PostgreSQL Source Code  git master
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros
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/smgr.h"
Include dependency graph for hashpage.c:

Go to the source code of this file.

Macros

#define USELOCKING(rel)   (!RELATION_IS_LOCAL(rel))
 

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)
 

Macro Definition Documentation

#define USELOCKING (   rel)    (!RELATION_IS_LOCAL(rel))

Definition at line 57 of file hashpage.c.

Function Documentation

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

Definition at line 998 of file hashpage.c.

References _hash_pageinit(), 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, RelationData::rd_node, RelationData::rd_smgr, RelationNeedsWAL, RelationOpenSmgr, and smgrextend().

Referenced by _hash_expandtable().

999 {
1000  BlockNumber lastblock;
1001  char zerobuf[BLCKSZ];
1002  Page page;
1003  HashPageOpaque ovflopaque;
1004 
1005  lastblock = firstblock + nblocks - 1;
1006 
1007  /*
1008  * Check for overflow in block number calculation; if so, we cannot extend
1009  * the index anymore.
1010  */
1011  if (lastblock < firstblock || lastblock == InvalidBlockNumber)
1012  return false;
1013 
1014  page = (Page) zerobuf;
1015 
1016  /*
1017  * Initialize the page. Just zeroing the page won't work; see
1018  * _hash_freeovflpage for similar usage. We take care to make the special
1019  * space valid for the benefit of tools such as pageinspect.
1020  */
1021  _hash_pageinit(page, BLCKSZ);
1022 
1023  ovflopaque = (HashPageOpaque) PageGetSpecialPointer(page);
1024 
1025  ovflopaque->hasho_prevblkno = InvalidBlockNumber;
1026  ovflopaque->hasho_nextblkno = InvalidBlockNumber;
1027  ovflopaque->hasho_bucket = -1;
1028  ovflopaque->hasho_flag = LH_UNUSED_PAGE;
1029  ovflopaque->hasho_page_id = HASHO_PAGE_ID;
1030 
1031  if (RelationNeedsWAL(rel))
1032  log_newpage(&rel->rd_node,
1033  MAIN_FORKNUM,
1034  lastblock,
1035  zerobuf,
1036  true);
1037 
1038  RelationOpenSmgr(rel);
1039  smgrextend(rel->rd_smgr, MAIN_FORKNUM, lastblock, zerobuf, false);
1040 
1041  return true;
1042 }
uint16 hasho_page_id
Definition: hash.h:93
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:606
struct SMgrRelationData * rd_smgr
Definition: rel.h:87
uint32 BlockNumber
Definition: block.h:31
#define LH_UNUSED_PAGE
Definition: hash.h:63
#define RelationOpenSmgr(relation)
Definition: rel.h:460
BlockNumber hasho_prevblkno
Definition: hash.h:89
RelFileNode rd_node
Definition: rel.h:85
Bucket hasho_bucket
Definition: hash.h:91
#define PageGetSpecialPointer(page)
Definition: bufpage.h:322
#define InvalidBlockNumber
Definition: block.h:33
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:96
#define HASHO_PAGE_ID
Definition: hash.h:109
#define RelationNeedsWAL(relation)
Definition: rel.h:505
void smgrextend(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool skipFsync)
Definition: smgr.c:600
uint16 hasho_flag
Definition: hash.h:92
XLogRecPtr log_newpage(RelFileNode *rnode, ForkNumber forkNum, BlockNumber blkno, Page page, bool page_std)
Definition: xloginsert.c:972
BlockNumber hasho_nextblkno
Definition: hash.h:90
Pointer Page
Definition: bufpage.h:74
void _hash_dropbuf ( Relation  rel,
Buffer  buf 
)

Definition at line 285 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().

286 {
288 }
void ReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:3309
static char * buf
Definition: pg_test_fsync.c:67
void _hash_dropscanbuf ( Relation  rel,
HashScanOpaque  so 
)

Definition at line 297 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().

298 {
299  /* release pin we hold on primary bucket page */
300  if (BufferIsValid(so->hashso_bucket_buf) &&
301  so->hashso_bucket_buf != so->currPos.buf)
304 
305  /* release pin we hold on primary bucket page of bucket being split */
310 
311  /* release any pin we still hold */
312  if (BufferIsValid(so->currPos.buf))
313  _hash_dropbuf(rel, so->currPos.buf);
314  so->currPos.buf = InvalidBuffer;
315 
316  /* reset split scan */
317  so->hashso_buc_populated = false;
318  so->hashso_buc_split = false;
319 }
#define InvalidBuffer
Definition: buf.h:25
void _hash_dropbuf(Relation rel, Buffer buf)
Definition: hashpage.c:285
Buffer hashso_bucket_buf
Definition: hash.h:174
bool hashso_buc_populated
Definition: hash.h:184
Buffer buf
Definition: hash.h:119
#define BufferIsValid(bufnum)
Definition: bufmgr.h:114
HashScanPosData currPos
Definition: hash.h:199
bool hashso_buc_split
Definition: hash.h:190
Buffer hashso_split_bucket_buf
Definition: hash.h:181
void _hash_expandtable ( Relation  rel,
Buffer  metabuf 
)

Definition at line 624 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().

