PostgreSQL Source Code  git master
hash.h File Reference
#include "access/amapi.h"
#include "access/itup.h"
#include "access/sdir.h"
#include "fmgr.h"
#include "lib/stringinfo.h"
#include "storage/bufmgr.h"
#include "storage/lockdefs.h"
#include "utils/hsearch.h"
#include "utils/relcache.h"
Include dependency graph for hash.h:
This graph shows which files directly or indirectly include this file:

Go to the source code of this file.

Data Structures

struct  HashPageOpaqueData
 
struct  HashScanPosItem
 
struct  HashScanPosData
 
struct  HashScanOpaqueData
 
struct  HashMetaPageData
 

Macros

#define InvalidBucket   ((Bucket) 0xFFFFFFFF)
 
#define BUCKET_TO_BLKNO(metap, B)   ((BlockNumber) ((B) + ((B) ? (metap)->hashm_spares[_hash_spareindex((B)+1)-1] : 0)) + 1)
 
#define ROTATE_HIGH_AND_LOW_32BITS(v)
 
#define LH_UNUSED_PAGE   (0)
 
#define LH_OVERFLOW_PAGE   (1 << 0)
 
#define LH_BUCKET_PAGE   (1 << 1)
 
#define LH_BITMAP_PAGE   (1 << 2)
 
#define LH_META_PAGE   (1 << 3)
 
#define LH_BUCKET_BEING_POPULATED   (1 << 4)
 
#define LH_BUCKET_BEING_SPLIT   (1 << 5)
 
#define LH_BUCKET_NEEDS_SPLIT_CLEANUP   (1 << 6)
 
#define LH_PAGE_HAS_DEAD_TUPLES   (1 << 7)
 
#define LH_PAGE_TYPE   (LH_OVERFLOW_PAGE | LH_BUCKET_PAGE | LH_BITMAP_PAGE | LH_META_PAGE)
 
#define H_NEEDS_SPLIT_CLEANUP(opaque)   (((opaque)->hasho_flag & LH_BUCKET_NEEDS_SPLIT_CLEANUP) != 0)
 
#define H_BUCKET_BEING_SPLIT(opaque)   (((opaque)->hasho_flag & LH_BUCKET_BEING_SPLIT) != 0)
 
#define H_BUCKET_BEING_POPULATED(opaque)   (((opaque)->hasho_flag & LH_BUCKET_BEING_POPULATED) != 0)
 
#define H_HAS_DEAD_TUPLES(opaque)   (((opaque)->hasho_flag & LH_PAGE_HAS_DEAD_TUPLES) != 0)
 
#define HASHO_PAGE_ID   0xFF80
 
#define HashScanPosIsPinned(scanpos)
 
#define HashScanPosIsValid(scanpos)
 
#define HashScanPosInvalidate(scanpos)
 
#define HASH_METAPAGE   0 /* metapage is always block 0 */
 
#define HASH_MAGIC   0x6440640
 
#define HASH_VERSION   4
 
#define HASH_MAX_BITMAPS   1024
 
#define HASH_SPLITPOINT_PHASE_BITS   2
 
#define HASH_SPLITPOINT_PHASES_PER_GRP   (1 << HASH_SPLITPOINT_PHASE_BITS)
 
#define HASH_SPLITPOINT_PHASE_MASK   (HASH_SPLITPOINT_PHASES_PER_GRP - 1)
 
#define HASH_SPLITPOINT_GROUPS_WITH_ONE_PHASE   10
 
#define HASH_MAX_SPLITPOINT_GROUP   32
 
#define HASH_MAX_SPLITPOINTS
 
#define HashMaxItemSize(page)
 
#define INDEX_MOVED_BY_SPLIT_MASK   INDEX_AM_RESERVED_BIT
 
#define HASH_MIN_FILLFACTOR   10
 
#define HASH_DEFAULT_FILLFACTOR   75
 
#define BYTE_TO_BIT   3 /* 2^3 bits/byte */
 
#define ALL_SET   ((uint32) ~0)
 
#define BMPGSZ_BYTE(metap)   ((metap)->hashm_bmsize)
 
#define BMPGSZ_BIT(metap)   ((metap)->hashm_bmsize << BYTE_TO_BIT)
 
#define BMPG_SHIFT(metap)   ((metap)->hashm_bmshift)
 
#define BMPG_MASK(metap)   (BMPGSZ_BIT(metap) - 1)
 
#define HashPageGetBitmap(page)   ((uint32 *) PageGetContents(page))
 
