PostgreSQL Source Code  git master
hashpage.c File Reference
#include "postgres.h"
#include "access/hash.h"
#include "access/hash_xlog.h"
#include "miscadmin.h"
#include "storage/lmgr.h"
#include "storage/predicate.h"
#include "storage/smgr.h"
Include dependency graph for hashpage.c:

Go to the source code of this file.

Functions

static bool _hash_alloc_buckets (Relation rel, BlockNumber firstblock, uint32 nblocks)
 
static void _hash_splitbucket (Relation rel, Buffer metabuf, Bucket obucket, Bucket nbucket, Buffer obuf, Buffer nbuf, HTAB *htab, uint32 maxbucket, uint32 highmask, uint32 lowmask)
 
static void log_split_page (Relation rel, Buffer buf)
 
Buffer _hash_getbuf (Relation rel, BlockNumber blkno, int access, int flags)
 
Buffer _hash_getbuf_with_condlock_cleanup (Relation rel, BlockNumber blkno, int flags)
 
Buffer _hash_getinitbuf (Relation rel, BlockNumber blkno)
 
void _hash_initbuf (Buffer buf, uint32 max_bucket, uint32 num_bucket, uint32 flag, bool initpage)
 
Buffer _hash_getnewbuf (Relation rel, BlockNumber blkno, ForkNumber forkNum)
 
Buffer _hash_getbuf_with_strategy (Relation rel, BlockNumber blkno, int access, int flags, BufferAccessStrategy bstrategy)
 
void _hash_relbuf (Relation rel, Buffer buf)
 
void _hash_dropbuf (Relation rel, Buffer buf)
 
void _hash_dropscanbuf (Relation rel, HashScanOpaque so)
 
uint32 _hash_init (Relation rel, double num_tuples, ForkNumber forkNum)
 
void _hash_init_metabuffer (Buffer buf, double num_tuples, RegProcedure procid, uint16 ffactor, bool initpage)
 
void _hash_pageinit (Page page, Size size)
 
void _hash_expandtable (Relation rel, Buffer metabuf)
 
void _hash_finish_split (Relation rel, Buffer metabuf, Buffer obuf, Bucket obucket, uint32 maxbucket, uint32 highmask, uint32 lowmask)
 
HashMetaPage _hash_getcachedmetap (Relation rel, Buffer *metabuf, bool force_refresh)
 
Buffer _hash_getbucketbuf_from_hashkey (Relation rel, uint32 hashkey, int access, HashMetaPage *cachedmetap)
 

Function Documentation

◆ _hash_alloc_buckets()

static bool _hash_alloc_buckets ( Relation  rel,
BlockNumber  firstblock,
uint32  nblocks 
)
static

Definition at line 989 of file hashpage.c.

References _hash_pageinit(), PGAlignedBlock::data, HashPageOpaqueData::hasho_bucket, HashPageOpaqueData::hasho_flag, HashPageOpaqueData::hasho_nextblkno, HashPageOpaqueData::hasho_page_id, HASHO_PAGE_ID, HashPageOpaqueData::hasho_prevblkno, InvalidBlockNumber, LH_UNUSED_PAGE, log_newpage(), MAIN_FORKNUM, PageGetSpecialPointer, PageSetChecksumInplace(), RelationData::rd_node, RelationData::rd_smgr, RelationNeedsWAL, RelationOpenSmgr, and smgrextend().

Referenced by _hash_expandtable().

990 {
991  BlockNumber lastblock;
992  PGAlignedBlock zerobuf;
993  Page page;
994  HashPageOpaque ovflopaque;
995 
996  lastblock = firstblock + nblocks - 1;
997 
998  /*
999  * Check for overflow in block number calculation; if so, we cannot extend
1000  * the index anymore.
1001  */
1002  if (lastblock < firstblock || lastblock == InvalidBlockNumber)
1003  return false;
1004 
1005  page = (Page) zerobuf.data;
1006 
1007  /*
1008  * Initialize the page. Just zeroing the page won't work; see
1009  * _hash_freeovflpage for similar usage. We take care to make the special
1010  * space valid for the benefit of tools such as pageinspect.
1011  */
1012  _hash_pageinit(page, BLCKSZ);
1013 
1014  ovflopaque = (HashPageOpaque) PageGetSpecialPointer(page);
1015 
1016  ovflopaque->hasho_prevblkno = InvalidBlockNumber;
1017  ovflopaque->hasho_nextblkno = InvalidBlockNumber;
1018  ovflopaque->hasho_bucket = -1;
1019  ovflopaque->hasho_flag = LH_UNUSED_PAGE;
1020  ovflopaque->hasho_page_id = HASHO_PAGE_ID;
1021 
1022  if (RelationNeedsWAL(rel))
1023  log_newpage(&rel->rd_node,
1024  MAIN_FORKNUM,
1025  lastblock,
1026  zerobuf.data,
1027  true);
1028 
1029  RelationOpenSmgr(rel);
1030  PageSetChecksumInplace(page, lastblock);
1031  smgrextend(rel->rd_smgr, MAIN_FORKNUM, lastblock, zerobuf.data, false);
1032 
1033  return true;
1034 }
uint16 hasho_page_id
Definition: hash.h:82
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:597
struct SMgrRelationData * rd_smgr
Definition: rel.h:56
uint32 BlockNumber
Definition: block.h:31
#define LH_UNUSED_PAGE
Definition: hash.h:52
#define RelationOpenSmgr(relation)
Definition: rel.h:480
char data[BLCKSZ]
Definition: c.h:1085
BlockNumber hasho_prevblkno
Definition: hash.h:78
RelFileNode rd_node
Definition: rel.h:54
Bucket hasho_bucket
Definition: hash.h:80
#define PageGetSpecialPointer(page)
Definition: bufpage.h:326
#define InvalidBlockNumber
Definition: block.h:33
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:85
#define HASHO_PAGE_ID
Definition: hash.h:98
void PageSetChecksumInplace(Page page, BlockNumber blkno)
Definition: bufpage.c:1198
#define RelationNeedsWAL(relation)
Definition: rel.h:525
void smgrextend(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
Definition: smgr.c:483
uint16 hasho_flag
Definition: hash.h:81
XLogRecPtr log_newpage(RelFileNode *rnode, ForkNumber forkNum, BlockNumber blkno, Page page, bool page_std)
Definition: xloginsert.c:972
BlockNumber hasho_nextblkno
Definition: hash.h:79
Pointer Page
Definition: bufpage.h:78

◆ _hash_dropbuf()

void _hash_dropbuf ( Relation  rel,
Buffer  buf 
)

Definition at line 275 of file hashpage.c.

References ReleaseBuffer().

Referenced by _hash_doinsert(), _hash_dropscanbuf(), _hash_expandtable(), _hash_finish_split(), _hash_first(), _hash_getbucketbuf_from_hashkey(), _hash_next(), _hash_readprev(), and hashbulkdelete().

276 {
278 }
void ReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:3365
static char * buf
Definition: pg_test_fsync.c:67

◆ _hash_dropscanbuf()

void _hash_dropscanbuf ( Relation  rel,
HashScanOpaque  so 
)

Definition at line 287 of file hashpage.c.

References _hash_dropbuf(), HashScanPosData::buf, BufferIsValid, HashScanOpaqueData::currPos, HashScanOpaqueData::hashso_buc_populated, HashScanOpaqueData::hashso_buc_split, HashScanOpaqueData::hashso_bucket_buf, HashScanOpaqueData::hashso_split_bucket_buf, and InvalidBuffer.

Referenced by _hash_next(), hashendscan(), and hashrescan().

288 {
289  /* release pin we hold on primary bucket page */
290  if (BufferIsValid(so->hashso_bucket_buf) &&
291  so->hashso_bucket_buf != so->currPos.buf)
294 
295  /* release pin we hold on primary bucket page of bucket being split */
300 
301  /* release any pin we still hold */
302  if (BufferIsValid(so->currPos.buf))
303  _hash_dropbuf(rel, so->currPos.buf);
304  so->currPos.buf = InvalidBuffer;
305 
306  /* reset split scan */
307  so->hashso_buc_populated = false;
308  so->hashso_buc_split = false;
309 }
#define InvalidBuffer
Definition: buf.h:25
void _hash_dropbuf(Relation rel, Buffer buf)
Definition: hashpage.c:275
Buffer hashso_bucket_buf
Definition: hash.h:161
bool hashso_buc_populated
Definition: hash.h:171
Buffer buf
Definition: hash.h:108
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
HashScanPosData currPos
Definition: hash.h:186
bool hashso_buc_split
Definition: hash.h:177
Buffer hashso_split_bucket_buf
Definition: hash.h:168

◆ _hash_expandtable()

void _hash_expandtable ( Relation  rel,
Buffer  metabuf 
)

Definition at line 615 of file hashpage.c.

References _hash_alloc_buckets(), _hash_checkpage(), _hash_dropbuf(), _hash_finish_split(), _hash_get_totalbuckets(), _hash_getbuf_with_condlock_cleanup(), _hash_getnewbuf(), _hash_relbuf(), _hash_spareindex(), _hash_splitbucket(), Assert, BUCKET_TO_BLKNO, BUFFER_LOCK_EXCLUSIVE, BUFFER_LOCK_UNLOCK, BufferGetPage, END_CRIT_SECTION, xl_hash_split_allocate_page::flags, H_BUCKET_BEING_SPLIT, H_NEEDS_SPLIT_CLEANUP, hashbucketcleanup(), HashMetaPageData::hashm_ffactor, HashMetaPageData::hashm_highmask, HashMetaPageData::hashm_lowmask, HashMetaPageData::hashm_maxbucket, HashMetaPageData::hashm_ntuples, HashMetaPageData::hashm_ovflpoint, HashMetaPageData::hashm_spares, HashPageOpaqueData::hasho_bucket, HashPageOpaqueData::hasho_flag, HashPageOpaqueData::hasho_nextblkno, HashPageOpaqueData::hasho_page_id, HASHO_PAGE_ID, HashPageOpaqueData::hasho_prevblkno, HashPageGetMeta, InvalidBlockNumber, IsBufferCleanupOK(), LH_BUCKET_BEING_POPULATED, LH_BUCKET_BEING_SPLIT, LH_BUCKET_PAGE, LH_META_PAGE, LockBuffer(), MAIN_FORKNUM, MarkBufferDirty(), xl_hash_split_allocate_page::new_bucket, xl_hash_split_allocate_page::new_bucket_flag, xl_hash_split_allocate_page::old_bucket_flag, PageGetSpecialPointer, PageSetLSN, REGBUF_STANDARD, REGBUF_WILL_INIT, RelationNeedsWAL, SizeOfHashSplitAllocPage, START_CRIT_SECTION, XLH_SPLIT_META_UPDATE_MASKS, XLH_SPLIT_META_UPDATE_SPLITPOINT, XLOG_HASH_SPLIT_ALLOCATE_PAGE, XLogBeginInsert(), XLogInsert(), XLogRegisterBufData(), XLogRegisterBuffer(), and XLogRegisterData().

Referenced by _hash_doinsert().