625 {
626  HashMetaPage metap;
627  Bucket old_bucket;
628  Bucket new_bucket;
629  uint32 spare_ndx;
630  BlockNumber start_oblkno;
631  BlockNumber start_nblkno;
632  Buffer buf_nblkno;
633  Buffer buf_oblkno;
634  Page opage;
635  Page npage;
636  HashPageOpaque oopaque;
637  HashPageOpaque nopaque;
638  uint32 maxbucket;
639  uint32 highmask;
640  uint32 lowmask;
641  bool metap_update_masks = false;
642  bool metap_update_splitpoint = false;
643 
644 restart_expand:
645 
646  /*
647  * Write-lock the meta page. It used to be necessary to acquire a
648  * heavyweight lock to begin a split, but that is no longer required.
649  */
651 
652  _hash_checkpage(rel, metabuf, LH_META_PAGE);
653  metap = HashPageGetMeta(BufferGetPage(metabuf));
654 
655  /*
656  * Check to see if split is still needed; someone else might have already
657  * done one while we waited for the lock.
658  *
659  * Make sure this stays in sync with _hash_doinsert()
660  */
661  if (metap->hashm_ntuples <=
662  (double) metap->hashm_ffactor * (metap->hashm_maxbucket + 1))
663  goto fail;
664 
665  /*
666  * Can't split anymore if maxbucket has reached its maximum possible
667  * value.
668  *
669  * Ideally we'd allow bucket numbers up to UINT_MAX-1 (no higher because
670  * the calculation maxbucket+1 mustn't overflow). Currently we restrict
671  * to half that because of overflow looping in _hash_log2() and
672  * insufficient space in hashm_spares[]. It's moot anyway because an
673  * index with 2^32 buckets would certainly overflow BlockNumber and hence
674  * _hash_alloc_buckets() would fail, but if we supported buckets smaller
675  * than a disk block then this would be an independent constraint.
676  *
677  * If you change this, see also the maximum initial number of buckets in
678  * _hash_init().
679  */
680  if (metap->hashm_maxbucket >= (uint32) 0x7FFFFFFE)
681  goto fail;
682 
683  /*
684  * Determine which bucket is to be split, and attempt to take cleanup lock
685  * on the old bucket. If we can't get the lock, give up.
686  *
687  * The cleanup lock protects us not only against other backends, but
688  * against our own backend as well.
689  *
690  * The cleanup lock is mainly to protect the split from concurrent
691  * inserts. See src/backend/access/hash/README, Lock Definitions for
692  * further details. Due to this locking restriction, if there is any
693  * pending scan, the split will give up which is not good, but harmless.
694  */
695  new_bucket = metap->hashm_maxbucket + 1;
696 
697  old_bucket = (new_bucket & metap->hashm_lowmask);
698 
699  start_oblkno = BUCKET_TO_BLKNO(metap, old_bucket);
700 
701  buf_oblkno = _hash_getbuf_with_condlock_cleanup(rel, start_oblkno, LH_BUCKET_PAGE);
702  if (!buf_oblkno)
703  goto fail;
704 
705  opage = BufferGetPage(buf_oblkno);
706  oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
707 
708  /*
709  * We want to finish the split from a bucket as there is no apparent
710  * benefit by not doing so and it will make the code complicated to finish
711  * the split that involves multiple buckets considering the case where new
712  * split also fails. We don't need to consider the new bucket for
713  * completing the split here as it is not possible that a re-split of new
714  * bucket starts when there is still a pending split from old bucket.
715  */
716  if (H_BUCKET_BEING_SPLIT(oopaque))
717  {
718  /*
719  * Copy bucket mapping info now; refer the comment in code below where
720  * we copy this information before calling _hash_splitbucket to see
721  * why this is okay.
722  */
723  maxbucket = metap->hashm_maxbucket;
724  highmask = metap->hashm_highmask;
725  lowmask = metap->hashm_lowmask;
726 
727  /*
728  * Release the lock on metapage and old_bucket, before completing the
729  * split.
730  */
731  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
732  LockBuffer(buf_oblkno, BUFFER_LOCK_UNLOCK);
733 
734  _hash_finish_split(rel, metabuf, buf_oblkno, old_bucket, maxbucket,
735  highmask, lowmask);
736 
737  /* release the pin on old buffer and retry for expand. */
738  _hash_dropbuf(rel, buf_oblkno);
739 
740  goto restart_expand;
741  }
742 
743  /*
744  * Clean the tuples remained from the previous split. This operation
745  * requires cleanup lock and we already have one on the old bucket, so
746  * let's do it. We also don't want to allow further splits from the bucket
747  * till the garbage of previous split is cleaned. This has two
748  * advantages; first, it helps in avoiding the bloat due to garbage and
749  * second is, during cleanup of bucket, we are always sure that the
750  * garbage tuples belong to most recently split bucket. On the contrary,
751  * if we allow cleanup of bucket after meta page is updated to indicate
752  * the new split and before the actual split, the cleanup operation won't
753  * be able to decide whether the tuple has been moved to the newly created
754  * bucket and ended up deleting such tuples.
755  */
756  if (H_NEEDS_SPLIT_CLEANUP(oopaque))
757  {
758  /*
759  * Copy bucket mapping info now; refer to the comment in code below
760  * where we copy this information before calling _hash_splitbucket to
761  * see why this is okay.
762  */
763  maxbucket = metap->hashm_maxbucket;
764  highmask = metap->hashm_highmask;
765  lowmask = metap->hashm_lowmask;
766 
767  /* Release the metapage lock. */
768  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
769 
770  hashbucketcleanup(rel, old_bucket, buf_oblkno, start_oblkno, NULL,
771  maxbucket, highmask, lowmask, NULL, NULL, true,
772  NULL, NULL);
773 
774  _hash_dropbuf(rel, buf_oblkno);
775 
776  goto restart_expand;
777  }
778 
779  /*
780  * There shouldn't be any active scan on new bucket.
781  *
782  * Note: it is safe to compute the new bucket's blkno here, even though we
783  * may still need to update the BUCKET_TO_BLKNO mapping. This is because
784  * the current value of hashm_spares[hashm_ovflpoint] correctly shows
785  * where we are going to put a new splitpoint's worth of buckets.
786  */
787  start_nblkno = BUCKET_TO_BLKNO(metap, new_bucket);
788 
789  /*
790  * If the split point is increasing we need to allocate a new batch of
791  * bucket pages.
792  */
793  spare_ndx = _hash_spareindex(new_bucket + 1);
794  if (spare_ndx > metap->hashm_ovflpoint)
795  {
796  uint32 buckets_to_add;
797 
798  Assert(spare_ndx == metap->hashm_ovflpoint + 1);
799 
800  /*
801  * We treat allocation of buckets as a separate WAL-logged action.
802  * Even if we fail after this operation, won't leak bucket pages;
803  * rather, the next split will consume this space. In any case, even
804  * without failure we don't use all the space in one split operation.
805  */
806  buckets_to_add = _hash_get_totalbuckets(spare_ndx) - new_bucket;
807  if (!_hash_alloc_buckets(rel, start_nblkno, buckets_to_add))
808  {
809  /* can't split due to BlockNumber overflow */
810  _hash_relbuf(rel, buf_oblkno);
811  goto fail;
812  }
813  }
814 
815  /*
816  * Physically allocate the new bucket's primary page. We want to do this
817  * before changing the metapage's mapping info, in case we can't get the
818  * disk space. Ideally, we don't need to check for cleanup lock on new
819  * bucket as no other backend could find this bucket unless meta page is
820  * updated. However, it is good to be consistent with old bucket locking.
821  */
822  buf_nblkno = _hash_getnewbuf(rel, start_nblkno, MAIN_FORKNUM);
823  if (!IsBufferCleanupOK(buf_nblkno))
824  {
825  _hash_relbuf(rel, buf_oblkno);
826  _hash_relbuf(rel, buf_nblkno);
827  goto fail;
828  }
829 
830  /*
831  * Since we are scribbling on the pages in the shared buffers, establish a
832  * critical section. Any failure in this next code leaves us with a big
833  * problem: the metapage is effectively corrupt but could get written back
834  * to disk.
835  */
837 
838  /*
839  * Okay to proceed with split. Update the metapage bucket mapping info.
840  */
841  metap->hashm_maxbucket = new_bucket;
842 
843  if (new_bucket > metap->hashm_highmask)
844  {
845  /* Starting a new doubling */
846  metap->hashm_lowmask = metap->hashm_highmask;
847  metap->hashm_highmask = new_bucket | metap->hashm_lowmask;
848  metap_update_masks = true;
849  }
850 
851  /*
852  * If the split point is increasing we need to adjust the hashm_spares[]
853  * array and hashm_ovflpoint so that future overflow pages will be created
854  * beyond this new batch of bucket pages.
855  */
856  if (spare_ndx > metap->hashm_ovflpoint)
857  {
858  metap->hashm_spares[spare_ndx] = metap->hashm_spares[metap->hashm_ovflpoint];
859  metap->hashm_ovflpoint = spare_ndx;
860  metap_update_splitpoint = true;
861  }
862 
863  MarkBufferDirty(metabuf);
864 
865  /*
866  * Copy bucket mapping info now; this saves re-accessing the meta page
867  * inside _hash_splitbucket's inner loop. Note that once we drop the
868  * split lock, other splits could begin, so these values might be out of
869  * date before _hash_splitbucket finishes. That's okay, since all it
870  * needs is to tell which of these two buckets to map hashkeys into.
871  */
872  maxbucket = metap->hashm_maxbucket;
873  highmask = metap->hashm_highmask;
874  lowmask = metap->hashm_lowmask;
875 
876  opage = BufferGetPage(buf_oblkno);
877  oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
878 
879  /*
880  * Mark the old bucket to indicate that split is in progress. (At
881  * operation end, we will clear the split-in-progress flag.) Also, for a
882  * primary bucket page, hasho_prevblkno stores the number of buckets that
883  * existed as of the last split, so we must update that value here.
884  */
885  oopaque->hasho_flag |= LH_BUCKET_BEING_SPLIT;
886  oopaque->hasho_prevblkno = maxbucket;
887 
888  MarkBufferDirty(buf_oblkno);
889 
890  npage = BufferGetPage(buf_nblkno);
891 
892  /*
893  * initialize the new bucket's primary page and mark it to indicate that
894  * split is in progress.
895  */
896  nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
897  nopaque->hasho_prevblkno = maxbucket;
899  nopaque->hasho_bucket = new_bucket;
901  nopaque->hasho_page_id = HASHO_PAGE_ID;
902 
903  MarkBufferDirty(buf_nblkno);
904 
905  /* XLOG stuff */
906  if (RelationNeedsWAL(rel))
907  {
909  XLogRecPtr recptr;
910 
911  xlrec.new_bucket = maxbucket;
912  xlrec.old_bucket_flag = oopaque->hasho_flag;
913  xlrec.new_bucket_flag = nopaque->hasho_flag;
914  xlrec.flags = 0;
915 
916  XLogBeginInsert();
917 
918  XLogRegisterBuffer(0, buf_oblkno, REGBUF_STANDARD);
919  XLogRegisterBuffer(1, buf_nblkno, REGBUF_WILL_INIT);
920  XLogRegisterBuffer(2, metabuf, REGBUF_STANDARD);
921 
922  if (metap_update_masks)
923  {
925  XLogRegisterBufData(2, (char *) &metap->hashm_lowmask, sizeof(uint32));
926  XLogRegisterBufData(2, (char *) &metap->hashm_highmask, sizeof(uint32));
927  }
928 
929  if (metap_update_splitpoint)
930  {
932  XLogRegisterBufData(2, (char *) &metap->hashm_ovflpoint,
933  sizeof(uint32));
935  (char *) &metap->hashm_spares[metap->hashm_ovflpoint],
936  sizeof(uint32));
937  }
938 
939  XLogRegisterData((char *) &xlrec, SizeOfHashSplitAllocPage);
940 
941  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_ALLOCATE_PAGE);
942 
943  PageSetLSN(BufferGetPage(buf_oblkno), recptr);
944  PageSetLSN(BufferGetPage(buf_nblkno), recptr);
945  PageSetLSN(BufferGetPage(metabuf), recptr);
946  }
947 
949 
950  /* drop lock, but keep pin */
951  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
952 
953  /* Relocate records to the new bucket */
954  _hash_splitbucket(rel, metabuf,
955  old_bucket, new_bucket,
956  buf_oblkno, buf_nblkno, NULL,
957  maxbucket, highmask, lowmask);
958 
959  /* all done, now release the pins on primary buckets. */
960  _hash_dropbuf(rel, buf_oblkno);
961  _hash_dropbuf(rel, buf_nblkno);
962 
963  return;
964 
965  /* Here if decide not to split or fail to acquire old bucket lock */
966 fail:
967 
968  /* We didn't write the metapage, so just drop lock */
969  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
970 }
void XLogRegisterBufData(uint8 block_id, char *data, int len)
Definition: xloginsert.c:361
uint16 hasho_page_id
Definition: hash.h:93
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:87
#define XLH_SPLIT_META_UPDATE_SPLITPOINT
Definition: hash_xlog.h:52
#define LH_META_PAGE
Definition: hash.h:67
void MarkBufferDirty(Buffer buffer)
Definition: bufmgr.c:1450
void XLogRegisterBuffer(uint8 block_id, Buffer buffer, uint8 flags)
Definition: xloginsert.c:213
#define END_CRIT_SECTION()
Definition: miscadmin.h:133
uint16 hashm_ffactor
Definition: hash.h:256
uint32 hashm_highmask
Definition: hash.h:262
#define REGBUF_WILL_INIT
Definition: xloginsert.h:32
#define START_CRIT_SECTION()
Definition: miscadmin.h:131
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:672
uint32 BlockNumber
Definition: block.h:31
Buffer _hash_getnewbuf(Relation rel, BlockNumber blkno, ForkNumber forkNum)
Definition: hashpage.c:206
void _hash_dropbuf(Relation rel, Buffer buf)
Definition: hashpage.c:285
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:89
uint32 hashm_lowmask
Definition: hash.h:263
#define BUCKET_TO_BLKNO(metap, B)
Definition: hash.h:38
static bool _hash_alloc_buckets(Relation rel, BlockNumber firstblock, uint32 nblocks)
Definition: hashpage.c:998
uint32 Bucket
Definition: hash.h:34
#define H_NEEDS_SPLIT_CLEANUP(opaque)
Definition: hash.h:98
BlockNumber hasho_prevblkno
Definition: hash.h:89
#define SizeOfHashSplitAllocPage
Definition: hash_xlog.h:119
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:1356
#define REGBUF_STANDARD
Definition: xloginsert.h:34
#define LH_BUCKET_BEING_SPLIT
Definition: hash.h:69
unsigned int uint32
Definition: c.h:258
Buffer _hash_getbuf_with_condlock_cleanup(Relation rel, BlockNumber blkno, int flags)
Definition: hashpage.c:104
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
bool IsBufferCleanupOK(Buffer buffer)
Definition: bufmgr.c:3774
uint32 hashm_ovflpoint
Definition: hash.h:264
void XLogRegisterData(char *data, int len)
Definition: xloginsert.c:323
#define LH_BUCKET_BEING_POPULATED
Definition: hash.h:68
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:255
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
uint32 hashm_spares[HASH_MAX_SPLITPOINTS]
Definition: hash.h:269
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:274
#define LH_BUCKET_PAGE
Definition: hash.h:65
#define H_BUCKET_BEING_SPLIT(opaque)
Definition: hash.h:99
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:1078
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:664
Bucket hasho_bucket
Definition: hash.h:91
#define PageGetSpecialPointer(page)
Definition: bufpage.h:322
#define InvalidBlockNumber
Definition: block.h:33
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:96
#define HASHO_PAGE_ID
Definition: hash.h:109
#define RelationNeedsWAL(relation)
Definition: rel.h:505
uint32 hashm_maxbucket
Definition: hash.h:261
uint16 hasho_flag
Definition: hash.h:92
#define XLOG_HASH_SPLIT_ALLOCATE_PAGE
Definition: hash_xlog.h:31
#define HashPageGetMeta(page)
Definition: hash.h:315
BlockNumber hasho_nextblkno
Definition: hash.h:90
#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:364
int Buffer
Definition: buf.h:23
Pointer Page
Definition: bufpage.h:74
void _hash_finish_split ( Relation  rel,
Buffer  metabuf,
Buffer  obuf,
Bucket  obucket,
uint32  maxbucket,
uint32  highmask,
uint32  lowmask 
)