#define HashGetMaxBitmapSize(page)
 
#define HashPageGetMeta(page)   ((HashMetaPage) PageGetContents(page))
 
#define BITS_PER_MAP   32 /* Number of bits in uint32 */
 
#define CLRBIT(A, N)   ((A)[(N)/BITS_PER_MAP] &= ~(1<<((N)%BITS_PER_MAP)))
 
#define SETBIT(A, N)   ((A)[(N)/BITS_PER_MAP] |= (1<<((N)%BITS_PER_MAP)))
 
#define ISSET(A, N)   ((A)[(N)/BITS_PER_MAP] & (1<<((N)%BITS_PER_MAP)))
 
#define HASH_READ   BUFFER_LOCK_SHARE
 
#define HASH_WRITE   BUFFER_LOCK_EXCLUSIVE
 
#define HASH_NOLOCK   (-1)
 
#define HTEqualStrategyNumber   1
 
#define HTMaxStrategyNumber   1
 
#define HASHSTANDARD_PROC   1
 
#define HASHEXTENDED_PROC   2
 
#define HASHNProcs   2
 

Typedefs

typedef uint32 Bucket
 
typedef struct HashPageOpaqueData HashPageOpaqueData
 
typedef HashPageOpaqueDataHashPageOpaque
 
typedef struct HashScanPosItem HashScanPosItem
 
typedef struct HashScanPosData HashScanPosData
 
typedef struct HashScanOpaqueData HashScanOpaqueData
 
typedef HashScanOpaqueDataHashScanOpaque
 
typedef struct HashMetaPageData HashMetaPageData
 
typedef HashMetaPageDataHashMetaPage
 
typedef struct HSpool HSpool
 

Functions

IndexBuildResulthashbuild (Relation heap, Relation index, struct IndexInfo *indexInfo)
 
void hashbuildempty (Relation index)
 
bool hashinsert (Relation rel, Datum *values, bool *isnull, ItemPointer ht_ctid, Relation heapRel, IndexUniqueCheck checkUnique, struct IndexInfo *indexInfo)
 
bool hashgettuple (IndexScanDesc scan, ScanDirection dir)
 
int64 hashgetbitmap (IndexScanDesc scan, TIDBitmap *tbm)
 
IndexScanDesc hashbeginscan (Relation rel, int nkeys, int norderbys)
 
void hashrescan (IndexScanDesc scan, ScanKey scankey, int nscankeys, ScanKey orderbys, int norderbys)
 
void hashendscan (IndexScanDesc scan)
 
IndexBulkDeleteResulthashbulkdelete (IndexVacuumInfo *info, IndexBulkDeleteResult *stats, IndexBulkDeleteCallback callback, void *callback_state)
 
IndexBulkDeleteResulthashvacuumcleanup (IndexVacuumInfo *info, IndexBulkDeleteResult *stats)
 
byteahashoptions (Datum reloptions, bool validate)
 
bool hashvalidate (Oid opclassoid)
 
Datum hash_any (register const unsigned char *k, register int keylen)
 
Datum hash_any_extended (register const unsigned char *k, register int keylen, uint64 seed)
 
Datum hash_uint32 (uint32 k)
 
Datum hash_uint32_extended (uint32 k, uint64 seed)
 
void _hash_doinsert (Relation rel, IndexTuple itup, Relation heapRel)
 
OffsetNumber _hash_pgaddtup (Relation rel, Buffer buf, Size itemsize, IndexTuple itup)
 
void _hash_pgaddmultitup (Relation rel, Buffer buf, IndexTuple *itups, OffsetNumber *itup_offsets, uint16 nitups)
 
Buffer _hash_addovflpage (Relation rel, Buffer metabuf, Buffer buf, bool retain_pin)
 
BlockNumber _hash_freeovflpage (Relation rel, Buffer bucketbuf, Buffer ovflbuf, Buffer wbuf, IndexTuple *itups, OffsetNumber *itup_offsets, Size *tups_size, uint16 nitups, BufferAccessStrategy bstrategy)
 
void _hash_initbitmapbuffer (Buffer buf, uint16 bmsize, bool initpage)
 
void _hash_squeezebucket (Relation rel, Bucket bucket, BlockNumber bucket_blkno, Buffer bucket_buf, BufferAccessStrategy bstrategy)
 
uint32 _hash_ovflblkno_to_bitno (HashMetaPage metap, BlockNumber ovflblkno)
 
Buffer _hash_getbuf (Relation rel, BlockNumber blkno, int access, int flags)
 