616 {
617  HashMetaPage metap;
618  Bucket old_bucket;
619  Bucket new_bucket;
620  uint32 spare_ndx;
621  BlockNumber start_oblkno;
622  BlockNumber start_nblkno;
623  Buffer buf_nblkno;
624  Buffer buf_oblkno;
625  Page opage;
626  Page npage;
627  HashPageOpaque oopaque;
628  HashPageOpaque nopaque;
629  uint32 maxbucket;
630  uint32 highmask;
631  uint32 lowmask;
632  bool metap_update_masks = false;
633  bool metap_update_splitpoint = false;
634 
635 restart_expand:
636 
637  /*
638  * Write-lock the meta page. It used to be necessary to acquire a
639  * heavyweight lock to begin a split, but that is no longer required.
640  */
642 
643  _hash_checkpage(rel, metabuf, LH_META_PAGE);
644  metap = HashPageGetMeta(BufferGetPage(metabuf));
645 
646  /*
647  * Check to see if split is still needed; someone else might have already
648  * done one while we waited for the lock.
649  *
650  * Make sure this stays in sync with _hash_doinsert()
651  */
652  if (metap->hashm_ntuples <=
653  (double) metap->hashm_ffactor * (metap->hashm_maxbucket + 1))
654  goto fail;
655 
656  /*
657  * Can't split anymore if maxbucket has reached its maximum possible
658  * value.
659  *
660  * Ideally we'd allow bucket numbers up to UINT_MAX-1 (no higher because
661  * the calculation maxbucket+1 mustn't overflow). Currently we restrict
662  * to half that because of overflow looping in _hash_log2() and
663  * insufficient space in hashm_spares[]. It's moot anyway because an
664  * index with 2^32 buckets would certainly overflow BlockNumber and hence
665  * _hash_alloc_buckets() would fail, but if we supported buckets smaller
666  * than a disk block then this would be an independent constraint.
667  *
668  * If you change this, see also the maximum initial number of buckets in
669  * _hash_init().
670  */
671  if (metap->hashm_maxbucket >= (uint32) 0x7FFFFFFE)
672  goto fail;
673 
674  /*
675  * Determine which bucket is to be split, and attempt to take cleanup lock
676  * on the old bucket. If we can't get the lock, give up.
677  *
678  * The cleanup lock protects us not only against other backends, but
679  * against our own backend as well.
680  *
681  * The cleanup lock is mainly to protect the split from concurrent
682  * inserts. See src/backend/access/hash/README, Lock Definitions for
683  * further details. Due to this locking restriction, if there is any
684  * pending scan, the split will give up which is not good, but harmless.
685  */
686  new_bucket = metap->hashm_maxbucket + 1;
687 
688  old_bucket = (new_bucket & metap->hashm_lowmask);
689 
690  start_oblkno = BUCKET_TO_BLKNO(metap, old_bucket);
691 
692  buf_oblkno = _hash_getbuf_with_condlock_cleanup(rel, start_oblkno, LH_BUCKET_PAGE);
693  if (!buf_oblkno)
694  goto fail;
695 
696  opage = BufferGetPage(buf_oblkno);
697  oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
698 
699  /*
700  * We want to finish the split from a bucket as there is no apparent
701  * benefit by not doing so and it will make the code complicated to finish
702  * the split that involves multiple buckets considering the case where new
703  * split also fails. We don't need to consider the new bucket for
704  * completing the split here as it is not possible that a re-split of new
705  * bucket starts when there is still a pending split from old bucket.
706  */
707  if (H_BUCKET_BEING_SPLIT(oopaque))
708  {
709  /*
710  * Copy bucket mapping info now; refer the comment in code below where
711  * we copy this information before calling _hash_splitbucket to see
712  * why this is okay.
713  */
714  maxbucket = metap->hashm_maxbucket;
715  highmask = metap->hashm_highmask;
716  lowmask = metap->hashm_lowmask;
717 
718  /*
719  * Release the lock on metapage and old_bucket, before completing the
720  * split.
721  */
722  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
723  LockBuffer(buf_oblkno, BUFFER_LOCK_UNLOCK);
724 
725  _hash_finish_split(rel, metabuf, buf_oblkno, old_bucket, maxbucket,
726  highmask, lowmask);
727 
728  /* release the pin on old buffer and retry for expand. */
729  _hash_dropbuf(rel, buf_oblkno);
730 
731  goto restart_expand;
732  }
733 
734  /*
735  * Clean the tuples remained from the previous split. This operation
736  * requires cleanup lock and we already have one on the old bucket, so
737  * let's do it. We also don't want to allow further splits from the bucket
738  * till the garbage of previous split is cleaned. This has two
739  * advantages; first, it helps in avoiding the bloat due to garbage and
740  * second is, during cleanup of bucket, we are always sure that the
741  * garbage tuples belong to most recently split bucket. On the contrary,
742  * if we allow cleanup of bucket after meta page is updated to indicate
743  * the new split and before the actual split, the cleanup operation won't
744  * be able to decide whether the tuple has been moved to the newly created
745  * bucket and ended up deleting such tuples.
746  */
747  if (H_NEEDS_SPLIT_CLEANUP(oopaque))
748  {
749  /*
750  * Copy bucket mapping info now; refer to the comment in code below
751  * where we copy this information before calling _hash_splitbucket to
752  * see why this is okay.
753  */
754  maxbucket = metap->hashm_maxbucket;
755  highmask = metap->hashm_highmask;
756  lowmask = metap->hashm_lowmask;
757 
758  /* Release the metapage lock. */
759  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
760 
761  hashbucketcleanup(rel, old_bucket, buf_oblkno, start_oblkno, NULL,
762  maxbucket, highmask, lowmask, NULL, NULL, true,
763  NULL, NULL);
764 
765  _hash_dropbuf(rel, buf_oblkno);
766 
767  goto restart_expand;
768  }
769 
770  /*
771  * There shouldn't be any active scan on new bucket.
772  *
773  * Note: it is safe to compute the new bucket's blkno here, even though we
774  * may still need to update the BUCKET_TO_BLKNO mapping. This is because
775  * the current value of hashm_spares[hashm_ovflpoint] correctly shows
776  * where we are going to put a new splitpoint's worth of buckets.
777  */
778  start_nblkno = BUCKET_TO_BLKNO(metap, new_bucket);
779 
780  /*
781  * If the split point is increasing we need to allocate a new batch of
782  * bucket pages.
783  */
784  spare_ndx = _hash_spareindex(new_bucket + 1);
785  if (spare_ndx > metap->hashm_ovflpoint)
786  {
787  uint32 buckets_to_add;
788 
789  Assert(spare_ndx == metap->hashm_ovflpoint + 1);
790 
791  /*
792  * We treat allocation of buckets as a separate WAL-logged action.
793  * Even if we fail after this operation, won't leak bucket pages;
794  * rather, the next split will consume this space. In any case, even
795  * without failure we don't use all the space in one split operation.
796  */
797  buckets_to_add = _hash_get_totalbuckets(spare_ndx) - new_bucket;
798  if (!_hash_alloc_buckets(rel, start_nblkno, buckets_to_add))
799  {
800  /* can't split due to BlockNumber overflow */
801  _hash_relbuf(rel, buf_oblkno);
802  goto fail;
803  }
804  }
805 
806  /*
807  * Physically allocate the new bucket's primary page. We want to do this
808  * before changing the metapage's mapping info, in case we can't get the
809  * disk space. Ideally, we don't need to check for cleanup lock on new
810  * bucket as no other backend could find this bucket unless meta page is
811  * updated. However, it is good to be consistent with old bucket locking.
812  */
813  buf_nblkno = _hash_getnewbuf(rel, start_nblkno, MAIN_FORKNUM);
814  if (!IsBufferCleanupOK(buf_nblkno))
815  {
816  _hash_relbuf(rel, buf_oblkno);
817  _hash_relbuf(rel, buf_nblkno);
818  goto fail;
819  }
820 
821  /*
822  * Since we are scribbling on the pages in the shared buffers, establish a
823  * critical section. Any failure in this next code leaves us with a big
824  * problem: the metapage is effectively corrupt but could get written back
825  * to disk.
826  */
828 
829  /*
830  * Okay to proceed with split. Update the metapage bucket mapping info.
831  */
832  metap->hashm_maxbucket = new_bucket;
833 
834  if (new_bucket > metap->hashm_highmask)
835  {
836  /* Starting a new doubling */
837  metap->hashm_lowmask = metap->hashm_highmask;
838  metap->hashm_highmask = new_bucket | metap->hashm_lowmask;
839  metap_update_masks = true;
840  }
841 
842  /*
843  * If the split point is increasing we need to adjust the hashm_spares[]
844  * array and hashm_ovflpoint so that future overflow pages will be created
845  * beyond this new batch of bucket pages.
846  */
847  if (spare_ndx > metap->hashm_ovflpoint)
848  {
849  metap->hashm_spares[spare_ndx] = metap->hashm_spares[metap->hashm_ovflpoint];
850  metap->hashm_ovflpoint = spare_ndx;
851  metap_update_splitpoint = true;
852  }
853 
854  MarkBufferDirty(metabuf);
855 
856  /*
857  * Copy bucket mapping info now; this saves re-accessing the meta page
858  * inside _hash_splitbucket's inner loop. Note that once we drop the
859  * split lock, other splits could begin, so these values might be out of
860  * date before _hash_splitbucket finishes. That's okay, since all it
861  * needs is to tell which of these two buckets to map hashkeys into.
862  */
863  maxbucket = metap->hashm_maxbucket;
864  highmask = metap->hashm_highmask;
865  lowmask = metap->hashm_lowmask;
866 
867  opage = BufferGetPage(buf_oblkno);
868  oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
869 
870  /*
871  * Mark the old bucket to indicate that split is in progress. (At
872  * operation end, we will clear the split-in-progress flag.) Also, for a
873  * primary bucket page, hasho_prevblkno stores the number of buckets that
874  * existed as of the last split, so we must update that value here.
875  */
876  oopaque->hasho_flag |= LH_BUCKET_BEING_SPLIT;
877  oopaque->hasho_prevblkno = maxbucket;
878 
879  MarkBufferDirty(buf_oblkno);
880 
881  npage = BufferGetPage(buf_nblkno);
882 
883  /*
884  * initialize the new bucket's primary page and mark it to indicate that
885  * split is in progress.
886  */
887  nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
888  nopaque->hasho_prevblkno = maxbucket;
890  nopaque->hasho_bucket = new_bucket;
892  nopaque->hasho_page_id = HASHO_PAGE_ID;
893 
894  MarkBufferDirty(buf_nblkno);
895 
896  /* XLOG stuff */
897  if (RelationNeedsWAL(rel))
898  {
900  XLogRecPtr recptr;
901 
902  xlrec.new_bucket = maxbucket;
903  xlrec.old_bucket_flag = oopaque->hasho_flag;
904  xlrec.new_bucket_flag = nopaque->hasho_flag;
905  xlrec.flags = 0;
906 
907  XLogBeginInsert();
908 
909  XLogRegisterBuffer(0, buf_oblkno, REGBUF_STANDARD);
910  XLogRegisterBuffer(1, buf_nblkno, REGBUF_WILL_INIT);
911  XLogRegisterBuffer(2, metabuf, REGBUF_STANDARD);
912 
913  if (metap_update_masks)
914  {
916  XLogRegisterBufData(2, (char *) &metap->hashm_lowmask, sizeof(uint32));
917  XLogRegisterBufData(2, (char *) &metap->hashm_highmask, sizeof(uint32));
918  }
919 
920  if (metap_update_splitpoint)
921  {
923  XLogRegisterBufData(2, (char *) &metap->hashm_ovflpoint,
924  sizeof(uint32));
926  (char *) &metap->hashm_spares[metap->hashm_ovflpoint],
927  sizeof(uint32));
928  }
929 
930  XLogRegisterData((char *) &xlrec, SizeOfHashSplitAllocPage);
931 
932  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_ALLOCATE_PAGE);
933 
934  PageSetLSN(BufferGetPage(buf_oblkno), recptr);
935  PageSetLSN(BufferGetPage(buf_nblkno), recptr);
936  PageSetLSN(BufferGetPage(metabuf), recptr);
937  }
938 
940 
941  /* drop lock, but keep pin */
942  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
943 
944  /* Relocate records to the new bucket */
945  _hash_splitbucket(rel, metabuf,
946  old_bucket, new_bucket,
947  buf_oblkno, buf_nblkno, NULL,
948  maxbucket, highmask, lowmask);
949 
950  /* all done, now release the pins on primary buckets. */
951  _hash_dropbuf(rel, buf_oblkno);
952  _hash_dropbuf(rel, buf_nblkno);
953 
954  return;
955 
956  /* Here if decide not to split or fail to acquire old bucket lock */
957 fail:
958 
959  /* We didn't write the metapage, so just drop lock */
960  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
961 }
void XLogRegisterBufData(uint8 block_id, char *data, int len)
Definition: xloginsert.c:361
uint16 hasho_page_id
Definition: hash.h:82
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:86
#define XLH_SPLIT_META_UPDATE_SPLITPOINT
Definition: hash_xlog.h:52
#define LH_META_PAGE
Definition: hash.h:56
void MarkBufferDirty(Buffer buffer)
Definition: bufmgr.c:1458
void XLogRegisterBuffer(uint8 block_id, Buffer buffer, uint8 flags)
Definition: xloginsert.c:213
#define END_CRIT_SECTION()
Definition: miscadmin.h:134
uint16 hashm_ffactor
Definition: hash.h:246
uint32 hashm_highmask
Definition: hash.h:252
#define REGBUF_WILL_INIT
Definition: xloginsert.h:33
#define START_CRIT_SECTION()
Definition: miscadmin.h:132
void hashbucketcleanup(Relation rel, Bucket cur_bucket, Buffer bucket_buf, BlockNumber bucket_blkno, BufferAccessStrategy bstrategy, uint32 maxbucket, uint32 highmask, uint32 lowmask, double *tuples_removed, double *num_index_tuples, bool split_cleanup, IndexBulkDeleteCallback callback, void *callback_state)
Definition: hash.c:677
uint32 BlockNumber
Definition: block.h:31
Buffer _hash_getnewbuf(Relation rel, BlockNumber blkno, ForkNumber forkNum)
Definition: hashpage.c:196
void _hash_dropbuf(Relation rel, Buffer buf)
Definition: hashpage.c:275
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:88
uint32 hashm_lowmask
Definition: hash.h:253
#define BUCKET_TO_BLKNO(metap, B)
Definition: hash.h:38
static bool _hash_alloc_buckets(Relation rel, BlockNumber firstblock, uint32 nblocks)
Definition: hashpage.c:989
uint32 Bucket
Definition: hash.h:34
#define H_NEEDS_SPLIT_CLEANUP(opaque)
Definition: hash.h:87
BlockNumber hasho_prevblkno
Definition: hash.h:78
#define SizeOfHashSplitAllocPage
Definition: hash_xlog.h:106
uint32 _hash_spareindex(uint32 num_bucket)
Definition: hashutil.c:157
uint32 _hash_get_totalbuckets(uint32 splitpoint_phase)
Definition: hashutil.c:189
void _hash_finish_split(Relation rel, Buffer metabuf, Buffer obuf, Bucket obucket, uint32 maxbucket, uint32 highmask, uint32 lowmask)
Definition: hashpage.c:1353
#define REGBUF_STANDARD
Definition: xloginsert.h:35
#define LH_BUCKET_BEING_SPLIT
Definition: hash.h:58
unsigned int uint32
Definition: c.h:359
Buffer _hash_getbuf_with_condlock_cleanup(Relation rel, BlockNumber blkno, int flags)
Definition: hashpage.c:94
#define BufferGetPage(buffer)
Definition: bufmgr.h:159
bool IsBufferCleanupOK(Buffer buffer)
Definition: bufmgr.c:3830
uint32 hashm_ovflpoint
Definition: hash.h:254
void XLogRegisterData(char *data, int len)
Definition: xloginsert.c:323
#define LH_BUCKET_BEING_POPULATED
Definition: hash.h:57
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:415
void _hash_checkpage(Relation rel, Buffer buf, int flags)
Definition: hashutil.c:225
double hashm_ntuples
Definition: hash.h:245
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3602
uint32 hashm_spares[HASH_MAX_SPLITPOINTS]
Definition: hash.h:259
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:264
#define LH_BUCKET_PAGE
Definition: hash.h:54
#define H_BUCKET_BEING_SPLIT(opaque)
Definition: hash.h:88
static void _hash_splitbucket(Relation rel, Buffer metabuf, Bucket obucket, Bucket nbucket, Buffer obuf, Buffer nbuf, HTAB *htab, uint32 maxbucket, uint32 highmask, uint32 lowmask)
Definition: hashpage.c:1070
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:733
Bucket hasho_bucket
Definition: hash.h:80
#define PageGetSpecialPointer(page)
Definition: bufpage.h:326
#define InvalidBlockNumber
Definition: block.h:33
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:85
#define HASHO_PAGE_ID
Definition: hash.h:98
#define RelationNeedsWAL(relation)
Definition: rel.h:525
uint32 hashm_maxbucket
Definition: hash.h:251
uint16 hasho_flag
Definition: hash.h:81
#define XLOG_HASH_SPLIT_ALLOCATE_PAGE
Definition: hash_xlog.h:31
#define HashPageGetMeta(page)
Definition: hash.h:305
BlockNumber hasho_nextblkno
Definition: hash.h:79
#define XLH_SPLIT_META_UPDATE_MASKS
Definition: hash_xlog.h:51
void XLogBeginInsert(void)
Definition: xloginsert.c:120
#define PageSetLSN(page, lsn)
Definition: bufpage.h:368
int Buffer
Definition: buf.h:23
Pointer Page
Definition: bufpage.h:78