Definition at line 1356 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().

1358 {
1359  HASHCTL hash_ctl;
1360  HTAB *tidhtab;
1361  Buffer bucket_nbuf = InvalidBuffer;
1362  Buffer nbuf;
1363  Page npage;
1364  BlockNumber nblkno;
1365  BlockNumber bucket_nblkno;
1366  HashPageOpaque npageopaque;
1367  Bucket nbucket;
1368  bool found;
1369 
1370  /* Initialize hash tables used to track TIDs */
1371  memset(&hash_ctl, 0, sizeof(hash_ctl));
1372  hash_ctl.keysize = sizeof(ItemPointerData);
1373  hash_ctl.entrysize = sizeof(ItemPointerData);
1374  hash_ctl.hcxt = CurrentMemoryContext;
1375 
1376  tidhtab =
1377  hash_create("bucket ctids",
1378  256, /* arbitrary initial size */
1379  &hash_ctl,
1381 
1382  bucket_nblkno = nblkno = _hash_get_newblock_from_oldbucket(rel, obucket);
1383 
1384  /*
1385  * Scan the new bucket and build hash table of TIDs
1386  */
1387  for (;;)
1388  {
1389  OffsetNumber noffnum;
1390  OffsetNumber nmaxoffnum;
1391 
1392  nbuf = _hash_getbuf(rel, nblkno, HASH_READ,
1394 
1395  /* remember the primary bucket buffer to acquire cleanup lock on it. */
1396  if (nblkno == bucket_nblkno)
1397  bucket_nbuf = nbuf;
1398 
1399  npage = BufferGetPage(nbuf);
1400  npageopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
1401 
1402  /* Scan each tuple in new page */
1403  nmaxoffnum = PageGetMaxOffsetNumber(npage);
1404  for (noffnum = FirstOffsetNumber;
1405  noffnum <= nmaxoffnum;
1406  noffnum = OffsetNumberNext(noffnum))
1407  {
1408  IndexTuple itup;
1409 
1410  /* Fetch the item's TID and insert it in hash table. */
1411  itup = (IndexTuple) PageGetItem(npage,
1412  PageGetItemId(npage, noffnum));
1413 
1414  (void) hash_search(tidhtab, &itup->t_tid, HASH_ENTER, &found);
1415 
1416  Assert(!found);
1417  }
1418 
1419  nblkno = npageopaque->hasho_nextblkno;
1420 
1421  /*
1422  * release our write lock without modifying buffer and ensure to
1423  * retain the pin on primary bucket.
1424  */
1425  if (nbuf == bucket_nbuf)
1427  else
1428  _hash_relbuf(rel, nbuf);
1429 
1430  /* Exit loop if no more overflow pages in new bucket */
1431  if (!BlockNumberIsValid(nblkno))
1432  break;
1433  }
1434 
1435  /*
1436  * Conditionally get the cleanup lock on old and new buckets to perform
1437  * the split operation. If we don't get the cleanup locks, silently give
1438  * up and next insertion on old bucket will try again to complete the
1439  * split.
1440  */
1442  {
1443  hash_destroy(tidhtab);
1444  return;
1445  }
1446  if (!ConditionalLockBufferForCleanup(bucket_nbuf))
1447  {
1449  hash_destroy(tidhtab);
1450  return;
1451  }
1452 
1453  npage = BufferGetPage(bucket_nbuf);
1454  npageopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
1455  nbucket = npageopaque->hasho_bucket;
1456 
1457  _hash_splitbucket(rel, metabuf, obucket,
1458  nbucket, obuf, bucket_nbuf, tidhtab,
1459  maxbucket, highmask, lowmask);
1460 
1461  _hash_dropbuf(rel, bucket_nbuf);
1462  hash_destroy(tidhtab);
1463 }
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:87
void hash_destroy(HTAB *hashp)
Definition: dynahash.c:810
#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:285
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:902
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:78
#define PageGetMaxOffsetNumber(page)
Definition: bufpage.h:353
uint16 OffsetNumber
Definition: off.h:24
#define HASH_READ
Definition: hash.h:331
uint32 Bucket
Definition: hash.h:34
Definition: dynahash.c:208
bool ConditionalLockBufferForCleanup(Buffer buffer)
Definition: bufmgr.c:3718
#define FirstOffsetNumber
Definition: off.h:27
IndexTupleData * IndexTuple
Definition: itup.h:53
MemoryContext CurrentMemoryContext
Definition: mcxt.c:37
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:231
#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:64
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
Size keysize
Definition: hsearch.h:72
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:274
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:65
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:1078
#define Assert(condition)
Definition: c.h:664
Bucket hasho_bucket
Definition: hash.h:91
struct ItemPointerData ItemPointerData
#define OffsetNumberNext(offsetNumber)
Definition: off.h:53
#define PageGetSpecialPointer(page)
Definition: bufpage.h:322
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:96
BlockNumber hasho_nextblkno
Definition: hash.h:90
int Buffer
Definition: buf.h:23
#define PageGetItem(page, itemId)
Definition: bufpage.h:336
Pointer Page
Definition: bufpage.h:74
Buffer _hash_getbucketbuf_from_hashkey ( Relation  rel,
uint32  hashkey,
int  access,
HashMetaPage cachedmetap 
)

Definition at line 1560 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().

