
515 
516  /*
517  * Truncation of a relation is WAL-logged at a higher-level, and we
518  * will be called at WAL replay. But if checksums are enabled, we need
519  * to still write a WAL record to protect against a torn page, if the
520  * page is flushed to disk before the truncation WAL record. We cannot
521  * use MarkBufferDirtyHint here, because that will not dirty the page
522  * during recovery.
523  */
524  MarkBufferDirty(mapBuffer);
526  log_newpage_buffer(mapBuffer, false);
527 
529 
530  UnlockReleaseBuffer(mapBuffer);
531  }
532  else
533  newnblocks = truncBlock;
534 
535  if (smgrnblocks(RelationGetSmgr(rel), VISIBILITYMAP_FORKNUM) <= newnblocks)
536  {
537  /* nothing to do, the file was already smaller than requested size */
538  return InvalidBlockNumber;
539  }
540 
541  return newnblocks;
542 }
543 
544 /*
545  * Read a visibility map page.
546  *
547  * If the page doesn't exist, InvalidBuffer is returned, or if 'extend' is
548  * true, the visibility map file is extended.
549  */
550 static Buffer
551 vm_readbuf(Relation rel, BlockNumber blkno, bool extend)
552 {
553  Buffer buf;
554  SMgrRelation reln;
555 
556  /*
557  * Caution: re-using this smgr pointer could fail if the relcache entry
558  * gets closed. It's safe as long as we only do smgr-level operations
559  * between here and the last use of the pointer.
560  */
561  reln = RelationGetSmgr(rel);
562 
563  /*
564  * If we haven't cached the size of the visibility map fork yet, check it
565  * first.
566  */
568  {
571  else
573  }
574 
575  /* Handle requests beyond EOF */
576  if (blkno >= reln->smgr_cached_nblocks[VISIBILITYMAP_FORKNUM])
577  {
578  if (extend)
579  vm_extend(rel, blkno + 1);
580  else
581  return InvalidBuffer;
582  }
583 
584  /*
585  * Use ZERO_ON_ERROR mode, and initialize the page if necessary. It's
586  * always safe to clear bits, so it's better to clear corrupt pages than
587  * error out.
588  *
589  * The initialize-the-page part is trickier than it looks, because of the
590  * possibility of multiple backends doing this concurrently, and our
591  * desire to not uselessly take the buffer lock in the normal path where
592  * the page is OK. We must take the lock to initialize the page, so
593  * recheck page newness after we have the lock, in case someone else
594  * already did it. Also, because we initially check PageIsNew with no
595  * lock, it's possible to fall through and return the buffer while someone
596  * else is still initializing the page (i.e., we might see pd_upper as set
597  * but other page header fields are still zeroes). This is harmless for
598  * callers that will take a buffer lock themselves, but some callers
599  * inspect the page without any lock at all. The latter is OK only so
600  * long as it doesn't depend on the page header having correct contents.
601  * Current usage is safe because PageGetContents() does not require that.
602  */
604  RBM_ZERO_ON_ERROR, NULL);
606  {
609  PageInit(BufferGetPage(buf), BLCKSZ, 0);
611  }
612  return buf;
613 }
614 
615 /*
616  * Ensure that the visibility map fork is at least vm_nblocks long, extending
617  * it if necessary with zeroed pages.
618  */
619 static void
620 vm_extend(Relation rel, BlockNumber vm_nblocks)
621 {
622  BlockNumber vm_nblocks_now;
623  PGAlignedBlock pg;
624  SMgrRelation reln;
625 
626  PageInit((Page) pg.data, BLCKSZ, 0);
627 
628  /*
629  * We use the relation extension lock to lock out other backends trying to
630  * extend the visibility map at the same time. It also locks out extension
631  * of the main fork, unnecessarily, but extending the visibility map
632  * happens seldom enough that it doesn't seem worthwhile to have a
633  * separate lock tag type for it.
634  *
635  * Note that another backend might have extended or created the relation
636  * by the time we get the lock.
637  */
639 
640  /*
641  * Caution: re-using this smgr pointer could fail if the relcache entry
642  * gets closed. It's safe as long as we only do smgr-level operations
643  * between here and the last use of the pointer.
644  */
645  reln = RelationGetSmgr(rel);
646 
647  /*
648  * Create the file first if it doesn't exist. If smgr_vm_nblocks is
649  * positive then it must exist, no need for an smgrexists call.
650  */
651  if ((reln->smgr_cached_nblocks[VISIBILITYMAP_FORKNUM] == 0 ||
654  smgrcreate(reln, VISIBILITYMAP_FORKNUM, false);
655 
656  /* Invalidate cache so that smgrnblocks() asks the kernel. */
658  vm_nblocks_now = smgrnblocks(reln, VISIBILITYMAP_FORKNUM);
659 
660  /* Now extend the file */
661  while (vm_nblocks_now < vm_nblocks)
662  {
663  PageSetChecksumInplace((Page) pg.data, vm_nblocks_now);
664 
665  smgrextend(reln, VISIBILITYMAP_FORKNUM, vm_nblocks_now, pg.data, false);
666  vm_nblocks_now++;
667  }
668 
669  /*
670  * Send a shared-inval message to force other backends to close any smgr
671  * references they may have for this rel, which we are about to change.
672  * This is a useful optimization because it means that backends don't have
673  * to keep checking for creation or extension of the file, which happens
674  * infrequently.