◆ _hash_finish_split()

void _hash_finish_split ( Relation  rel,
Buffer  metabuf,
Buffer  obuf,
Bucket  obucket,
uint32  maxbucket,
uint32  highmask,
uint32  lowmask 
)

Definition at line 1353 of file hashpage.c.

References _hash_dropbuf(), _hash_get_newblock_from_oldbucket(), _hash_getbuf(), _hash_relbuf(), _hash_splitbucket(), Assert, BlockNumberIsValid, BUFFER_LOCK_UNLOCK, BufferGetPage, ConditionalLockBufferForCleanup(), CurrentMemoryContext, HASHCTL::entrysize, FirstOffsetNumber, HASH_BLOBS, HASH_CONTEXT, hash_create(), hash_destroy(), HASH_ELEM, HASH_ENTER, HASH_READ, hash_search(), HashPageOpaqueData::hasho_bucket, HashPageOpaqueData::hasho_nextblkno, HASHCTL::hcxt, InvalidBuffer, HASHCTL::keysize, LH_BUCKET_PAGE, LH_OVERFLOW_PAGE, LockBuffer(), OffsetNumberNext, PageGetItem, PageGetItemId, PageGetMaxOffsetNumber, PageGetSpecialPointer, and IndexTupleData::t_tid.

Referenced by _hash_doinsert(), and _hash_expandtable().

1355 {
1356  HASHCTL hash_ctl;
1357  HTAB *tidhtab;
1358  Buffer bucket_nbuf = InvalidBuffer;
1359  Buffer nbuf;
1360  Page npage;
1361  BlockNumber nblkno;
1362  BlockNumber bucket_nblkno;
1363  HashPageOpaque npageopaque;
1364  Bucket nbucket;
1365  bool found;
1366 
1367  /* Initialize hash tables used to track TIDs */
1368  memset(&hash_ctl, 0, sizeof(hash_ctl));
1369  hash_ctl.keysize = sizeof(ItemPointerData);
1370  hash_ctl.entrysize = sizeof(ItemPointerData);
1371  hash_ctl.hcxt = CurrentMemoryContext;
1372 
1373  tidhtab =
1374  hash_create("bucket ctids",
1375  256, /* arbitrary initial size */
1376  &hash_ctl,
1378 
1379  bucket_nblkno = nblkno = _hash_get_newblock_from_oldbucket(rel, obucket);
1380 
1381  /*
1382  * Scan the new bucket and build hash table of TIDs
1383  */
1384  for (;;)
1385  {
1386  OffsetNumber noffnum;
1387  OffsetNumber nmaxoffnum;
1388 
1389  nbuf = _hash_getbuf(rel, nblkno, HASH_READ,
1391 
1392  /* remember the primary bucket buffer to acquire cleanup lock on it. */
1393  if (nblkno == bucket_nblkno)
1394  bucket_nbuf = nbuf;
1395 
1396  npage = BufferGetPage(nbuf);
1397  npageopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
1398 
1399  /* Scan each tuple in new page */
1400  nmaxoffnum = PageGetMaxOffsetNumber(npage);
1401  for (noffnum = FirstOffsetNumber;
1402  noffnum <= nmaxoffnum;
1403  noffnum = OffsetNumberNext(noffnum))
1404  {
1405  IndexTuple itup;
1406 
1407  /* Fetch the item's TID and insert it in hash table. */
1408  itup = (IndexTuple) PageGetItem(npage,
1409  PageGetItemId(npage, noffnum));
1410 
1411  (void) hash_search(tidhtab, &itup->t_tid, HASH_ENTER, &found);
1412 
1413  Assert(!found);
1414  }
1415 
1416  nblkno = npageopaque->hasho_nextblkno;
1417 
1418  /*
1419  * release our write lock without modifying buffer and ensure to
1420  * retain the pin on primary bucket.
1421  */
1422  if (nbuf == bucket_nbuf)
1424  else
1425  _hash_relbuf(rel, nbuf);
1426 
1427  /* Exit loop if no more overflow pages in new bucket */
1428  if (!BlockNumberIsValid(nblkno))
1429  break;
1430  }
1431 
1432  /*
1433  * Conditionally get the cleanup lock on old and new buckets to perform
1434  * the split operation. If we don't get the cleanup locks, silently give
1435  * up and next insertion on old bucket will try again to complete the
1436  * split.
1437  */
1439  {
1440  hash_destroy(tidhtab);
1441  return;
1442  }
1443  if (!ConditionalLockBufferForCleanup(bucket_nbuf))
1444  {
1446  hash_destroy(tidhtab);
1447  return;
1448  }
1449 
1450  npage = BufferGetPage(bucket_nbuf);
1451  npageopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
1452  nbucket = npageopaque->hasho_bucket;
1453 
1454  _hash_splitbucket(rel, metabuf, obucket,
1455  nbucket, obuf, bucket_nbuf, tidhtab,
1456  maxbucket, highmask, lowmask);
1457 
1458  _hash_dropbuf(rel, bucket_nbuf);
1459  hash_destroy(tidhtab);
1460 }
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:86
void hash_destroy(HTAB *hashp)
Definition: dynahash.c:814
#define HASH_CONTEXT
Definition: hsearch.h:93
#define HASH_ELEM
Definition: hsearch.h:87
MemoryContext hcxt
Definition: hsearch.h:78
ItemPointerData t_tid
Definition: itup.h:37
#define InvalidBuffer
Definition: buf.h:25
Size entrysize
Definition: hsearch.h:73
uint32 BlockNumber
Definition: block.h:31
void _hash_dropbuf(Relation rel, Buffer buf)
Definition: hashpage.c:275
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:906
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:68
#define PageGetMaxOffsetNumber(page)
Definition: bufpage.h:357
uint16 OffsetNumber
Definition: off.h:24
#define HASH_READ
Definition: hash.h:321
uint32 Bucket
Definition: hash.h:34
Definition: dynahash.c:208
bool ConditionalLockBufferForCleanup(Buffer buffer)
Definition: bufmgr.c:3774
#define FirstOffsetNumber
Definition: off.h:27
IndexTupleData * IndexTuple
Definition: itup.h:53
MemoryContext CurrentMemoryContext
Definition: mcxt.c:38
#define BufferGetPage(buffer)
Definition: bufmgr.h:159
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:235
#define HASH_BLOBS
Definition: hsearch.h:88
HTAB * hash_create(const char *tabname, long nelem, HASHCTL *info, int flags)
Definition: dynahash.c:316
#define LH_OVERFLOW_PAGE
Definition: hash.h:53
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3602
Size keysize
Definition: hsearch.h:72
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:264
BlockNumber _hash_get_newblock_from_oldbucket(Relation rel, Bucket old_bucket)
Definition: hashutil.c:469
#define BlockNumberIsValid(blockNumber)
Definition: block.h:70
#define LH_BUCKET_PAGE
Definition: hash.h:54
static void _hash_splitbucket(Relation rel, Buffer metabuf, Bucket obucket, Bucket nbucket, Buffer obuf, Buffer nbuf, HTAB *htab, uint32 maxbucket, uint32 highmask, uint32 lowmask)
Definition: hashpage.c:1070
#define Assert(condition)
Definition: c.h:733
Bucket hasho_bucket
Definition: hash.h:80
struct ItemPointerData ItemPointerData
#define OffsetNumberNext(offsetNumber)
Definition: off.h:52
#define PageGetSpecialPointer(page)
Definition: bufpage.h:326
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:85
BlockNumber hasho_nextblkno
Definition: hash.h:79
int Buffer
Definition: buf.h:23
#define PageGetItem(page, itemId)
Definition: bufpage.h:340
Pointer Page
Definition: bufpage.h:78