1562 {
1563  HashMetaPage metap;
1564  Buffer buf;
1565  Buffer metabuf = InvalidBuffer;
1566  Page page;
1567  Bucket bucket;
1568  BlockNumber blkno;
1569  HashPageOpaque opaque;
1570 
1571  /* We read from target bucket buffer, hence locking is must. */
1572  Assert(access == HASH_READ || access == HASH_WRITE);
1573 
1574  metap = _hash_getcachedmetap(rel, &metabuf, false);
1575  Assert(metap != NULL);
1576 
1577  /*
1578  * Loop until we get a lock on the correct target bucket.
1579  */
1580  for (;;)
1581  {
1582  /*
1583  * Compute the target bucket number, and convert to block number.
1584  */
1585  bucket = _hash_hashkey2bucket(hashkey,
1586  metap->hashm_maxbucket,
1587  metap->hashm_highmask,
1588  metap->hashm_lowmask);
1589 
1590  blkno = BUCKET_TO_BLKNO(metap, bucket);
1591 
1592  /* Fetch the primary bucket page for the bucket */
1593  buf = _hash_getbuf(rel, blkno, access, LH_BUCKET_PAGE);
1594  page = BufferGetPage(buf);
1595  opaque = (HashPageOpaque) PageGetSpecialPointer(page);
1596  Assert(opaque->hasho_bucket == bucket);
1598 
1599  /*
1600  * If this bucket hasn't been split, we're done.
1601  */
1602  if (opaque->hasho_prevblkno <= metap->hashm_maxbucket)
1603  break;
1604 
1605  /* Drop lock on this buffer, update cached metapage, and retry. */
1606  _hash_relbuf(rel, buf);
1607  metap = _hash_getcachedmetap(rel, &metabuf, true);
1608  Assert(metap != NULL);
1609  }
1610 
1611  if (BufferIsValid(metabuf))
1612  _hash_dropbuf(rel, metabuf);
1613 
1614  if (cachedmetap)
1615  *cachedmetap = metap;
1616 
1617  return buf;
1618 }
Bucket _hash_hashkey2bucket(uint32 hashkey, uint32 maxbucket, uint32 highmask, uint32 lowmask)
Definition: hashutil.c:125
uint32 hashm_highmask
Definition: hash.h:262
#define InvalidBuffer
Definition: buf.h:25
uint32 BlockNumber
Definition: block.h:31
void _hash_dropbuf(Relation rel, Buffer buf)
Definition: hashpage.c:285
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:78
uint32 hashm_lowmask
Definition: hash.h:263
#define BUCKET_TO_BLKNO(metap, B)
Definition: hash.h:38
#define HASH_READ
Definition: hash.h:331
uint32 Bucket
Definition: hash.h:34
BlockNumber hasho_prevblkno
Definition: hash.h:89
static char * buf
Definition: pg_test_fsync.c:67
#define HASH_WRITE
Definition: hash.h:332
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
HashMetaPage _hash_getcachedmetap(Relation rel, Buffer *metabuf, bool force_refresh)
Definition: hashpage.c:1502
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:274
#define LH_BUCKET_PAGE
Definition: hash.h:65
#define Assert(condition)
Definition: c.h:664
Bucket hasho_bucket
Definition: hash.h:91
#define PageGetSpecialPointer(page)
Definition: bufpage.h:322
#define InvalidBlockNumber
Definition: block.h:33
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:96
#define BufferIsValid(bufnum)
Definition: bufmgr.h:114
uint32 hashm_maxbucket
Definition: hash.h:261
int Buffer
Definition: buf.h:23
Pointer Page
Definition: bufpage.h:74
Buffer _hash_getbuf ( Relation  rel,
BlockNumber  blkno,
int  access,
int  flags 
)

Definition at line 78 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().

79 {
80  Buffer buf;
81 
82  if (blkno == P_NEW)
83  elog(ERROR, "hash AM does not use P_NEW");
84 
85  buf = ReadBuffer(rel, blkno);
86 
87  if (access != HASH_NOLOCK)
88  LockBuffer(buf, access);
89 
90  /* ref count and lock type are correct */
91 
92  _hash_checkpage(rel, buf, flags);
93 
94  return buf;
95 }
#define P_NEW
Definition: bufmgr.h:82
#define ERROR
Definition: elog.h:43
static char * buf
Definition: pg_test_fsync.c:67
#define HASH_NOLOCK
Definition: hash.h:333
void _hash_checkpage(Relation rel, Buffer buf, int flags)
Definition: hashutil.c:225
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
Buffer ReadBuffer(Relation reln, BlockNumber blockNum)
Definition: bufmgr.c:594
#define elog
Definition: elog.h:219
int Buffer
Definition: buf.h:23
Buffer _hash_getbuf_with_condlock_cleanup ( Relation  rel,
BlockNumber  blkno,
int  flags 
)

Definition at line 104 of file hashpage.c.

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

Referenced by _hash_expandtable().

105 {
106  Buffer buf;
107 
108  if (blkno == P_NEW)
109  elog(ERROR, "hash AM does not use P_NEW");
110 
111  buf = ReadBuffer(rel, blkno);
112 
114  {
115  ReleaseBuffer(buf);
116  return InvalidBuffer;
117  }
118 
119  /* ref count and lock type are correct */
120 
121  _hash_checkpage(rel, buf, flags);
122 
123  return buf;
124 }
#define InvalidBuffer
Definition: buf.h:25
void ReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:3309
#define P_NEW
Definition: bufmgr.h:82
bool ConditionalLockBufferForCleanup(Buffer buffer)
Definition: bufmgr.c:3718
#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:594
#define elog
Definition: elog.h:219
int Buffer
Definition: buf.h:23
Buffer _hash_getbuf_with_strategy ( Relation  rel,
BlockNumber  blkno,
int  access,
int  flags,
BufferAccessStrategy  bstrategy 
)

Definition at line 247 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().

250 {
251  Buffer buf;
252 
253  if (blkno == P_NEW)
254  elog(ERROR, "hash AM does not use P_NEW");
255 
256  buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_NORMAL, bstrategy);
257 
258  if (access != HASH_NOLOCK)
259  LockBuffer(buf, access);
260 
261  /* ref count and lock type are correct */
262 
263  _hash_checkpage(rel, buf, flags);
264 
265  return buf;
266 }
Buffer ReadBufferExtended(Relation reln, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy)
Definition: bufmgr.c:640
#define P_NEW
Definition: bufmgr.h:82
#define ERROR
Definition: elog.h:43
static char * buf
Definition: pg_test_fsync.c:67
#define HASH_NOLOCK
Definition: hash.h:333
void _hash_checkpage(Relation rel, Buffer buf, int flags)
Definition: hashutil.c:225
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
#define elog
Definition: elog.h:219
int Buffer
Definition: buf.h:23
HashMetaPage _hash_getcachedmetap ( Relation  rel,
Buffer metabuf,
bool  force_refresh 
)

Definition at line 1502 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().

1503 {
1504  Page page;
1505 
1506  Assert(metabuf);
1507  if (force_refresh || rel->rd_amcache == NULL)
1508  {
1509  char *cache = NULL;
1510 
1511  /*
1512  * It's important that we don't set rd_amcache to an invalid value.
1513  * Either MemoryContextAlloc or _hash_getbuf could fail, so don't
1514  * install a pointer to the newly-allocated storage in the actual
1515  * relcache entry until both have succeeeded.
1516  */
1517  if (rel->rd_amcache == NULL)
1518  cache = MemoryContextAlloc(rel->rd_indexcxt,
1519  sizeof(HashMetaPageData));
1520 
1521  /* Read the metapage. */
1522  if (BufferIsValid(*metabuf))
1523  LockBuffer(*metabuf, BUFFER_LOCK_SHARE);
1524  else
1525  *metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ,
1526  LH_META_PAGE);
1527  page = BufferGetPage(*metabuf);
1528 
1529  /* Populate the cache. */
1530  if (rel->rd_amcache == NULL)
1531  rel->rd_amcache = cache;
1532  memcpy(rel->rd_amcache, HashPageGetMeta(page),
1533  sizeof(HashMetaPageData));
1534 
1535  /* Release metapage lock, but keep the pin. */
1536  LockBuffer(*metabuf, BUFFER_LOCK_UNLOCK);
1537  }
1538 
1539  return (HashMetaPage) rel->rd_amcache;
1540 }
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:87
#define LH_META_PAGE
Definition: hash.h:67
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:78
#define HASH_READ
Definition: hash.h:331
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define HASH_METAPAGE
Definition: hash.h:208
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
#define Assert(condition)
Definition: c.h:664
#define BufferIsValid(bufnum)
Definition: bufmgr.h:114
#define HashPageGetMeta(page)
Definition: hash.h:315
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:706
MemoryContext rd_indexcxt
Definition: rel.h:179
#define BUFFER_LOCK_SHARE
Definition: bufmgr.h:88
void * rd_amcache
Definition: rel.h:192
Pointer Page
Definition: bufpage.h:74
Buffer _hash_getinitbuf ( Relation  rel,
BlockNumber  blkno 
)

Definition at line 143 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().