Buffer _hash_getbuf_with_condlock_cleanup (Relation rel, BlockNumber blkno, int flags)
 
HashMetaPage _hash_getcachedmetap (Relation rel, Buffer *metabuf, bool force_refresh)
 
Buffer _hash_getbucketbuf_from_hashkey (Relation rel, uint32 hashkey, int access, HashMetaPage *cachedmetap)
 
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)
 
bool _hash_next (IndexScanDesc scan, ScanDirection dir)
 
bool _hash_first (IndexScanDesc scan, ScanDirection dir)
 
HSpool_h_spoolinit (Relation heap, Relation index, uint32 num_buckets)
 
void _h_spooldestroy (HSpool *hspool)
 
void _h_spool (HSpool *hspool, ItemPointer self, Datum *values, bool *isnull)
 
void _h_indexbuild (HSpool *hspool, Relation heapRel)
 
bool _hash_checkqual (IndexScanDesc scan, IndexTuple itup)
 
uint32 _hash_datum2hashkey (Relation rel, Datum key)
 
uint32 _hash_datum2hashkey_type (Relation rel, Datum key, Oid keytype)
 
Bucket _hash_hashkey2bucket (uint32 hashkey, uint32 maxbucket, uint32 highmask, uint32 lowmask)
 
uint32 _hash_log2 (uint32 num)
 
uint32 _hash_spareindex (uint32 num_bucket)
 
uint32 _hash_get_totalbuckets (uint32 splitpoint_phase)
 
void _hash_checkpage (Relation rel, Buffer buf, int flags)
 
uint32 _hash_get_indextuple_hashkey (IndexTuple itup)
 
bool _hash_convert_tuple (Relation index, Datum *user_values, bool *user_isnull, Datum *index_values, bool *index_isnull)
 
OffsetNumber _hash_binsearch (Page page, uint32 hash_value)
 
OffsetNumber _hash_binsearch_last (Page page, uint32 hash_value)
 
BlockNumber _hash_get_oldblock_from_newbucket (Relation rel, Bucket new_bucket)
 
BlockNumber _hash_get_newblock_from_oldbucket (Relation rel, Bucket old_bucket)
 
Bucket _hash_get_newbucket_from_oldbucket (Relation rel, Bucket old_bucket, uint32 lowmask, uint32 maxbucket)
 
void _hash_kill_items (IndexScanDesc scan)
 
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 bucket_has_garbage, IndexBulkDeleteCallback callback, void *callback_state)
 

Macro Definition Documentation

◆ ALL_SET

#define ALL_SET   ((uint32) ~0)

Definition at line 292 of file hash.h.

Referenced by _hash_addovflpage().

◆ BITS_PER_MAP

#define BITS_PER_MAP   32 /* Number of bits in uint32 */

Definition at line 319 of file hash.h.

Referenced by _hash_addovflpage(), and _hash_firstfreebit().

◆ BMPG_MASK

#define BMPG_MASK (   metap)    (BMPGSZ_BIT(metap) - 1)

◆ BMPG_SHIFT

#define BMPG_SHIFT (   metap)    ((metap)->hashm_bmshift)

◆ BMPGSZ_BIT

#define BMPGSZ_BIT (   metap)    ((metap)->hashm_bmsize << BYTE_TO_BIT)

Definition at line 302 of file hash.h.

Referenced by _hash_addovflpage().

◆ BMPGSZ_BYTE

#define BMPGSZ_BYTE (   metap)    ((metap)->hashm_bmsize)

Definition at line 301 of file hash.h.

◆ BUCKET_TO_BLKNO

#define BUCKET_TO_BLKNO (   metap,
 
)    ((BlockNumber) ((B) + ((B) ? (metap)->hashm_spares[_hash_spareindex((B)+1)-1] : 0)) + 1)

◆ BYTE_TO_BIT

#define BYTE_TO_BIT   3 /* 2^3 bits/byte */

Definition at line 291 of file hash.h.

Referenced by _hash_init_metabuffer().

◆ CLRBIT

#define CLRBIT (   A,
 
)    ((A)[(N)/BITS_PER_MAP] &= ~(1<<((N)%BITS_PER_MAP)))

Definition at line 322 of file hash.h.

◆ H_BUCKET_BEING_POPULATED

#define H_BUCKET_BEING_POPULATED (   opaque)    (((opaque)->hasho_flag & LH_BUCKET_BEING_POPULATED) != 0)

Definition at line 100 of file hash.h.

Referenced by _hash_first().