◆ _hash_getbucketbuf_from_hashkey()

Buffer _hash_getbucketbuf_from_hashkey ( Relation  rel,
uint32  hashkey,
int  access,
HashMetaPage cachedmetap 
)

Definition at line 1557 of file hashpage.c.

References _hash_dropbuf(), _hash_getbuf(), _hash_getcachedmetap(), _hash_hashkey2bucket(), _hash_relbuf(), Assert, BUCKET_TO_BLKNO, buf, BufferGetPage, BufferIsValid, HASH_READ, HASH_WRITE, HashMetaPageData::hashm_highmask, HashMetaPageData::hashm_lowmask, HashMetaPageData::hashm_maxbucket, HashPageOpaqueData::hasho_bucket, HashPageOpaqueData::hasho_prevblkno, InvalidBlockNumber, InvalidBuffer, LH_BUCKET_PAGE, and PageGetSpecialPointer.

Referenced by _hash_doinsert(), and _hash_first().

1559 {
1560  HashMetaPage metap;
1561  Buffer buf;
1562  Buffer metabuf = InvalidBuffer;
1563  Page page;
1564  Bucket bucket;
1565  BlockNumber blkno;
1566  HashPageOpaque opaque;
1567 
1568  /* We read from target bucket buffer, hence locking is must. */
1569  Assert(access == HASH_READ || access == HASH_WRITE);
1570 
1571  metap = _hash_getcachedmetap(rel, &metabuf, false);
1572  Assert(metap != NULL);
1573 
1574  /*
1575  * Loop until we get a lock on the correct target bucket.
1576  */
1577  for (;;)
1578  {
1579  /*
1580  * Compute the target bucket number, and convert to block number.
1581  */
1582  bucket = _hash_hashkey2bucket(hashkey,
1583  metap->hashm_maxbucket,
1584  metap->hashm_highmask,
1585  metap->hashm_lowmask);
1586 
1587  blkno = BUCKET_TO_BLKNO(metap, bucket);
1588 
1589  /* Fetch the primary bucket page for the bucket */
1590  buf = _hash_getbuf(rel, blkno, access, LH_BUCKET_PAGE);
1591  page = BufferGetPage(buf);
1592  opaque = (HashPageOpaque) PageGetSpecialPointer(page);
1593  Assert(opaque->hasho_bucket == bucket);
1595 
1596  /*
1597  * If this bucket hasn't been split, we're done.
1598  */
1599  if (opaque->hasho_prevblkno <= metap->hashm_maxbucket)
1600  break;
1601 
1602  /* Drop lock on this buffer, update cached metapage, and retry. */
1603  _hash_relbuf(rel, buf);
1604  metap = _hash_getcachedmetap(rel, &metabuf, true);
1605  Assert(metap != NULL);
1606  }
1607 
1608  if (BufferIsValid(metabuf))
1609  _hash_dropbuf(rel, metabuf);
1610 
1611  if (cachedmetap)
1612  *cachedmetap = metap;
1613 
1614  return buf;
1615 }
Bucket _hash_hashkey2bucket(uint32 hashkey, uint32 maxbucket, uint32 highmask, uint32 lowmask)
Definition: hashutil.c:125
uint32 hashm_highmask
Definition: hash.h:252
#define InvalidBuffer
Definition: buf.h:25
uint32 BlockNumber
Definition: block.h:31
void _hash_dropbuf(Relation rel, Buffer buf)
Definition: hashpage.c:275
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:68
uint32 hashm_lowmask
Definition: hash.h:253
#define BUCKET_TO_BLKNO(metap, B)
Definition: hash.h:38
#define HASH_READ
Definition: hash.h:321
uint32 Bucket
Definition: hash.h:34
BlockNumber hasho_prevblkno
Definition: hash.h:78
static char * buf
Definition: pg_test_fsync.c:67
#define HASH_WRITE
Definition: hash.h:322
#define BufferGetPage(buffer)
Definition: bufmgr.h:159
HashMetaPage _hash_getcachedmetap(Relation rel, Buffer *metabuf, bool force_refresh)
Definition: hashpage.c:1499
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:264
#define LH_BUCKET_PAGE
Definition: hash.h:54
#define Assert(condition)
Definition: c.h:733
Bucket hasho_bucket
Definition: hash.h:80
#define PageGetSpecialPointer(page)
Definition: bufpage.h:326
#define InvalidBlockNumber
Definition: block.h:33
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:85
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
uint32 hashm_maxbucket
Definition: hash.h:251
int Buffer
Definition: buf.h:23
Pointer Page
Definition: bufpage.h:78

◆ _hash_getbuf()

Buffer _hash_getbuf ( Relation  rel,
BlockNumber  blkno,
int  access,
int  flags 
)

Definition at line 68 of file hashpage.c.

References _hash_checkpage(), buf, elog, ERROR, HASH_NOLOCK, LockBuffer(), P_NEW, and ReadBuffer().

Referenced by _hash_addovflpage(), _hash_doinsert(), _hash_finish_split(), _hash_first(), _hash_freeovflpage(), _hash_get_newblock_from_oldbucket(), _hash_get_oldblock_from_newbucket(), _hash_getbucketbuf_from_hashkey(), _hash_getcachedmetap(), _hash_kill_items(), _hash_next(), _hash_readnext(), _hash_readprev(), _hash_splitbucket(), hash_bitmap_info(), hashbulkdelete(), and pgstathashindex().

69 {
70  Buffer buf;
71 
72  if (blkno == P_NEW)
73  elog(ERROR, "hash AM does not use P_NEW");
74 
75  buf = ReadBuffer(rel, blkno);
76 
77  if (access != HASH_NOLOCK)
78  LockBuffer(buf, access);
79 
80  /* ref count and lock type are correct */
81 
82  _hash_checkpage(rel, buf, flags);
83 
84  return buf;
85 }
#define P_NEW
Definition: bufmgr.h:81
#define ERROR
Definition: elog.h:43
static char * buf
Definition: pg_test_fsync.c:67
#define HASH_NOLOCK
Definition: hash.h:323
void _hash_checkpage(Relation rel, Buffer buf, int flags)
Definition: hashutil.c:225
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3602
Buffer ReadBuffer(Relation reln, BlockNumber blockNum)
Definition: bufmgr.c:596
#define elog(elevel,...)
Definition: elog.h:228
int Buffer
Definition: buf.h:23

◆ _hash_getbuf_with_condlock_cleanup()

Buffer _hash_getbuf_with_condlock_cleanup ( Relation  rel,
BlockNumber  blkno,
int  flags 
)

Definition at line 94 of file hashpage.c.

References _hash_checkpage(), buf, ConditionalLockBufferForCleanup(), elog, ERROR, InvalidBuffer, P_NEW, ReadBuffer(), and ReleaseBuffer().

Referenced by _hash_expandtable().

95 {
96  Buffer buf;
97 
98  if (blkno == P_NEW)
99  elog(ERROR, "hash AM does not use P_NEW");
100 
101  buf = ReadBuffer(rel, blkno);
102 
104  {
105  ReleaseBuffer(buf);
106  return InvalidBuffer;
107  }
108 
109  /* ref count and lock type are correct */
110 
111  _hash_checkpage(rel, buf, flags);
112 
113  return buf;
114 }
#define InvalidBuffer
Definition: buf.h:25
void ReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:3365
#define P_NEW
Definition: bufmgr.h:81
bool ConditionalLockBufferForCleanup(Buffer buffer)
Definition: bufmgr.c:3774
#define ERROR
Definition: elog.h:43
static char * buf
Definition: pg_test_fsync.c:67
void _hash_checkpage(Relation rel, Buffer buf, int flags)
Definition: hashutil.c:225
Buffer ReadBuffer(Relation reln, BlockNumber blockNum)
Definition: bufmgr.c:596
#define elog(elevel,...)
Definition: elog.h:228
int Buffer
Definition: buf.h:23

◆ _hash_getbuf_with_strategy()

Buffer _hash_getbuf_with_strategy ( Relation  rel,
BlockNumber  blkno,
int  access,
int  flags,
BufferAccessStrategy  bstrategy 
)

Definition at line 237 of file hashpage.c.

References _hash_checkpage(), buf, elog, ERROR, HASH_NOLOCK, LockBuffer(), MAIN_FORKNUM, P_NEW, RBM_NORMAL, and ReadBufferExtended().

Referenced by _hash_freeovflpage(), _hash_squeezebucket(), hashbucketcleanup(), and pgstat_hash_page().

240 {
241  Buffer buf;
242 
243  if (blkno == P_NEW)
244  elog(ERROR, "hash AM does not use P_NEW");
245 
246  buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_NORMAL, bstrategy);
247 
248  if (access != HASH_NOLOCK)
249  LockBuffer(buf, access);
250 
251  /* ref count and lock type are correct */
252 
253  _hash_checkpage(rel, buf, flags);
254 
255  return buf;
256 }
Buffer ReadBufferExtended(Relation reln, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy)
Definition: bufmgr.c:642
#define P_NEW
Definition: bufmgr.h:81
#define ERROR
Definition: elog.h:43
static char * buf
Definition: pg_test_fsync.c:67
#define HASH_NOLOCK
Definition: hash.h:323
void _hash_checkpage(Relation rel, Buffer buf, int flags)
Definition: hashutil.c:225
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3602
#define elog(elevel,...)
Definition: elog.h:228
int Buffer
Definition: buf.h:23

◆ _hash_getcachedmetap()

HashMetaPage _hash_getcachedmetap ( Relation  rel,
Buffer metabuf,
bool  force_refresh 
)

Definition at line 1499 of file hashpage.c.

References _hash_getbuf(), Assert, BUFFER_LOCK_SHARE, BUFFER_LOCK_UNLOCK, BufferGetPage, BufferIsValid, HASH_METAPAGE, HASH_READ, HashPageGetMeta, LH_META_PAGE, LockBuffer(), MemoryContextAlloc(), RelationData::rd_amcache, and RelationData::rd_indexcxt.

Referenced by _hash_getbucketbuf_from_hashkey(), and hashbulkdelete().