144 {
145  Buffer buf;
146 
147  if (blkno == P_NEW)
148  elog(ERROR, "hash AM does not use P_NEW");
149 
151  NULL);
152 
153  /* ref count and lock type are correct */
154 
155  /* initialize the page */
157 
158  return buf;
159 }
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:606
Buffer ReadBufferExtended(Relation reln, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy)
Definition: bufmgr.c:640
#define P_NEW
Definition: bufmgr.h:82
#define ERROR
Definition: elog.h:43
static char * buf
Definition: pg_test_fsync.c:67
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define BufferGetPageSize(buffer)
Definition: bufmgr.h:147
#define elog
Definition: elog.h:219
int Buffer
Definition: buf.h:23
Buffer _hash_getnewbuf ( Relation  rel,
BlockNumber  blkno,
ForkNumber  forkNum 
)

Definition at line 206 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().

207 {
208  BlockNumber nblocks = RelationGetNumberOfBlocksInFork(rel, forkNum);
209  Buffer buf;
210 
211  if (blkno == P_NEW)
212  elog(ERROR, "hash AM does not use P_NEW");
213  if (blkno > nblocks)
214  elog(ERROR, "access to noncontiguous page in hash index \"%s\"",
216 
217  /* smgr insists we use P_NEW to extend the relation */
218  if (blkno == nblocks)
219  {
220  buf = ReadBufferExtended(rel, forkNum, P_NEW, RBM_NORMAL, NULL);
221  if (BufferGetBlockNumber(buf) != blkno)
222  elog(ERROR, "unexpected hash relation size: %u, should be %u",
223  BufferGetBlockNumber(buf), blkno);
224  LockBuffer(buf, HASH_WRITE);
225  }
226  else
227  {
228  buf = ReadBufferExtended(rel, forkNum, blkno, RBM_ZERO_AND_LOCK,
229  NULL);
230  }
231 
232  /* ref count and lock type are correct */
233 
234  /* initialize the page */
236 
237  return buf;
238 }
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:606
Buffer ReadBufferExtended(Relation reln, ForkNumber forkNum, BlockNumber blockNum, ReadBufferMode mode, BufferAccessStrategy strategy)
Definition: bufmgr.c:640
uint32 BlockNumber
Definition: block.h:31
#define P_NEW
Definition: bufmgr.h:82
#define ERROR
Definition: elog.h:43
static char * buf
Definition: pg_test_fsync.c:67
#define HASH_WRITE
Definition: hash.h:332
#define RelationGetRelationName(relation)
Definition: rel.h:436
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define BufferGetPageSize(buffer)
Definition: bufmgr.h:147
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
BlockNumber RelationGetNumberOfBlocksInFork(Relation relation, ForkNumber forkNum)
Definition: bufmgr.c:2788
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:2605
#define elog
Definition: elog.h:219
int Buffer
Definition: buf.h:23
uint32 _hash_init ( Relation  rel,
double  num_tuples,
ForkNumber  forkNum 
)

Definition at line 335 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().

336 {
337  Buffer metabuf;
338  Buffer buf;
339  Buffer bitmapbuf;
340  Page pg;
341  HashMetaPage metap;
342  RegProcedure procid;
343  int32 data_width;
344  int32 item_width;
345  int32 ffactor;
346  uint32 num_buckets;
347  uint32 i;
348  bool use_wal;
349 
350  /* safety check */
351  if (RelationGetNumberOfBlocksInFork(rel, forkNum) != 0)
352  elog(ERROR, "cannot initialize non-empty hash index \"%s\"",
354 
355  /*
356  * WAL log creation of pages if the relation is persistent, or this is the
357  * init fork. Init forks for unlogged relations always need to be WAL
358  * logged.
359  */
360  use_wal = RelationNeedsWAL(rel) || forkNum == INIT_FORKNUM;
361 
362  /*
363  * Determine the target fill factor (in tuples per bucket) for this index.
364  * The idea is to make the fill factor correspond to pages about as full
365  * as the user-settable fillfactor parameter says. We can compute it
366  * exactly since the index datatype (i.e. uint32 hash key) is fixed-width.
367  */
368  data_width = sizeof(uint32);
369  item_width = MAXALIGN(sizeof(IndexTupleData)) + MAXALIGN(data_width) +
370  sizeof(ItemIdData); /* include the line pointer */
371  ffactor = RelationGetTargetPageUsage(rel, HASH_DEFAULT_FILLFACTOR) / item_width;
372  /* keep to a sane range */
373  if (ffactor < 10)
374  ffactor = 10;
375 
376  procid = index_getprocid(rel, 1, HASHSTANDARD_PROC);
377 
378  /*
379  * We initialize the metapage, the first N bucket pages, and the first
380  * bitmap page in sequence, using _hash_getnewbuf to cause smgrextend()
381  * calls to occur. This ensures that the smgr level has the right idea of
382  * the physical index length.
383  *
384  * Critical section not required, because on error the creation of the
385  * whole relation will be rolled back.
386  */
387  metabuf = _hash_getnewbuf(rel, HASH_METAPAGE, forkNum);
388  _hash_init_metabuffer(metabuf, num_tuples, procid, ffactor, false);
389  MarkBufferDirty(metabuf);
390 
391  pg = BufferGetPage(metabuf);
392  metap = HashPageGetMeta(pg);
393 
394  /* XLOG stuff */
395  if (use_wal)
396  {
398  XLogRecPtr recptr;
399 
400  xlrec.num_tuples = num_tuples;
401  xlrec.procid = metap->hashm_procid;
402  xlrec.ffactor = metap->hashm_ffactor;
403 
404  XLogBeginInsert();
405  XLogRegisterData((char *) &xlrec, SizeOfHashInitMetaPage);
407 
408  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_INIT_META_PAGE);
409 
410  PageSetLSN(BufferGetPage(metabuf), recptr);
411  }
412 
413  num_buckets = metap->hashm_maxbucket + 1;
414 
415  /*
416  * Release buffer lock on the metapage while we initialize buckets.
417  * Otherwise, we'll be in interrupt holdoff and the CHECK_FOR_INTERRUPTS
418  * won't accomplish anything. It's a bad idea to hold buffer locks for
419  * long intervals in any case, since that can block the bgwriter.
420  */
421  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
422 
423  /*
424  * Initialize and WAL Log the first N buckets
425  */
426  for (i = 0; i < num_buckets; i++)
427  {
428  BlockNumber blkno;
429 
430  /* Allow interrupts, in case N is huge */
432 
433  blkno = BUCKET_TO_BLKNO(metap, i);
434  buf = _hash_getnewbuf(rel, blkno, forkNum);
435  _hash_initbuf(buf, metap->hashm_maxbucket, i, LH_BUCKET_PAGE, false);
436  MarkBufferDirty(buf);
437 
438  if (use_wal)
439  log_newpage(&rel->rd_node,
440  forkNum,
441  blkno,
442  BufferGetPage(buf),
443  true);
444  _hash_relbuf(rel, buf);
445  }
446 
447  /* Now reacquire buffer lock on metapage */
449 
450  /*
451  * Initialize bitmap page
452  */
453  bitmapbuf = _hash_getnewbuf(rel, num_buckets + 1, forkNum);
454  _hash_initbitmapbuffer(bitmapbuf, metap->hashm_bmsize, false);
455  MarkBufferDirty(bitmapbuf);
456 
457  /* add the new bitmap page to the metapage's list of bitmaps */
458  /* metapage already has a write lock */
459  if (metap->hashm_nmaps >= HASH_MAX_BITMAPS)
460  ereport(ERROR,
461  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
462  errmsg("out of overflow pages in hash index \"%s\"",
463  RelationGetRelationName(rel))));
464 
465  metap->hashm_mapp[metap->hashm_nmaps] = num_buckets + 1;
466 
467  metap->hashm_nmaps++;
468  MarkBufferDirty(metabuf);
469 
470  /* XLOG stuff */
471  if (use_wal)
472  {
474  XLogRecPtr recptr;
475 
476  xlrec.bmsize = metap->hashm_bmsize;
477 
478  XLogBeginInsert();
479  XLogRegisterData((char *) &xlrec, SizeOfHashInitBitmapPage);
480  XLogRegisterBuffer(0, bitmapbuf, REGBUF_WILL_INIT);
481 
482  /*
483  * This is safe only because nobody else can be modifying the index at
484  * this stage; it's only visible to the transaction that is creating
485  * it.
486  */
487  XLogRegisterBuffer(1, metabuf, REGBUF_STANDARD);
488 
489  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_INIT_BITMAP_PAGE);
490 
491  PageSetLSN(BufferGetPage(bitmapbuf), recptr);
492  PageSetLSN(BufferGetPage(metabuf), recptr);
493  }
494 
495  /* all done */
496  _hash_relbuf(rel, bitmapbuf);
497  _hash_relbuf(rel, metabuf);
498 
499  return num_buckets;
500 }
#define HASH_DEFAULT_FILLFACTOR
Definition: hash.h:288
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:87
RegProcedure hashm_procid
Definition: hash.h:268
void MarkBufferDirty(Buffer buffer)
Definition: bufmgr.c:1450
void XLogRegisterBuffer(uint8 block_id, Buffer buffer, uint8 flags)
Definition: xloginsert.c:213
regproc RegProcedure
Definition: c.h:389
#define XLOG_HASH_INIT_BITMAP_PAGE
Definition: hash_xlog.h:28
uint16 hashm_ffactor
Definition: hash.h:256
#define REGBUF_WILL_INIT
Definition: xloginsert.h:32
int errcode(int sqlerrcode)
Definition: elog.c:575
uint32 BlockNumber
Definition: block.h:31
Buffer _hash_getnewbuf(Relation rel, BlockNumber blkno, ForkNumber forkNum)
Definition: hashpage.c:206
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:89
signed int int32
Definition: c.h:246
#define BUCKET_TO_BLKNO(metap, B)
Definition: hash.h:38
#define ERROR
Definition: elog.h:43
#define SizeOfHashInitMetaPage
Definition: hash_xlog.h:236
void _hash_init_metabuffer(Buffer buf, double num_tuples, RegProcedure procid, uint16 ffactor, bool initpage)
Definition: hashpage.c:506
#define SizeOfHashInitBitmapPage
Definition: hash_xlog.h:252
uint32 hashm_nmaps
Definition: hash.h:267
static char * buf
Definition: pg_test_fsync.c:67
#define REGBUF_STANDARD
Definition: xloginsert.h:34
#define RelationGetRelationName(relation)
Definition: rel.h:436
unsigned int uint32
Definition: c.h:258
struct ItemIdData ItemIdData
#define XLOG_HASH_INIT_META_PAGE
Definition: hash_xlog.h:27
#define HASHSTANDARD_PROC
Definition: hash.h:353
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define ereport(elevel, rest)
Definition: elog.h:122
#define HASH_MAX_BITMAPS
Definition: hash.h:237
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:208
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:274
void _hash_initbitmapbuffer(Buffer buf, uint16 bmsize, bool initpage)
Definition: hashovfl.c:739
#define LH_BUCKET_PAGE
Definition: hash.h:65
BlockNumber RelationGetNumberOfBlocksInFork(Relation relation, ForkNumber forkNum)
Definition: bufmgr.c:2788
RelFileNode rd_node
Definition: rel.h:85
#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:165
#define MAXALIGN(LEN)
Definition: c.h:576
#define RelationNeedsWAL(relation)
Definition: rel.h:505
uint32 hashm_maxbucket
Definition: hash.h:261
RegProcedure procid
Definition: hash_xlog.h:232
#define HashPageGetMeta(page)
Definition: hash.h:315
int errmsg(const char *fmt,...)
Definition: elog.c:797
XLogRecPtr log_newpage(RelFileNode *rnode, ForkNumber forkNum, BlockNumber blkno, Page page, bool page_std)
Definition: xloginsert.c:972
int i
uint16 hashm_bmsize
Definition: hash.h:258
#define CHECK_FOR_INTERRUPTS()
Definition: miscadmin.h:98
#define elog
Definition: elog.h:219
void XLogBeginInsert(void)
Definition: xloginsert.c:120
#define PageSetLSN(page, lsn)
Definition: bufpage.h:364
BlockNumber hashm_mapp[HASH_MAX_BITMAPS]
Definition: hash.h:271
int Buffer
Definition: buf.h:23
Pointer Page
Definition: bufpage.h:74
RegProcedure index_getprocid(Relation irel, AttrNumber attnum, uint16 procnum)
Definition: indexam.c:821
void _hash_init_metabuffer ( Buffer  buf,
double  num_tuples,
RegProcedure  procid,
uint16  ffactor,
bool  initpage 
)