675  */
677 
679 }
uint32 BlockNumber
Definition: block.h:31
#define InvalidBlockNumber
Definition: block.h:33
int Buffer
Definition: buf.h:23
#define InvalidBuffer
Definition: buf.h:25
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:2763
void ReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:3931
void UnlockReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:3954
void MarkBufferDirty(Buffer buffer)
Definition: bufmgr.c:1583
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:4172
Buffer ReadBufferExtended(Relation reln, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy)
Definition: bufmgr.c:759
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:105
static Page BufferGetPage(Buffer buffer)
Definition: bufmgr.h:280
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:107
@ RBM_ZERO_ON_ERROR
Definition: bufmgr.h:44
static bool BufferIsValid(Buffer bufnum)
Definition: bufmgr.h:228
void PageSetChecksumInplace(Page page, BlockNumber blkno)
Definition: bufpage.c:1539
void PageInit(Page page, Size pageSize, Size specialSize)
Definition: bufpage.c:42
static char * PageGetContents(Page page)
Definition: bufpage.h:254
Pointer Page
Definition: bufpage.h:78
static bool PageIsNew(Page page)
Definition: bufpage.h:230
static bool PageIsAllVisible(Page page)
Definition: bufpage.h:426
static void PageSetLSN(Page page, XLogRecPtr lsn)
Definition: bufpage.h:388
unsigned int uint32
Definition: c.h:442
unsigned char uint8
Definition: c.h:440
#define MemSet(start, val, len)
Definition: c.h:953
#define StaticAssertStmt(condition, errmessage)
Definition: c.h:869
uint32 TransactionId
Definition: c.h:588
#define DEBUG1
Definition: elog.h:26
#define ERROR
Definition: elog.h:35
XLogRecPtr log_heap_visible(RelFileLocator rlocator, Buffer heap_buffer, Buffer vm_buffer, TransactionId snapshotConflictHorizon, uint8 vmflags)
Definition: heapam.c:8251
void CacheInvalidateSmgr(RelFileLocatorBackend rlocator)
Definition: inval.c:1462
int i
Definition: isn.c:73
Assert(fmt[strlen(fmt) - 1] !='\n')
void LockRelationForExtension(Relation relation, LOCKMODE lockmode)
Definition: lmgr.c:431
void UnlockRelationForExtension(Relation relation, LOCKMODE lockmode)
Definition: lmgr.c:481
#define ExclusiveLock
Definition: lockdefs.h:42
#define START_CRIT_SECTION()
Definition: miscadmin.h:148
#define END_CRIT_SECTION()
Definition: miscadmin.h:150
int pg_popcount64(uint64 word)
Definition: pg_bitutils.c:284
static char * buf
Definition: pg_test_fsync.c:67
static SMgrRelation RelationGetSmgr(Relation rel)
Definition: rel.h:569
#define RelationGetRelationName(relation)
Definition: rel.h:535
#define RelationNeedsWAL(relation)
Definition: rel.h:626
@ VISIBILITYMAP_FORKNUM
Definition: relpath.h:52
BlockNumber smgrnblocks(SMgrRelation reln, ForkNumber forknum)
Definition: smgr.c:579
void smgrextend(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
Definition: smgr.c:493
void smgrcreate(SMgrRelation reln, ForkNumber forknum, bool isRedo)
Definition: smgr.c:369
bool smgrexists(SMgrRelation reln, ForkNumber forknum)
Definition: smgr.c:247
RelFileLocator rd_locator
Definition: rel.h:56
BlockNumber smgr_cached_nblocks[MAX_FORKNUM+1]
Definition: smgr.h:54
RelFileLocatorBackend smgr_rlocator
Definition: smgr.h:42
char data[BLCKSZ]
Definition: c.h:1079
static void vm_extend(Relation rel, BlockNumber vm_nblocks)
#define MAPSIZE
#define VISIBLE_MASK64
void visibilitymap_set(Relation rel, BlockNumber heapBlk, Buffer heapBuf, XLogRecPtr recptr, Buffer vmBuf, TransactionId cutoff_xid, uint8 flags)
bool visibilitymap_pin_ok(BlockNumber heapBlk, Buffer vmbuf)
bool visibilitymap_clear(Relation rel, BlockNumber heapBlk, Buffer vmbuf, uint8 flags)
#define HEAPBLK_TO_OFFSET(x)
void visibilitymap_pin(Relation rel, BlockNumber heapBlk, Buffer *vmbuf)
uint8 visibilitymap_get_status(Relation rel, BlockNumber heapBlk, Buffer *vmbuf)
BlockNumber visibilitymap_prepare_truncate(Relation rel, BlockNumber nheapblocks)
void visibilitymap_count(Relation rel, BlockNumber *all_visible, BlockNumber *all_frozen)
static Buffer vm_readbuf(Relation rel, BlockNumber blkno, bool extend)
#define FROZEN_MASK64
#define HEAPBLK_TO_MAPBLOCK(x)
#define HEAPBLK_TO_MAPBYTE(x)
#define VISIBILITYMAP_VALID_BITS
#define XLogHintBitIsNeeded()
Definition: xlog.h:115
#define XLogRecPtrIsInvalid(r)
Definition: xlogdefs.h:29
uint64 XLogRecPtr
Definition: xlogdefs.h:21
XLogRecPtr log_newpage_buffer(Buffer buffer, bool page_std)
Definition: xloginsert.c:1191
bool InRecovery
Definition: xlogutils.c:53