1500 {
1501  Page page;
1502 
1503  Assert(metabuf);
1504  if (force_refresh || rel->rd_amcache == NULL)
1505  {
1506  char *cache = NULL;
1507 
1508  /*
1509  * It's important that we don't set rd_amcache to an invalid value.
1510  * Either MemoryContextAlloc or _hash_getbuf could fail, so don't
1511  * install a pointer to the newly-allocated storage in the actual
1512  * relcache entry until both have succeeded.
1513  */
1514  if (rel->rd_amcache == NULL)
1515  cache = MemoryContextAlloc(rel->rd_indexcxt,
1516  sizeof(HashMetaPageData));
1517 
1518  /* Read the metapage. */
1519  if (BufferIsValid(*metabuf))
1520  LockBuffer(*metabuf, BUFFER_LOCK_SHARE);
1521  else
1522  *metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ,
1523  LH_META_PAGE);
1524  page = BufferGetPage(*metabuf);
1525 
1526  /* Populate the cache. */
1527  if (rel->rd_amcache == NULL)
1528  rel->rd_amcache = cache;
1529  memcpy(rel->rd_amcache, HashPageGetMeta(page),
1530  sizeof(HashMetaPageData));
1531 
1532  /* Release metapage lock, but keep the pin. */
1533  LockBuffer(*metabuf, BUFFER_LOCK_UNLOCK);
1534  }
1535 
1536  return (HashMetaPage) rel->rd_amcache;
1537 }
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:86
#define LH_META_PAGE
Definition: hash.h:56
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:68
#define HASH_READ
Definition: hash.h:321
#define BufferGetPage(buffer)
Definition: bufmgr.h:159
#define HASH_METAPAGE
Definition: hash.h:195
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3602
#define Assert(condition)
Definition: c.h:733
#define BufferIsValid(bufnum)
Definition: bufmgr.h:113
#define HashPageGetMeta(page)
Definition: hash.h:305
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:796
MemoryContext rd_indexcxt
Definition: rel.h:155
#define BUFFER_LOCK_SHARE
Definition: bufmgr.h:87
void * rd_amcache
Definition: rel.h:179
Pointer Page
Definition: bufpage.h:78

◆ _hash_getinitbuf()

Buffer _hash_getinitbuf ( Relation  rel,
BlockNumber  blkno 
)

Definition at line 133 of file hashpage.c.

References _hash_pageinit(), buf, BufferGetPage, BufferGetPageSize, elog, ERROR, MAIN_FORKNUM, P_NEW, RBM_ZERO_AND_LOCK, and ReadBufferExtended().

Referenced by _hash_addovflpage().

134 {
135  Buffer buf;
136 
137  if (blkno == P_NEW)
138  elog(ERROR, "hash AM does not use P_NEW");
139 
141  NULL);
142 
143  /* ref count and lock type are correct */
144 
145  /* initialize the page */
147 
148  return buf;
149 }
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:597
Buffer ReadBufferExtended(Relation reln, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy)
Definition: bufmgr.c:642
#define P_NEW
Definition: bufmgr.h:81
#define ERROR
Definition: elog.h:43
static char * buf
Definition: pg_test_fsync.c:67
#define BufferGetPage(buffer)
Definition: bufmgr.h:159
#define BufferGetPageSize(buffer)
Definition: bufmgr.h:146
#define elog(elevel,...)
Definition: elog.h:228
int Buffer
Definition: buf.h:23

◆ _hash_getnewbuf()

Buffer _hash_getnewbuf ( Relation  rel,
BlockNumber  blkno,
ForkNumber  forkNum 
)

Definition at line 196 of file hashpage.c.

References _hash_pageinit(), buf, BufferGetBlockNumber(), BufferGetPage, BufferGetPageSize, elog, ERROR, HASH_WRITE, LockBuffer(), P_NEW, RBM_NORMAL, RBM_ZERO_AND_LOCK, ReadBufferExtended(), RelationGetNumberOfBlocksInFork(), and RelationGetRelationName.

Referenced by _hash_addovflpage(), _hash_expandtable(), and _hash_init().

197 {
198  BlockNumber nblocks = RelationGetNumberOfBlocksInFork(rel, forkNum);
199  Buffer buf;
200 
201  if (blkno == P_NEW)
202  elog(ERROR, "hash AM does not use P_NEW");
203  if (blkno > nblocks)
204  elog(ERROR, "access to noncontiguous page in hash index \"%s\"",
206 
207  /* smgr insists we use P_NEW to extend the relation */
208  if (blkno == nblocks)
209  {
210  buf = ReadBufferExtended(rel, forkNum, P_NEW, RBM_NORMAL, NULL);
211  if (BufferGetBlockNumber(buf) != blkno)
212  elog(ERROR, "unexpected hash relation size: %u, should be %u",
213  BufferGetBlockNumber(buf), blkno);
214  LockBuffer(buf, HASH_WRITE);
215  }
216  else
217  {
218  buf = ReadBufferExtended(rel, forkNum, blkno, RBM_ZERO_AND_LOCK,
219  NULL);
220  }
221 
222  /* ref count and lock type are correct */
223 
224  /* initialize the page */
226 
227  return buf;
228 }
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:597
Buffer ReadBufferExtended(Relation reln, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy)
Definition: bufmgr.c:642
uint32 BlockNumber
Definition: block.h:31
#define P_NEW
Definition: bufmgr.h:81
#define ERROR
Definition: elog.h:43
static char * buf
Definition: pg_test_fsync.c:67
#define HASH_WRITE
Definition: hash.h:322
#define RelationGetRelationName(relation)
Definition: rel.h:457
#define BufferGetPage(buffer)
Definition: bufmgr.h:159
#define BufferGetPageSize(buffer)
Definition: bufmgr.h:146
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3602
BlockNumber RelationGetNumberOfBlocksInFork(Relation relation, ForkNumber forkNum)
Definition: bufmgr.c:2800
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:2613
#define elog(elevel,...)
Definition: elog.h:228
int Buffer
Definition: buf.h:23

◆ _hash_init()

uint32 _hash_init ( Relation  rel,
double  num_tuples,
ForkNumber  forkNum 
)

Definition at line 325 of file hashpage.c.

References _hash_getnewbuf(), _hash_init_metabuffer(), _hash_initbitmapbuffer(), _hash_initbuf(), _hash_relbuf(), xl_hash_init_bitmap_page::bmsize, BUCKET_TO_BLKNO, buf, BUFFER_LOCK_EXCLUSIVE, BUFFER_LOCK_UNLOCK, BufferGetPage, CHECK_FOR_INTERRUPTS, elog, ereport, errcode(), errmsg(), ERROR, xl_hash_init_meta_page::ffactor, HASH_DEFAULT_FILLFACTOR, HASH_MAX_BITMAPS, HASH_METAPAGE, HashMetaPageData::hashm_bmsize, HashMetaPageData::hashm_ffactor, HashMetaPageData::hashm_mapp, HashMetaPageData::hashm_maxbucket, HashMetaPageData::hashm_nmaps, HashMetaPageData::hashm_procid, HashPageGetMeta, HASHSTANDARD_PROC, i, index_getprocid(), INIT_FORKNUM, LH_BUCKET_PAGE, LockBuffer(), log_newpage(), MarkBufferDirty(), MAXALIGN, xl_hash_init_meta_page::num_tuples, PageSetLSN, xl_hash_init_meta_page::procid, RelationData::rd_node, REGBUF_STANDARD, REGBUF_WILL_INIT, RelationGetNumberOfBlocksInFork(), RelationGetRelationName, RelationGetTargetPageUsage, RelationNeedsWAL, SizeOfHashInitBitmapPage, SizeOfHashInitMetaPage, XLOG_HASH_INIT_BITMAP_PAGE, XLOG_HASH_INIT_META_PAGE, XLogBeginInsert(), XLogInsert(), XLogRegisterBuffer(), and XLogRegisterData().

Referenced by hashbuild(), and hashbuildempty().