Definition at line 506 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().

508 {
509  HashMetaPage metap;
510  HashPageOpaque pageopaque;
511  Page page;
512  double dnumbuckets;
513  uint32 num_buckets;
514  uint32 spare_index;
515  uint32 i;
516 
517  /*
518  * Choose the number of initial bucket pages to match the fill factor
519  * given the estimated number of tuples. We round up the result to the
520  * total number of buckets which has to be allocated before using its
521  * _hashm_spare element. However always force at least 2 bucket pages. The
522  * upper limit is determined by considerations explained in
523  * _hash_expandtable().
524  */
525  dnumbuckets = num_tuples / ffactor;
526  if (dnumbuckets <= 2.0)
527  num_buckets = 2;
528  else if (dnumbuckets >= (double) 0x40000000)
529  num_buckets = 0x40000000;
530  else
531  num_buckets = _hash_get_totalbuckets(_hash_spareindex(dnumbuckets));
532 
533  spare_index = _hash_spareindex(num_buckets);
534  Assert(spare_index < HASH_MAX_SPLITPOINTS);
535 
536  page = BufferGetPage(buf);
537  if (initpage)
539 
540  pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
541  pageopaque->hasho_prevblkno = InvalidBlockNumber;
542  pageopaque->hasho_nextblkno = InvalidBlockNumber;
543  pageopaque->hasho_bucket = -1;
544  pageopaque->hasho_flag = LH_META_PAGE;
545  pageopaque->hasho_page_id = HASHO_PAGE_ID;
546 
547  metap = HashPageGetMeta(page);
548 
549  metap->hashm_magic = HASH_MAGIC;
550  metap->hashm_version = HASH_VERSION;
551  metap->hashm_ntuples = 0;
552  metap->hashm_nmaps = 0;
553  metap->hashm_ffactor = ffactor;
554  metap->hashm_bsize = HashGetMaxBitmapSize(page);
555  /* find largest bitmap array size that will fit in page size */
556  for (i = _hash_log2(metap->hashm_bsize); i > 0; --i)
557  {
558  if ((1 << i) <= metap->hashm_bsize)
559  break;
560  }
561  Assert(i > 0);
562  metap->hashm_bmsize = 1 << i;
563  metap->hashm_bmshift = i + BYTE_TO_BIT;
564  Assert((1 << BMPG_SHIFT(metap)) == (BMPG_MASK(metap) + 1));
565 
566  /*
567  * Label the index with its primary hash support function's OID. This is
568  * pretty useless for normal operation (in fact, hashm_procid is not used
569  * anywhere), but it might be handy for forensic purposes so we keep it.
570  */
571  metap->hashm_procid = procid;
572 
573  /*
574  * We initialize the index with N buckets, 0 .. N-1, occupying physical
575  * blocks 1 to N. The first freespace bitmap page is in block N+1.
576  */
577  metap->hashm_maxbucket = num_buckets - 1;
578 
579  /*
580  * Set highmask as next immediate ((2 ^ x) - 1), which should be
581  * sufficient to cover num_buckets.
582  */
583  metap->hashm_highmask = (1 << (_hash_log2(num_buckets + 1))) - 1;
584  metap->hashm_lowmask = (metap->hashm_highmask >> 1);
585 
586  MemSet(metap->hashm_spares, 0, sizeof(metap->hashm_spares));
587  MemSet(metap->hashm_mapp, 0, sizeof(metap->hashm_mapp));
588 
589  /* Set up mapping for one spare page after the initial splitpoints */
590  metap->hashm_spares[spare_index] = 1;
591  metap->hashm_ovflpoint = spare_index;
592  metap->hashm_firstfree = 0;
593 
594  /*
595  * Set pd_lower just past the end of the metadata. This is to log full
596  * page image of metapage in xloginsert.c.
597  */
598  ((PageHeader) page)->pd_lower =
599  ((char *) metap + sizeof(HashMetaPageData)) - (char *) page;
600 }
uint16 hashm_bmshift
Definition: hash.h:260
uint16 hasho_page_id
Definition: hash.h:93
RegProcedure hashm_procid
Definition: hash.h:268
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:606
#define LH_META_PAGE
Definition: hash.h:67
#define BYTE_TO_BIT
Definition: hash.h:293
uint32 hashm_magic
Definition: hash.h:253
uint16 hashm_ffactor
Definition: hash.h:256
uint32 hashm_highmask
Definition: hash.h:262
#define MemSet(start, val, len)
Definition: c.h:846
#define HASH_VERSION
Definition: hash.h:211
uint32 hashm_lowmask
Definition: hash.h:263
#define HASH_MAX_SPLITPOINTS
Definition: hash.h:246
#define HASH_MAGIC
Definition: hash.h:210
#define HashGetMaxBitmapSize(page)
Definition: hash.h:311
BlockNumber hasho_prevblkno
Definition: hash.h:89
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:254
uint32 hashm_nmaps
Definition: hash.h:267
static char * buf
Definition: pg_test_fsync.c:67
#define BMPG_MASK(metap)
Definition: hash.h:306
unsigned int uint32
Definition: c.h:258
#define BMPG_SHIFT(metap)
Definition: hash.h:305
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
uint32 hashm_ovflpoint
Definition: hash.h:264
uint16 hashm_bsize
Definition: hash.h:257
double hashm_ntuples
Definition: hash.h:255
#define BufferGetPageSize(buffer)
Definition: bufmgr.h:147
uint32 hashm_firstfree
Definition: hash.h:266
uint32 hashm_spares[HASH_MAX_SPLITPOINTS]
Definition: hash.h:269
PageHeaderData * PageHeader
Definition: bufpage.h:162
#define Assert(condition)
Definition: c.h:664
Bucket hasho_bucket
Definition: hash.h:91
#define PageGetSpecialPointer(page)
Definition: bufpage.h:322
#define InvalidBlockNumber
Definition: block.h:33
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:96
#define HASHO_PAGE_ID
Definition: hash.h:109
uint32 hashm_maxbucket
Definition: hash.h:261
uint16 hasho_flag
Definition: hash.h:92
#define HashPageGetMeta(page)
Definition: hash.h:315
uint32 _hash_log2(uint32 num)
Definition: hashutil.c:141
int i
BlockNumber hasho_nextblkno
Definition: hash.h:90
uint16 hashm_bmsize
Definition: hash.h:258
BlockNumber hashm_mapp[HASH_MAX_BITMAPS]
Definition: hash.h:271
Pointer Page
Definition: bufpage.h:74
void _hash_initbuf ( Buffer  buf,
uint32  max_bucket,
uint32  num_bucket,
uint32  flag,
bool  initpage 
)