◆ H_BUCKET_BEING_SPLIT

#define H_BUCKET_BEING_SPLIT (   opaque)    (((opaque)->hasho_flag & LH_BUCKET_BEING_SPLIT) != 0)

Definition at line 99 of file hash.h.

Referenced by _hash_doinsert(), _hash_expandtable(), and hashbulkdelete().

◆ H_HAS_DEAD_TUPLES

#define H_HAS_DEAD_TUPLES (   opaque)    (((opaque)->hasho_flag & LH_PAGE_HAS_DEAD_TUPLES) != 0)

Definition at line 101 of file hash.h.

Referenced by _hash_doinsert(), and hashbucketcleanup().

◆ H_NEEDS_SPLIT_CLEANUP

#define H_NEEDS_SPLIT_CLEANUP (   opaque)    (((opaque)->hasho_flag & LH_BUCKET_NEEDS_SPLIT_CLEANUP) != 0)

Definition at line 98 of file hash.h.

Referenced by _hash_expandtable(), and hashbulkdelete().

◆ HASH_DEFAULT_FILLFACTOR

#define HASH_DEFAULT_FILLFACTOR   75

Definition at line 286 of file hash.h.

Referenced by _hash_init().

◆ HASH_MAGIC

#define HASH_MAGIC   0x6440640

Definition at line 208 of file hash.h.

Referenced by _hash_checkpage(), _hash_init_metabuffer(), and verify_hash_page().

◆ HASH_MAX_BITMAPS

#define HASH_MAX_BITMAPS   1024

Definition at line 235 of file hash.h.

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

◆ HASH_MAX_SPLITPOINT_GROUP

#define HASH_MAX_SPLITPOINT_GROUP   32

Definition at line 243 of file hash.h.

◆ HASH_MAX_SPLITPOINTS

#define HASH_MAX_SPLITPOINTS
Value:
HASH_SPLITPOINT_GROUPS_WITH_ONE_PHASE)
#define HASH_MAX_SPLITPOINT_GROUP
Definition: hash.h:243
#define HASH_SPLITPOINT_PHASES_PER_GRP
Definition: hash.h:238
#define HASH_SPLITPOINT_GROUPS_WITH_ONE_PHASE
Definition: hash.h:240

Definition at line 244 of file hash.h.

Referenced by _hash_init_metabuffer(), and hash_metapage_info().

◆ HASH_METAPAGE

◆ HASH_MIN_FILLFACTOR

#define HASH_MIN_FILLFACTOR   10

Definition at line 285 of file hash.h.

◆ HASH_NOLOCK

#define HASH_NOLOCK   (-1)

Definition at line 331 of file hash.h.

Referenced by _hash_doinsert(), _hash_getbuf(), _hash_getbuf_with_strategy(), and hashbulkdelete().

◆ HASH_READ

◆ HASH_SPLITPOINT_GROUPS_WITH_ONE_PHASE

#define HASH_SPLITPOINT_GROUPS_WITH_ONE_PHASE   10

Definition at line 240 of file hash.h.

Referenced by _hash_get_totalbuckets(), and _hash_spareindex().

◆ HASH_SPLITPOINT_PHASE_BITS

#define HASH_SPLITPOINT_PHASE_BITS   2

Definition at line 237 of file hash.h.

Referenced by _hash_get_totalbuckets(), and _hash_spareindex().

◆ HASH_SPLITPOINT_PHASE_MASK

#define HASH_SPLITPOINT_PHASE_MASK   (HASH_SPLITPOINT_PHASES_PER_GRP - 1)

Definition at line 239 of file hash.h.

Referenced by _hash_get_totalbuckets(), and _hash_spareindex().

◆ HASH_SPLITPOINT_PHASES_PER_GRP

#define HASH_SPLITPOINT_PHASES_PER_GRP   (1 << HASH_SPLITPOINT_PHASE_BITS)

Definition at line 238 of file hash.h.

◆ HASH_VERSION

#define HASH_VERSION   4

Definition at line 209 of file hash.h.

Referenced by _hash_checkpage(), _hash_init_metabuffer(), and verify_hash_page().

◆ HASH_WRITE

◆ HASHEXTENDED_PROC

◆ HashGetMaxBitmapSize

#define HashGetMaxBitmapSize (   page)
Value:
(PageGetPageSize((Page) page) - \
#define SizeOfPageHeaderData
Definition: bufpage.h:212
#define PageGetPageSize(page)
Definition: bufpage.h:264
#define MAXALIGN(LEN)
Definition: c.h:652
Pointer Page
Definition: bufpage.h:74

Definition at line 309 of file hash.h.