326 {
327  Buffer metabuf;
328  Buffer buf;
329  Buffer bitmapbuf;
330  Page pg;
331  HashMetaPage metap;
332  RegProcedure procid;
333  int32 data_width;
334  int32 item_width;
335  int32 ffactor;
336  uint32 num_buckets;
337  uint32 i;
338  bool use_wal;
339 
340  /* safety check */
341  if (RelationGetNumberOfBlocksInFork(rel, forkNum) != 0)
342  elog(ERROR, "cannot initialize non-empty hash index \"%s\"",
344 
345  /*
346  * WAL log creation of pages if the relation is persistent, or this is the
347  * init fork. Init forks for unlogged relations always need to be WAL
348  * logged.
349  */
350  use_wal = RelationNeedsWAL(rel) || forkNum == INIT_FORKNUM;
351 
352  /*
353  * Determine the target fill factor (in tuples per bucket) for this index.
354  * The idea is to make the fill factor correspond to pages about as full
355  * as the user-settable fillfactor parameter says. We can compute it
356  * exactly since the index datatype (i.e. uint32 hash key) is fixed-width.
357  */
358  data_width = sizeof(uint32);
359  item_width = MAXALIGN(sizeof(IndexTupleData)) + MAXALIGN(data_width) +
360  sizeof(ItemIdData); /* include the line pointer */
361  ffactor = RelationGetTargetPageUsage(rel, HASH_DEFAULT_FILLFACTOR) / item_width;
362  /* keep to a sane range */
363  if (ffactor < 10)
364  ffactor = 10;
365 
366  procid = index_getprocid(rel, 1, HASHSTANDARD_PROC);
367 
368  /*
369  * We initialize the metapage, the first N bucket pages, and the first
370  * bitmap page in sequence, using _hash_getnewbuf to cause smgrextend()
371  * calls to occur. This ensures that the smgr level has the right idea of
372  * the physical index length.
373  *
374  * Critical section not required, because on error the creation of the
375  * whole relation will be rolled back.
376  */
377  metabuf = _hash_getnewbuf(rel, HASH_METAPAGE, forkNum);
378  _hash_init_metabuffer(metabuf, num_tuples, procid, ffactor, false);
379  MarkBufferDirty(metabuf);
380 
381  pg = BufferGetPage(metabuf);
382  metap = HashPageGetMeta(pg);
383 
384  /* XLOG stuff */
385  if (use_wal)
386  {
388  XLogRecPtr recptr;
389 
390  xlrec.num_tuples = num_tuples;
391  xlrec.procid = metap->hashm_procid;
392  xlrec.ffactor = metap->hashm_ffactor;
393 
394  XLogBeginInsert();
395  XLogRegisterData((char *) &xlrec, SizeOfHashInitMetaPage);
397 
398  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_INIT_META_PAGE);
399 
400  PageSetLSN(BufferGetPage(metabuf), recptr);
401  }
402 
403  num_buckets = metap->hashm_maxbucket + 1;
404 
405  /*
406  * Release buffer lock on the metapage while we initialize buckets.
407  * Otherwise, we'll be in interrupt holdoff and the CHECK_FOR_INTERRUPTS
408  * won't accomplish anything. It's a bad idea to hold buffer locks for
409  * long intervals in any case, since that can block the bgwriter.
410  */
411  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
412 
413  /*
414  * Initialize and WAL Log the first N buckets
415  */
416  for (i = 0; i < num_buckets; i++)
417  {
418  BlockNumber blkno;
419 
420  /* Allow interrupts, in case N is huge */
422 
423  blkno = BUCKET_TO_BLKNO(metap, i);
424  buf = _hash_getnewbuf(rel, blkno, forkNum);
425  _hash_initbuf(buf, metap->hashm_maxbucket, i, LH_BUCKET_PAGE, false);
426  MarkBufferDirty(buf);
427 
428  if (use_wal)
429  log_newpage(&rel->rd_node,
430  forkNum,
431  blkno,
432  BufferGetPage(buf),
433  true);
434  _hash_relbuf(rel, buf);
435  }
436 
437  /* Now reacquire buffer lock on metapage */
439 
440  /*
441  * Initialize bitmap page
442  */
443  bitmapbuf = _hash_getnewbuf(rel, num_buckets + 1, forkNum);
444  _hash_initbitmapbuffer(bitmapbuf, metap->hashm_bmsize, false);
445  MarkBufferDirty(bitmapbuf);
446 
447  /* add the new bitmap page to the metapage's list of bitmaps */
448  /* metapage already has a write lock */
449  if (metap->hashm_nmaps >= HASH_MAX_BITMAPS)
450  ereport(ERROR,
451  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
452  errmsg("out of overflow pages in hash index \"%s\"",
453  RelationGetRelationName(rel))));
454 
455  metap->hashm_mapp[metap->hashm_nmaps] = num_buckets + 1;
456 
457  metap->hashm_nmaps++;
458  MarkBufferDirty(metabuf);
459 
460  /* XLOG stuff */
461  if (use_wal)
462  {
464  XLogRecPtr recptr;
465 
466  xlrec.bmsize = metap->hashm_bmsize;
467 
468  XLogBeginInsert();
469  XLogRegisterData((char *) &xlrec, SizeOfHashInitBitmapPage);
470  XLogRegisterBuffer(0, bitmapbuf, REGBUF_WILL_INIT);
471 
472  /*
473  * This is safe only because nobody else can be modifying the index at
474  * this stage; it's only visible to the transaction that is creating
475  * it.
476  */
477  XLogRegisterBuffer(1, metabuf, REGBUF_STANDARD);
478 
479  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_INIT_BITMAP_PAGE);
480 
481  PageSetLSN(BufferGetPage(bitmapbuf), recptr);
482  PageSetLSN(BufferGetPage(metabuf), recptr);
483  }
484 
485  /* all done */
486  _hash_relbuf(rel, bitmapbuf);
487  _hash_relbuf(rel, metabuf);
488 
489  return num_buckets;
490 }
#define HASH_DEFAULT_FILLFACTOR
Definition: hash.h:278
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:86
RegProcedure hashm_procid
Definition: hash.h:258
void MarkBufferDirty(Buffer buffer)
Definition: bufmgr.c:1458
void XLogRegisterBuffer(uint8 block_id, Buffer buffer, uint8 flags)
Definition: xloginsert.c:213
regproc RegProcedure
Definition: c.h:506
#define XLOG_HASH_INIT_BITMAP_PAGE
Definition: hash_xlog.h:28
uint16 hashm_ffactor
Definition: hash.h:246
#define REGBUF_WILL_INIT
Definition: xloginsert.h:33
int errcode(int sqlerrcode)
Definition: elog.c:608
uint32 BlockNumber
Definition: block.h:31
Buffer _hash_getnewbuf(Relation rel, BlockNumber blkno, ForkNumber forkNum)
Definition: hashpage.c:196
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:88
signed int int32
Definition: c.h:347
#define BUCKET_TO_BLKNO(metap, B)
Definition: hash.h:38
#define ERROR
Definition: elog.h:43
#define SizeOfHashInitMetaPage
Definition: hash_xlog.h:223
void _hash_init_metabuffer(Buffer buf, double num_tuples, RegProcedure procid, uint16 ffactor, bool initpage)
Definition: hashpage.c:496
#define SizeOfHashInitBitmapPage
Definition: hash_xlog.h:239
uint32 hashm_nmaps
Definition: hash.h:257
static char * buf
Definition: pg_test_fsync.c:67
#define REGBUF_STANDARD
Definition: xloginsert.h:35
#define RelationGetRelationName(relation)
Definition: rel.h:457
unsigned int uint32
Definition: c.h:359
struct ItemIdData ItemIdData
#define XLOG_HASH_INIT_META_PAGE
Definition: hash_xlog.h:27
#define HASHSTANDARD_PROC
Definition: hash.h:337
#define BufferGetPage(buffer)
Definition: bufmgr.h:159
#define ereport(elevel, rest)
Definition: elog.h:141
#define HASH_MAX_BITMAPS
Definition: hash.h:227
void XLogRegisterData(char *data, int len)
Definition: xloginsert.c:323
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:415
#define HASH_METAPAGE
Definition: hash.h:195
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3602
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:264
void _hash_initbitmapbuffer(Buffer buf, uint16 bmsize, bool initpage)
Definition: hashovfl.c:740
#define LH_BUCKET_PAGE
Definition: hash.h:54
BlockNumber RelationGetNumberOfBlocksInFork(Relation relation, ForkNumber forkNum)
Definition: bufmgr.c:2800
RelFileNode rd_node
Definition: rel.h:54
#define RelationGetTargetPageUsage(relation, defaultff)
Definition: rel.h:300
uint64 XLogRecPtr
Definition: xlogdefs.h:21
void _hash_initbuf(Buffer buf, uint32 max_bucket, uint32 num_bucket, uint32 flag, bool initpage)
Definition: hashpage.c:155
#define MAXALIGN(LEN)
Definition: c.h:686
#define RelationNeedsWAL(relation)
Definition: rel.h:525
uint32 hashm_maxbucket
Definition: hash.h:251
RegProcedure procid
Definition: hash_xlog.h:219
#define HashPageGetMeta(page)
Definition: hash.h:305
int errmsg(const char *fmt,...)
Definition: elog.c:822
XLogRecPtr log_newpage(RelFileNode *rnode, ForkNumber forkNum, BlockNumber blkno, Page page, bool page_std)
Definition: xloginsert.c:972
#define elog(elevel,...)
Definition: elog.h:228
int i
uint16 hashm_bmsize
Definition: hash.h:248
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:99
void XLogBeginInsert(void)
Definition: xloginsert.c:120
#define PageSetLSN(page, lsn)
Definition: bufpage.h:368
BlockNumber hashm_mapp[HASH_MAX_BITMAPS]
Definition: hash.h:261
int Buffer
Definition: buf.h:23
Pointer Page
Definition: bufpage.h:78
RegProcedure index_getprocid(Relation irel, AttrNumber attnum, uint16 procnum)
Definition: indexam.c:760

◆ _hash_init_metabuffer()

void _hash_init_metabuffer ( Buffer  buf,
double  num_tuples,
RegProcedure  procid,
uint16  ffactor,
bool  initpage 
)

Definition at line 496 of file hashpage.c.

References _hash_get_totalbuckets(), _hash_log2(), _hash_pageinit(), _hash_spareindex(), Assert, BMPG_MASK, BMPG_SHIFT, BufferGetPage, BufferGetPageSize, BYTE_TO_BIT, HASH_MAGIC, HASH_MAX_SPLITPOINTS, HASH_VERSION, HashGetMaxBitmapSize, HashMetaPageData::hashm_bmshift, HashMetaPageData::hashm_bmsize, HashMetaPageData::hashm_bsize, HashMetaPageData::hashm_ffactor, HashMetaPageData::hashm_firstfree, HashMetaPageData::hashm_highmask, HashMetaPageData::hashm_lowmask, HashMetaPageData::hashm_magic, HashMetaPageData::hashm_mapp, HashMetaPageData::hashm_maxbucket, HashMetaPageData::hashm_nmaps, HashMetaPageData::hashm_ntuples, HashMetaPageData::hashm_ovflpoint, HashMetaPageData::hashm_procid, HashMetaPageData::hashm_spares, HashMetaPageData::hashm_version, HashPageOpaqueData::hasho_bucket, HashPageOpaqueData::hasho_flag, HashPageOpaqueData::hasho_nextblkno, HashPageOpaqueData::hasho_page_id, HASHO_PAGE_ID, HashPageOpaqueData::hasho_prevblkno, HashPageGetMeta, i, InvalidBlockNumber, LH_META_PAGE, MemSet, and PageGetSpecialPointer.

Referenced by _hash_init(), and hash_xlog_init_meta_page().

498 {
499  HashMetaPage metap;
500  HashPageOpaque pageopaque;
501  Page page;
502  double dnumbuckets;
503  uint32 num_buckets;
504  uint32 spare_index;
505  uint32 i;
506 
507  /*
508  * Choose the number of initial bucket pages to match the fill factor
509  * given the estimated number of tuples. We round up the result to the
510  * total number of buckets which has to be allocated before using its
511  * hashm_spares element. However always force at least 2 bucket pages. The
512  * upper limit is determined by considerations explained in
513  * _hash_expandtable().
514  */
515  dnumbuckets = num_tuples / ffactor;
516  if (dnumbuckets <= 2.0)
517  num_buckets = 2;
518  else if (dnumbuckets >= (double) 0x40000000)
519  num_buckets = 0x40000000;
520  else
521  num_buckets = _hash_get_totalbuckets(_hash_spareindex(dnumbuckets));
522 
523  spare_index = _hash_spareindex(num_buckets);
524  Assert(spare_index < HASH_MAX_SPLITPOINTS);
525 
526  page = BufferGetPage(buf);
527  if (initpage)
529 
530  pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
531  pageopaque->hasho_prevblkno = InvalidBlockNumber;
532  pageopaque->hasho_nextblkno = InvalidBlockNumber;
533  pageopaque->hasho_bucket = -1;
534  pageopaque->hasho_flag = LH_META_PAGE;
535  pageopaque->hasho_page_id = HASHO_PAGE_ID;
536 
537  metap = HashPageGetMeta(page);
538 
539  metap->hashm_magic = HASH_MAGIC;
540  metap->hashm_version = HASH_VERSION;
541  metap->hashm_ntuples = 0;
542  metap->hashm_nmaps = 0;
543  metap->hashm_ffactor = ffactor;
544  metap->hashm_bsize = HashGetMaxBitmapSize(page);
545  /* find largest bitmap array size that will fit in page size */
546  for (i = _hash_log2(metap->hashm_bsize); i > 0; --i)
547  {
548  if ((1 << i) <= metap->hashm_bsize)
549  break;
550  }
551  Assert(i > 0);
552  metap->hashm_bmsize = 1 << i;
553  metap->hashm_bmshift = i + BYTE_TO_BIT;
554  Assert((1 << BMPG_SHIFT(metap)) == (BMPG_MASK(metap) + 1));
555 
556  /*
557  * Label the index with its primary hash support function's OID. This is
558  * pretty useless for normal operation (in fact, hashm_procid is not used
559  * anywhere), but it might be handy for forensic purposes so we keep it.
560  */
561  metap->hashm_procid = procid;
562 
563  /*
564  * We initialize the index with N buckets, 0 .. N-1, occupying physical
565  * blocks 1 to N. The first freespace bitmap page is in block N+1.
566  */
567  metap->hashm_maxbucket = num_buckets - 1;
568 
569  /*
570  * Set highmask as next immediate ((2 ^ x) - 1), which should be
571  * sufficient to cover num_buckets.
572  */
573  metap->hashm_highmask = (1 << (_hash_log2(num_buckets + 1))) - 1;
574  metap->hashm_lowmask = (metap->hashm_highmask >> 1);
575 
576  MemSet(metap->hashm_spares, 0, sizeof(metap->hashm_spares));
577  MemSet(metap->hashm_mapp, 0, sizeof(metap->hashm_mapp));
578 
579  /* Set up mapping for one spare page after the initial splitpoints */
580  metap->hashm_spares[spare_index] = 1;
581  metap->hashm_ovflpoint = spare_index;
582  metap->hashm_firstfree = 0;
583 
584  /*
585  * Set pd_lower just past the end of the metadata. This is essential,
586  * because without doing so, metadata will be lost if xlog.c compresses
587  * the page.
588  */
589  ((PageHeader) page)->pd_lower =
590  ((char *) metap + sizeof(HashMetaPageData)) - (char *) page;
591 }
uint16 hashm_bmshift
Definition: hash.h:250
uint16 hasho_page_id
Definition: hash.h:82
RegProcedure hashm_procid
Definition: hash.h:258
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:597
#define LH_META_PAGE
Definition: hash.h:56
#define BYTE_TO_BIT
Definition: hash.h:283
uint32 hashm_magic
Definition: hash.h:243
uint16 hashm_ffactor
Definition: hash.h:246
uint32 hashm_highmask
Definition: hash.h:252
#define MemSet(start, val, len)
Definition: c.h:956
#define HASH_VERSION
Definition: hash.h:198
uint32 hashm_lowmask
Definition: hash.h:253
#define HASH_MAX_SPLITPOINTS
Definition: hash.h:236
#define HASH_MAGIC
Definition: hash.h:197
#define HashGetMaxBitmapSize(page)
Definition: hash.h:301
BlockNumber hasho_prevblkno
Definition: hash.h:78
uint32 _hash_spareindex(uint32 num_bucket)
Definition: hashutil.c:157
uint32 _hash_get_totalbuckets(uint32 splitpoint_phase)
Definition: hashutil.c:189
uint32 hashm_version
Definition: hash.h:244
uint32 hashm_nmaps
Definition: hash.h:257
static char * buf
Definition: pg_test_fsync.c:67
#define BMPG_MASK(metap)
Definition: hash.h:296
unsigned int uint32
Definition: c.h:359
#define BMPG_SHIFT(metap)
Definition: hash.h:295
#define BufferGetPage(buffer)
Definition: bufmgr.h:159
uint32 hashm_ovflpoint
Definition: hash.h:254
uint16 hashm_bsize
Definition: hash.h:247
double hashm_ntuples
Definition: hash.h:245
#define BufferGetPageSize(buffer)
Definition: bufmgr.h:146
uint32 hashm_firstfree
Definition: hash.h:256
uint32 hashm_spares[HASH_MAX_SPLITPOINTS]
Definition: hash.h:259
PageHeaderData * PageHeader
Definition: bufpage.h:166
#define Assert(condition)
Definition: c.h:733
Bucket hasho_bucket
Definition: hash.h:80
#define PageGetSpecialPointer(page)
Definition: bufpage.h:326
#define InvalidBlockNumber
Definition: block.h:33
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:85
#define HASHO_PAGE_ID
Definition: hash.h:98
uint32 hashm_maxbucket
Definition: hash.h:251
uint16 hasho_flag
Definition: hash.h:81
#define HashPageGetMeta(page)
Definition: hash.h:305
uint32 _hash_log2(uint32 num)
Definition: hashutil.c:141
int i
BlockNumber hasho_nextblkno
Definition: hash.h:79
uint16 hashm_bmsize
Definition: hash.h:248
BlockNumber hashm_mapp[HASH_MAX_BITMAPS]
Definition: hash.h:261
Pointer Page
Definition: bufpage.h:78