Definition at line 165 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().

167 {
168  HashPageOpaque pageopaque;
169  Page page;
170 
171  page = BufferGetPage(buf);
172 
173  /* initialize the page */
174  if (initpage)
176 
177  pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
178 
179  /*
180  * Set hasho_prevblkno with current hashm_maxbucket. This value will be
181  * used to validate cached HashMetaPageData. See
182  * _hash_getbucketbuf_from_hashkey().
183  */
184  pageopaque->hasho_prevblkno = max_bucket;
185  pageopaque->hasho_nextblkno = InvalidBlockNumber;
186  pageopaque->hasho_bucket = num_bucket;
187  pageopaque->hasho_flag = flag;
188  pageopaque->hasho_page_id = HASHO_PAGE_ID;
189 }
uint16 hasho_page_id
Definition: hash.h:93
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:606
BlockNumber hasho_prevblkno
Definition: hash.h:89
static char * buf
Definition: pg_test_fsync.c:67
char * flag(int b)
Definition: test-ctype.c:33
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define BufferGetPageSize(buffer)
Definition: bufmgr.h:147
Bucket hasho_bucket
Definition: hash.h:91
#define PageGetSpecialPointer(page)
Definition: bufpage.h:322
#define InvalidBlockNumber
Definition: block.h:33
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:96
#define HASHO_PAGE_ID
Definition: hash.h:109
uint16 hasho_flag
Definition: hash.h:92
BlockNumber hasho_nextblkno
Definition: hash.h:90
Pointer Page
Definition: bufpage.h:74
void _hash_pageinit ( Page  page,
Size  size 
)

Definition at line 606 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().

607 {
608  PageInit(page, size, sizeof(HashPageOpaqueData));
609 }
void PageInit(Page page, Size pageSize, Size specialSize)
Definition: bufpage.c:41
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 1078 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, IndexTupleDSize, 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(), 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().