◆ _hash_initbuf()

void _hash_initbuf ( Buffer  buf,
uint32  max_bucket,
uint32  num_bucket,
uint32  flag,
bool  initpage 
)

Definition at line 155 of file hashpage.c.

References _hash_pageinit(), BufferGetPage, BufferGetPageSize, flag(), HashPageOpaqueData::hasho_bucket, HashPageOpaqueData::hasho_flag, HashPageOpaqueData::hasho_nextblkno, HashPageOpaqueData::hasho_page_id, HASHO_PAGE_ID, HashPageOpaqueData::hasho_prevblkno, InvalidBlockNumber, and PageGetSpecialPointer.

Referenced by _hash_init(), hash_xlog_add_ovfl_page(), and hash_xlog_split_allocate_page().

157 {
158  HashPageOpaque pageopaque;
159  Page page;
160 
161  page = BufferGetPage(buf);
162 
163  /* initialize the page */
164  if (initpage)
166 
167  pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
168 
169  /*
170  * Set hasho_prevblkno with current hashm_maxbucket. This value will be
171  * used to validate cached HashMetaPageData. See
172  * _hash_getbucketbuf_from_hashkey().
173  */
174  pageopaque->hasho_prevblkno = max_bucket;
175  pageopaque->hasho_nextblkno = InvalidBlockNumber;
176  pageopaque->hasho_bucket = num_bucket;
177  pageopaque->hasho_flag = flag;
178  pageopaque->hasho_page_id = HASHO_PAGE_ID;
179 }
uint16 hasho_page_id
Definition: hash.h:82
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:597
BlockNumber hasho_prevblkno
Definition: hash.h:78
static char * buf
Definition: pg_test_fsync.c:67
char * flag(int b)
Definition: test-ctype.c:33
#define BufferGetPage(buffer)
Definition: bufmgr.h:159
#define BufferGetPageSize(buffer)
Definition: bufmgr.h:146
Bucket hasho_bucket
Definition: hash.h:80
#define PageGetSpecialPointer(page)
Definition: bufpage.h:326
#define InvalidBlockNumber
Definition: block.h:33
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:85
#define HASHO_PAGE_ID
Definition: hash.h:98
uint16 hasho_flag
Definition: hash.h:81
BlockNumber hasho_nextblkno
Definition: hash.h:79
Pointer Page
Definition: bufpage.h:78

◆ _hash_pageinit()

void _hash_pageinit ( Page  page,
Size  size 
)

Definition at line 597 of file hashpage.c.

References PageInit().

Referenced by _hash_alloc_buckets(), _hash_freeovflpage(), _hash_getinitbuf(), _hash_getnewbuf(), _hash_init_metabuffer(), _hash_initbitmapbuffer(), _hash_initbuf(), and hash_xlog_squeeze_page().

598 {
599  PageInit(page, size, sizeof(HashPageOpaqueData));
600 }
void PageInit(Page page, Size pageSize, Size specialSize)
Definition: bufpage.c:42

◆ _hash_relbuf()

◆ _hash_splitbucket()

static void _hash_splitbucket ( Relation  rel,
Buffer  metabuf,
Bucket  obucket,
Bucket  nbucket,
Buffer  obuf,
Buffer  nbuf,
HTAB htab,
uint32  maxbucket,
uint32  highmask,
uint32  lowmask 
)
static

Definition at line 1070 of file hashpage.c.

References _hash_addovflpage(), _hash_get_indextuple_hashkey(), _hash_getbuf(), _hash_hashkey2bucket(), _hash_pgaddmultitup(), _hash_relbuf(), Assert, BlockNumberIsValid, BUFFER_LOCK_EXCLUSIVE, BUFFER_LOCK_UNLOCK, BufferGetBlockNumber(), BufferGetPage, CopyIndexTuple(), END_CRIT_SECTION, FirstOffsetNumber, HASH_FIND, HASH_READ, hash_search(), hashbucketcleanup(), HashPageOpaqueData::hasho_flag, HashPageOpaqueData::hasho_nextblkno, i, INDEX_MOVED_BY_SPLIT_MASK, IndexTupleSize, IsBufferCleanupOK(), ItemIdIsDead, LH_BUCKET_BEING_POPULATED, LH_BUCKET_BEING_SPLIT, LH_BUCKET_NEEDS_SPLIT_CLEANUP, LH_OVERFLOW_PAGE, LockBuffer(), log_split_page(), MarkBufferDirty(), MAXALIGN, MaxIndexTuplesPerPage, xl_hash_split_complete::new_bucket_flag, OffsetNumberNext, xl_hash_split_complete::old_bucket_flag, PageGetFreeSpaceForMultipleTuples(), PageGetItem, PageGetItemId, PageGetMaxOffsetNumber, PageGetSpecialPointer, PageSetLSN, pfree(), PredicateLockPageSplit(), REGBUF_STANDARD, RelationNeedsWAL, SizeOfHashSplitComplete, START_CRIT_SECTION, IndexTupleData::t_info, IndexTupleData::t_tid, XLOG_HASH_SPLIT_COMPLETE, XLogBeginInsert(), XLogInsert(), XLogRegisterBuffer(), and XLogRegisterData().

Referenced by _hash_expandtable(), and _hash_finish_split().