1088 {
1089  Buffer bucket_obuf;
1090  Buffer bucket_nbuf;
1091  Page opage;
1092  Page npage;
1093  HashPageOpaque oopaque;
1094  HashPageOpaque nopaque;
1095  OffsetNumber itup_offsets[MaxIndexTuplesPerPage];
1097  Size all_tups_size = 0;
1098  int i;
1099  uint16 nitups = 0;
1100 
1101  bucket_obuf = obuf;
1102  opage = BufferGetPage(obuf);
1103  oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
1104 
1105  bucket_nbuf = nbuf;
1106  npage = BufferGetPage(nbuf);
1107  nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
1108 
1109  /*
1110  * Partition the tuples in the old bucket between the old bucket and the
1111  * new bucket, advancing along the old bucket's overflow bucket chain and
1112  * adding overflow pages to the new bucket as needed. Outer loop iterates
1113  * once per page in old bucket.
1114  */
1115  for (;;)
1116  {
1117  BlockNumber oblkno;
1118  OffsetNumber ooffnum;
1119  OffsetNumber omaxoffnum;
1120 
1121  /* Scan each tuple in old page */
1122  omaxoffnum = PageGetMaxOffsetNumber(opage);
1123  for (ooffnum = FirstOffsetNumber;
1124  ooffnum <= omaxoffnum;
1125  ooffnum = OffsetNumberNext(ooffnum))
1126  {
1127  IndexTuple itup;
1128  Size itemsz;
1129  Bucket bucket;
1130  bool found = false;
1131 
1132  /* skip dead tuples */
1133  if (ItemIdIsDead(PageGetItemId(opage, ooffnum)))
1134  continue;
1135 
1136  /*
1137  * Before inserting a tuple, probe the hash table containing TIDs
1138  * of tuples belonging to new bucket, if we find a match, then
1139  * skip that tuple, else fetch the item's hash key (conveniently
1140  * stored in the item) and determine which bucket it now belongs
1141  * in.
1142  */
1143  itup = (IndexTuple) PageGetItem(opage,
1144  PageGetItemId(opage, ooffnum));
1145 
1146  if (htab)
1147  (void) hash_search(htab, &itup->t_tid, HASH_FIND, &found);
1148 
1149  if (found)
1150  continue;
1151 
1153  maxbucket, highmask, lowmask);
1154 
1155  if (bucket == nbucket)
1156  {
1157  IndexTuple new_itup;
1158 
1159  /*
1160  * make a copy of index tuple as we have to scribble on it.
1161  */
1162  new_itup = CopyIndexTuple(itup);
1163 
1164  /*
1165  * mark the index tuple as moved by split, such tuples are
1166  * skipped by scan if there is split in progress for a bucket.
1167  */
1168  new_itup->t_info |= INDEX_MOVED_BY_SPLIT_MASK;
1169 
1170  /*
1171  * insert the tuple into the new bucket. if it doesn't fit on
1172  * the current page in the new bucket, we must allocate a new
1173  * overflow page and place the tuple on that page instead.
1174  */
1175  itemsz = IndexTupleDSize(*new_itup);
1176  itemsz = MAXALIGN(itemsz);
1177 
1178  if (PageGetFreeSpaceForMultipleTuples(npage, nitups + 1) < (all_tups_size + itemsz))
1179  {
1180  /*
1181  * Change the shared buffer state in critical section,
1182  * otherwise any error could make it unrecoverable.
1183  */
1185 
1186  _hash_pgaddmultitup(rel, nbuf, itups, itup_offsets, nitups);
1187  MarkBufferDirty(nbuf);
1188  /* log the split operation before releasing the lock */
1189  log_split_page(rel, nbuf);
1190 
1191  END_CRIT_SECTION();
1192 
1193  /* drop lock, but keep pin */
1195 
1196  /* be tidy */
1197  for (i = 0; i < nitups; i++)
1198  pfree(itups[i]);
1199  nitups = 0;
1200  all_tups_size = 0;
1201 
1202  /* chain to a new overflow page */
1203  nbuf = _hash_addovflpage(rel, metabuf, nbuf, (nbuf == bucket_nbuf) ? true : false);
1204  npage = BufferGetPage(nbuf);
1205  nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
1206  }
1207 
1208  itups[nitups++] = new_itup;
1209  all_tups_size += itemsz;
1210  }
1211  else
1212  {
1213  /*
1214  * the tuple stays on this page, so nothing to do.
1215  */
1216  Assert(bucket == obucket);
1217  }
1218  }
1219 
1220  oblkno = oopaque->hasho_nextblkno;
1221 
1222  /* retain the pin on the old primary bucket */
1223  if (obuf == bucket_obuf)
1225  else
1226  _hash_relbuf(rel, obuf);
1227 
1228  /* Exit loop if no more overflow pages in old bucket */
1229  if (!BlockNumberIsValid(oblkno))
1230  {
1231  /*
1232  * Change the shared buffer state in critical section, otherwise
1233  * any error could make it unrecoverable.
1234  */
1236 
1237  _hash_pgaddmultitup(rel, nbuf, itups, itup_offsets, nitups);
1238  MarkBufferDirty(nbuf);
1239  /* log the split operation before releasing the lock */
1240  log_split_page(rel, nbuf);
1241 
1242  END_CRIT_SECTION();
1243 
1244  if (nbuf == bucket_nbuf)
1246  else
1247  _hash_relbuf(rel, nbuf);
1248 
1249  /* be tidy */
1250  for (i = 0; i < nitups; i++)
1251  pfree(itups[i]);
1252  break;
1253  }
1254 
1255  /* Else, advance to next old page */
1256  obuf = _hash_getbuf(rel, oblkno, HASH_READ, LH_OVERFLOW_PAGE);
1257  opage = BufferGetPage(obuf);
1258  oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
1259  }
1260 
1261  /*
1262  * We're at the end of the old bucket chain, so we're done partitioning
1263  * the tuples. Mark the old and new buckets to indicate split is
1264  * finished.
1265  *
1266  * To avoid deadlocks due to locking order of buckets, first lock the old
1267  * bucket and then the new bucket.
1268  */
1269  LockBuffer(bucket_obuf, BUFFER_LOCK_EXCLUSIVE);
1270  opage = BufferGetPage(bucket_obuf);
1271  oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
1272 
1273  LockBuffer(bucket_nbuf, BUFFER_LOCK_EXCLUSIVE);
1274  npage = BufferGetPage(bucket_nbuf);
1275  nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
1276 
1278 
1279  oopaque->hasho_flag &= ~LH_BUCKET_BEING_SPLIT;
1281 
1282  /*
1283  * After the split is finished, mark the old bucket to indicate that it
1284  * contains deletable tuples. We will clear split-cleanup flag after
1285  * deleting such tuples either at the end of split or at the next split
1286  * from old bucket or at the time of vacuum.
1287  */
1289 
1290  /*
1291  * now write the buffers, here we don't release the locks as caller is
1292  * responsible to release locks.
1293  */
1294  MarkBufferDirty(bucket_obuf);
1295  MarkBufferDirty(bucket_nbuf);
1296 
1297  if (RelationNeedsWAL(rel))
1298  {
1299  XLogRecPtr recptr;
1300  xl_hash_split_complete xlrec;
1301 
1302  xlrec.old_bucket_flag = oopaque->hasho_flag;
1303  xlrec.new_bucket_flag = nopaque->hasho_flag;
1304 
1305  XLogBeginInsert();
1306 
1307  XLogRegisterData((char *) &xlrec, SizeOfHashSplitComplete);
1308 
1309  XLogRegisterBuffer(0, bucket_obuf, REGBUF_STANDARD);
1310  XLogRegisterBuffer(1, bucket_nbuf, REGBUF_STANDARD);
1311 
1312  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_COMPLETE);
1313 
1314  PageSetLSN(BufferGetPage(bucket_obuf), recptr);
1315  PageSetLSN(BufferGetPage(bucket_nbuf), recptr);
1316  }
1317 
1318  END_CRIT_SECTION();
1319 
1320  /*
1321  * If possible, clean up the old bucket. We might not be able to do this
1322  * if someone else has a pin on it, but if not then we can go ahead. This
1323  * isn't absolutely necessary, but it reduces bloat; if we don't do it
1324  * now, VACUUM will do it eventually, but maybe not until new overflow
1325  * pages have been allocated. Note that there's no need to clean up the
1326  * new bucket.
1327  */
1328  if (IsBufferCleanupOK(bucket_obuf))
1329  {
1330  LockBuffer(bucket_nbuf, BUFFER_LOCK_UNLOCK);
1331  hashbucketcleanup(rel, obucket, bucket_obuf,
1332  BufferGetBlockNumber(bucket_obuf), NULL,
1333  maxbucket, highmask, lowmask, NULL, NULL, true,
1334  NULL, NULL);
1335  }
1336  else
1337  {
1338  LockBuffer(bucket_nbuf, BUFFER_LOCK_UNLOCK);
1339  LockBuffer(bucket_obuf, BUFFER_LOCK_UNLOCK);
1340  }
1341 }
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:87
void _hash_pgaddmultitup(Relation rel, Buffer buf, IndexTuple *itups, OffsetNumber *itup_offsets, uint16 nitups)
Definition: hashinsert.c:297
Bucket _hash_hashkey2bucket(uint32 hashkey, uint32 maxbucket, uint32 highmask, uint32 lowmask)
Definition: hashutil.c:125
void MarkBufferDirty(Buffer buffer)
Definition: bufmgr.c:1450
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:133
#define START_CRIT_SECTION()
Definition: miscadmin.h:131
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:672
uint32 BlockNumber
Definition: block.h:31
void * hash_search(HTAB *hashp, const void *keyPtr, HASHACTION action, bool *foundPtr)
Definition: dynahash.c:902
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:78
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:89
#define LH_BUCKET_NEEDS_SPLIT_CLEANUP
Definition: hash.h:70
#define ItemIdIsDead(itemId)
Definition: itemid.h:112
#define PageGetMaxOffsetNumber(page)
Definition: bufpage.h:353
uint16 OffsetNumber
Definition: off.h:24
#define HASH_READ
Definition: hash.h:331
uint32 Bucket
Definition: hash.h:34
unsigned short uint16
Definition: c.h:257
void pfree(void *pointer)
Definition: mcxt.c:949
uint32 _hash_get_indextuple_hashkey(IndexTuple itup)
Definition: hashutil.c:299
IndexTuple CopyIndexTuple(IndexTuple source)
Definition: indextuple.c:438
#define IndexTupleDSize(itup)
Definition: itup.h:71
#define FirstOffsetNumber
Definition: off.h:27
IndexTupleData * IndexTuple
Definition: itup.h:53
#define REGBUF_STANDARD
Definition: xloginsert.h:34
#define LH_BUCKET_BEING_SPLIT
Definition: hash.h:69
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
bool IsBufferCleanupOK(Buffer buffer)
Definition: bufmgr.c:3774
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:231
void XLogRegisterData(char *data, int len)
Definition: xloginsert.c:323
#define LH_BUCKET_BEING_POPULATED
Definition: hash.h:68
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:415
#define LH_OVERFLOW_PAGE
Definition: hash.h:64
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
Size PageGetFreeSpaceForMultipleTuples(Page page, int ntups)
Definition: bufpage.c:609
static void log_split_page(Relation rel, Buffer buf)
Definition: hashpage.c:1475
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:274
#define BlockNumberIsValid(blockNumber)
Definition: block.h:70
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:664
#define SizeOfHashSplitComplete
Definition: hash_xlog.h:136
#define OffsetNumberNext(offsetNumber)
Definition: off.h:53
size_t Size
Definition: c.h:350
#define PageGetSpecialPointer(page)
Definition: bufpage.h:322
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:96
Buffer _hash_addovflpage(Relation rel, Buffer metabuf, Buffer buf, bool retain_pin)
Definition: hashovfl.c:111
#define MAXALIGN(LEN)
Definition: c.h:576
#define RelationNeedsWAL(relation)
Definition: rel.h:505
#define XLOG_HASH_SPLIT_COMPLETE
Definition: hash_xlog.h:33
uint16 hasho_flag
Definition: hash.h:92
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:2605
#define MaxIndexTuplesPerPage
Definition: itup.h:137
int i
BlockNumber hasho_nextblkno
Definition: hash.h:90
unsigned short t_info
Definition: itup.h:49
#define INDEX_MOVED_BY_SPLIT_MASK
Definition: hash.h:285
void XLogBeginInsert(void)
Definition: xloginsert.c:120
#define PageSetLSN(page, lsn)
Definition: bufpage.h:364
int Buffer
Definition: buf.h:23
#define PageGetItem(page, itemId)
Definition: bufpage.h:336
Pointer Page
Definition: bufpage.h:74
static void log_split_page ( Relation  rel,
Buffer  buf 
)
static

Definition at line 1475 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().

1476 {
1477  if (RelationNeedsWAL(rel))
1478  {
1479  XLogRecPtr recptr;
1480 
1481  XLogBeginInsert();
1482 
1484 
1485  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_PAGE);
1486 
1487  PageSetLSN(BufferGetPage(buf), recptr);
1488  }
1489 }
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:34
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define REGBUF_FORCE_IMAGE
Definition: xloginsert.h:30
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:415
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define RelationNeedsWAL(relation)
Definition: rel.h:505
void XLogBeginInsert(void)
Definition: xloginsert.c:120
#define PageSetLSN(page, lsn)
Definition: bufpage.h:364