1080 {
1081  Buffer bucket_obuf;
1082  Buffer bucket_nbuf;
1083  Page opage;
1084  Page npage;
1085  HashPageOpaque oopaque;
1086  HashPageOpaque nopaque;
1087  OffsetNumber itup_offsets[MaxIndexTuplesPerPage];
1089  Size all_tups_size = 0;
1090  int i;
1091  uint16 nitups = 0;
1092 
1093  bucket_obuf = obuf;
1094  opage = BufferGetPage(obuf);
1095  oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
1096 
1097  bucket_nbuf = nbuf;
1098  npage = BufferGetPage(nbuf);
1099  nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
1100 
1101  /* Copy the predicate locks from old bucket to new bucket. */
1103  BufferGetBlockNumber(bucket_obuf),
1104  BufferGetBlockNumber(bucket_nbuf));
1105 
1106  /*
1107  * Partition the tuples in the old bucket between the old bucket and the
1108  * new bucket, advancing along the old bucket's overflow bucket chain and
1109  * adding overflow pages to the new bucket as needed. Outer loop iterates
1110  * once per page in old bucket.
1111  */
1112  for (;;)
1113  {
1114  BlockNumber oblkno;
1115  OffsetNumber ooffnum;
1116  OffsetNumber omaxoffnum;
1117 
1118  /* Scan each tuple in old page */
1119  omaxoffnum = PageGetMaxOffsetNumber(opage);
1120  for (ooffnum = FirstOffsetNumber;
1121  ooffnum <= omaxoffnum;
1122  ooffnum = OffsetNumberNext(ooffnum))
1123  {
1124  IndexTuple itup;
1125  Size itemsz;
1126  Bucket bucket;
1127  bool found = false;
1128 
1129  /* skip dead tuples */
1130  if (ItemIdIsDead(PageGetItemId(opage, ooffnum)))
1131  continue;
1132 
1133  /*
1134  * Before inserting a tuple, probe the hash table containing TIDs
1135  * of tuples belonging to new bucket, if we find a match, then
1136  * skip that tuple, else fetch the item's hash key (conveniently
1137  * stored in the item) and determine which bucket it now belongs
1138  * in.
1139  */
1140  itup = (IndexTuple) PageGetItem(opage,
1141  PageGetItemId(opage, ooffnum));
1142 
1143  if (htab)
1144  (void) hash_search(htab, &itup->t_tid, HASH_FIND, &found);
1145 
1146  if (found)
1147  continue;
1148 
1150  maxbucket, highmask, lowmask);
1151 
1152  if (bucket == nbucket)
1153  {
1154  IndexTuple new_itup;
1155 
1156  /*
1157  * make a copy of index tuple as we have to scribble on it.
1158  */
1159  new_itup = CopyIndexTuple(itup);
1160 
1161  /*
1162  * mark the index tuple as moved by split, such tuples are
1163  * skipped by scan if there is split in progress for a bucket.
1164  */
1165  new_itup->t_info |= INDEX_MOVED_BY_SPLIT_MASK;
1166 
1167  /*
1168  * insert the tuple into the new bucket. if it doesn't fit on
1169  * the current page in the new bucket, we must allocate a new
1170  * overflow page and place the tuple on that page instead.
1171  */
1172  itemsz = IndexTupleSize(new_itup);
1173  itemsz = MAXALIGN(itemsz);
1174 
1175  if (PageGetFreeSpaceForMultipleTuples(npage, nitups + 1) < (all_tups_size + itemsz))
1176  {
1177  /*
1178  * Change the shared buffer state in critical section,
1179  * otherwise any error could make it unrecoverable.
1180  */
1182 
1183  _hash_pgaddmultitup(rel, nbuf, itups, itup_offsets, nitups);
1184  MarkBufferDirty(nbuf);
1185  /* log the split operation before releasing the lock */
1186  log_split_page(rel, nbuf);
1187 
1188  END_CRIT_SECTION();
1189 
1190  /* drop lock, but keep pin */
1192 
1193  /* be tidy */
1194  for (i = 0; i < nitups; i++)
1195  pfree(itups[i]);
1196  nitups = 0;
1197  all_tups_size = 0;
1198 
1199  /* chain to a new overflow page */
1200  nbuf = _hash_addovflpage(rel, metabuf, nbuf, (nbuf == bucket_nbuf) ? true : false);
1201  npage = BufferGetPage(nbuf);
1202  nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
1203  }
1204 
1205  itups[nitups++] = new_itup;
1206  all_tups_size += itemsz;
1207  }
1208  else
1209  {
1210  /*
1211  * the tuple stays on this page, so nothing to do.
1212  */
1213  Assert(bucket == obucket);
1214  }
1215  }
1216 
1217  oblkno = oopaque->hasho_nextblkno;
1218 
1219  /* retain the pin on the old primary bucket */
1220  if (obuf == bucket_obuf)
1222  else
1223  _hash_relbuf(rel, obuf);
1224 
1225  /* Exit loop if no more overflow pages in old bucket */
1226  if (!BlockNumberIsValid(oblkno))
1227  {
1228  /*
1229  * Change the shared buffer state in critical section, otherwise
1230  * any error could make it unrecoverable.
1231  */
1233 
1234  _hash_pgaddmultitup(rel, nbuf, itups, itup_offsets, nitups);
1235  MarkBufferDirty(nbuf);
1236  /* log the split operation before releasing the lock */
1237  log_split_page(rel, nbuf);
1238 
1239  END_CRIT_SECTION();
1240 
1241  if (nbuf == bucket_nbuf)
1243  else
1244  _hash_relbuf(rel, nbuf);
1245 
1246  /* be tidy */
1247  for (i = 0; i < nitups; i++)
1248  pfree(itups[i]);
1249  break;
1250  }
1251 
1252  /* Else, advance to next old page */
1253  obuf = _hash_getbuf(rel, oblkno, HASH_READ, LH_OVERFLOW_PAGE);
1254  opage = BufferGetPage(obuf);
1255  oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
1256  }
1257 
1258  /*
1259  * We're at the end of the old bucket chain, so we're done partitioning
1260  * the tuples. Mark the old and new buckets to indicate split is
1261  * finished.
1262  *
1263  * To avoid deadlocks due to locking order of buckets, first lock the old
1264  * bucket and then the new bucket.
1265  */
1266  LockBuffer(bucket_obuf, BUFFER_LOCK_EXCLUSIVE);
1267  opage = BufferGetPage(bucket_obuf);
1268  oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
1269 
1270  LockBuffer(bucket_nbuf, BUFFER_LOCK_EXCLUSIVE);
1271  npage = BufferGetPage(bucket_nbuf);
1272  nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
1273 
1275 
1276  oopaque->hasho_flag &= ~LH_BUCKET_BEING_SPLIT;
1278 
1279  /*
1280  * After the split is finished, mark the old bucket to indicate that it
1281  * contains deletable tuples. We will clear split-cleanup flag after
1282  * deleting such tuples either at the end of split or at the next split
1283  * from old bucket or at the time of vacuum.
1284  */
1286 
1287  /*
1288  * now write the buffers, here we don't release the locks as caller is
1289  * responsible to release locks.
1290  */
1291  MarkBufferDirty(bucket_obuf);
1292  MarkBufferDirty(bucket_nbuf);
1293 
1294  if (RelationNeedsWAL(rel))
1295  {
1296  XLogRecPtr recptr;
1297  xl_hash_split_complete xlrec;
1298 
1299  xlrec.old_bucket_flag = oopaque->hasho_flag;
1300  xlrec.new_bucket_flag = nopaque->hasho_flag;
1301 
1302  XLogBeginInsert();
1303 
1304  XLogRegisterData((char *) &xlrec, SizeOfHashSplitComplete);
1305 
1306  XLogRegisterBuffer(0, bucket_obuf, REGBUF_STANDARD);
1307  XLogRegisterBuffer(1, bucket_nbuf, REGBUF_STANDARD);
1308 
1309  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_COMPLETE);
1310 
1311  PageSetLSN(BufferGetPage(bucket_obuf), recptr);
1312  PageSetLSN(BufferGetPage(bucket_nbuf), recptr);
1313  }
1314 
1315  END_CRIT_SECTION();
1316 
1317  /*
1318  * If possible, clean up the old bucket. We might not be able to do this
1319  * if someone else has a pin on it, but if not then we can go ahead. This
1320  * isn't absolutely necessary, but it reduces bloat; if we don't do it
1321  * now, VACUUM will do it eventually, but maybe not until new overflow
1322  * pages have been allocated. Note that there's no need to clean up the
1323  * new bucket.
1324  */
1325  if (IsBufferCleanupOK(bucket_obuf))
1326  {
1327  LockBuffer(bucket_nbuf, BUFFER_LOCK_UNLOCK);
1328  hashbucketcleanup(rel, obucket, bucket_obuf,
1329  BufferGetBlockNumber(bucket_obuf), NULL,
1330  maxbucket, highmask, lowmask, NULL, NULL, true,
1331  NULL, NULL);
1332  }
1333  else
1334  {
1335  LockBuffer(bucket_nbuf, BUFFER_LOCK_UNLOCK);
1336  LockBuffer(bucket_obuf, BUFFER_LOCK_UNLOCK);
1337  }
1338 }
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:86
void _hash_pgaddmultitup(Relation rel, Buffer buf, IndexTuple *itups, OffsetNumber *itup_offsets, uint16 nitups)
Definition: hashinsert.c:299
Bucket _hash_hashkey2bucket(uint32 hashkey, uint32 maxbucket, uint32 highmask, uint32 lowmask)
Definition: hashutil.c:125
void MarkBufferDirty(Buffer buffer)
Definition: bufmgr.c:1458
void XLogRegisterBuffer(uint8 block_id, Buffer buffer, uint8 flags)
Definition: xloginsert.c:213
ItemPointerData t_tid
Definition: itup.h:37
#define END_CRIT_SECTION()
Definition: miscadmin.h:134
#define START_CRIT_SECTION()
Definition: miscadmin.h:132
void hashbucketcleanup(Relation rel, Bucket cur_bucket, Buffer bucket_buf, BlockNumber bucket_blkno, BufferAccessStrategy bstrategy, uint32 maxbucket, uint32 highmask, uint32 lowmask, double *tuples_removed, double *num_index_tuples, bool split_cleanup, IndexBulkDeleteCallback callback, void *callback_state)
Definition: hash.c:677
uint32 BlockNumber
Definition: block.h:31
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:906
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:68
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:88
#define LH_BUCKET_NEEDS_SPLIT_CLEANUP
Definition: hash.h:59
#define ItemIdIsDead(itemId)
Definition: itemid.h:113
#define PageGetMaxOffsetNumber(page)
Definition: bufpage.h:357
uint16 OffsetNumber
Definition: off.h:24
#define HASH_READ
Definition: hash.h:321
uint32 Bucket
Definition: hash.h:34
unsigned short uint16
Definition: c.h:358
void pfree(void *pointer)
Definition: mcxt.c:1056
uint32 _hash_get_indextuple_hashkey(IndexTuple itup)
Definition: hashutil.c:299
IndexTuple CopyIndexTuple(IndexTuple source)
Definition: indextuple.c:509
#define FirstOffsetNumber
Definition: off.h:27
IndexTupleData * IndexTuple
Definition: itup.h:53
#define REGBUF_STANDARD
Definition: xloginsert.h:35
#define LH_BUCKET_BEING_SPLIT
Definition: hash.h:58
#define BufferGetPage(buffer)
Definition: bufmgr.h:159
bool IsBufferCleanupOK(Buffer buffer)
Definition: bufmgr.c:3830
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:235
void XLogRegisterData(char *data, int len)
Definition: xloginsert.c:323
#define LH_BUCKET_BEING_POPULATED
Definition: hash.h:57
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:415
#define LH_OVERFLOW_PAGE
Definition: hash.h:53
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3602
Size PageGetFreeSpaceForMultipleTuples(Page page, int ntups)
Definition: bufpage.c:608
static void log_split_page(Relation rel, Buffer buf)
Definition: hashpage.c:1472
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:264
#define BlockNumberIsValid(blockNumber)
Definition: block.h:70
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:733
#define SizeOfHashSplitComplete
Definition: hash_xlog.h:123
#define OffsetNumberNext(offsetNumber)
Definition: off.h:52
size_t Size
Definition: c.h:467
#define PageGetSpecialPointer(page)
Definition: bufpage.h:326
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:85
Buffer _hash_addovflpage(Relation rel, Buffer metabuf, Buffer buf, bool retain_pin)
Definition: hashovfl.c:111
#define MAXALIGN(LEN)
Definition: c.h:686
#define RelationNeedsWAL(relation)
Definition: rel.h:525
#define XLOG_HASH_SPLIT_COMPLETE
Definition: hash_xlog.h:33
uint16 hasho_flag
Definition: hash.h:81
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:2613
#define MaxIndexTuplesPerPage
Definition: itup.h:145
int i
BlockNumber hasho_nextblkno
Definition: hash.h:79
unsigned short t_info
Definition: itup.h:49
#define INDEX_MOVED_BY_SPLIT_MASK
Definition: hash.h:275
void XLogBeginInsert(void)
Definition: xloginsert.c:120
#define PageSetLSN(page, lsn)
Definition: bufpage.h:368
int Buffer
Definition: buf.h:23
void PredicateLockPageSplit(Relation relation, BlockNumber oldblkno, BlockNumber newblkno)
Definition: predicate.c:3102
#define PageGetItem(page, itemId)
Definition: bufpage.h:340
Pointer Page
Definition: bufpage.h:78
#define IndexTupleSize(itup)
Definition: itup.h:71

◆ log_split_page()

static void log_split_page ( Relation  rel,
Buffer  buf 
)
static

Definition at line 1472 of file hashpage.c.

References BufferGetPage, PageSetLSN, REGBUF_FORCE_IMAGE, REGBUF_STANDARD, RelationNeedsWAL, XLOG_HASH_SPLIT_PAGE, XLogBeginInsert(), XLogInsert(), and XLogRegisterBuffer().

Referenced by _hash_splitbucket().

1473 {
1474  if (RelationNeedsWAL(rel))
1475  {
1476  XLogRecPtr recptr;
1477 
1478  XLogBeginInsert();
1479 
1481 
1482  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_PAGE);
1483 
1484  PageSetLSN(BufferGetPage(buf), recptr);
1485  }
1486 }
void XLogRegisterBuffer(uint8 block_id, Buffer buffer, uint8 flags)
Definition: xloginsert.c:213
#define XLOG_HASH_SPLIT_PAGE
Definition: hash_xlog.h:32
static char * buf
Definition: pg_test_fsync.c:67
#define REGBUF_STANDARD
Definition: xloginsert.h:35
#define BufferGetPage(buffer)
Definition: bufmgr.h:159
#define REGBUF_FORCE_IMAGE
Definition: xloginsert.h:31
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:415
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define RelationNeedsWAL(relation)
Definition: rel.h:525
void XLogBeginInsert(void)
Definition: xloginsert.c:120
#define PageSetLSN(page, lsn)
Definition: bufpage.h:368