Referenced by _hash_init_metabuffer().

◆ HashMaxItemSize

#define HashMaxItemSize (   page)
Value:
sizeof(ItemIdData) - \
#define SizeOfPageHeaderData
Definition: bufpage.h:212
#define PageGetPageSize(page)
Definition: bufpage.h:264
#define MAXALIGN(LEN)
Definition: c.h:652
#define MAXALIGN_DOWN(LEN)
Definition: c.h:664

Definition at line 277 of file hash.h.

Referenced by _hash_doinsert().

◆ HASHNProcs

#define HASHNProcs   2

Definition at line 353 of file hash.h.

Referenced by hashhandler().

◆ HASHO_PAGE_ID

◆ HashPageGetBitmap

#define HashPageGetBitmap (   page)    ((uint32 *) PageGetContents(page))

◆ HashPageGetMeta

◆ HashScanPosInvalidate

#define HashScanPosInvalidate (   scanpos)
Value:
do { \
(scanpos).buf = InvalidBuffer; \
(scanpos).currPage = InvalidBlockNumber; \
(scanpos).nextPage = InvalidBlockNumber; \
(scanpos).prevPage = InvalidBlockNumber; \
(scanpos).firstItem = 0; \
(scanpos).lastItem = 0; \
(scanpos).itemIndex = 0; \
} while (0);
#define InvalidBuffer
Definition: buf.h:25
static char * buf
Definition: pg_test_fsync.c:67
#define InvalidBlockNumber
Definition: block.h:33

Definition at line 152 of file hash.h.

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

◆ HashScanPosIsPinned

#define HashScanPosIsPinned (   scanpos)
Value:
( \
AssertMacro(BlockNumberIsValid((scanpos).currPage) || \
!BufferIsValid((scanpos).buf)), \
BufferIsValid((scanpos).buf) \
)
static char * buf
Definition: pg_test_fsync.c:67
#define BlockNumberIsValid(blockNumber)
Definition: block.h:70
#define BufferIsValid(bufnum)
Definition: bufmgr.h:114

Definition at line 138 of file hash.h.

Referenced by _hash_kill_items().

◆ HashScanPosIsValid

#define HashScanPosIsValid (   scanpos)
Value:
( \
AssertMacro(BlockNumberIsValid((scanpos).currPage) || \
!BufferIsValid((scanpos).buf)), \
BlockNumberIsValid((scanpos).currPage) \
)
static char * buf
Definition: pg_test_fsync.c:67
#define BlockNumberIsValid(blockNumber)
Definition: block.h:70
#define BufferIsValid(bufnum)
Definition: bufmgr.h:114

Definition at line 145 of file hash.h.

Referenced by _hash_kill_items(), hashendscan(), hashgettuple(), and hashrescan().

◆ HASHSTANDARD_PROC

◆ HTEqualStrategyNumber

◆ HTMaxStrategyNumber

#define HTMaxStrategyNumber   1

Definition at line 337 of file hash.h.

Referenced by gen_prune_steps_from_opexps(), hashhandler(), and hashvalidate().

◆ INDEX_MOVED_BY_SPLIT_MASK

#define INDEX_MOVED_BY_SPLIT_MASK   INDEX_AM_RESERVED_BIT

Definition at line 283 of file hash.h.

Referenced by _hash_load_qualified_items(), and _hash_splitbucket().

◆ InvalidBucket

#define InvalidBucket   ((Bucket) 0xFFFFFFFF)

Definition at line 36 of file hash.h.

Referenced by hashbucketcleanup().

◆ ISSET

#define ISSET (   A,
 
)    ((A)[(N)/BITS_PER_MAP] & (1<<((N)%BITS_PER_MAP)))

Definition at line 324 of file hash.h.

Referenced by _hash_freeovflpage(), and hash_bitmap_info().

◆ LH_BITMAP_PAGE

◆ LH_BUCKET_BEING_POPULATED

#define LH_BUCKET_BEING_POPULATED   (1 << 4)

Definition at line 68 of file hash.h.

Referenced by _hash_expandtable(), and _hash_splitbucket().

◆ LH_BUCKET_BEING_SPLIT

#define LH_BUCKET_BEING_SPLIT   (1 << 5)

Definition at line 69 of file hash.h.

Referenced by _hash_expandtable(), and _hash_splitbucket().

◆ LH_BUCKET_NEEDS_SPLIT_CLEANUP

#define LH_BUCKET_NEEDS_SPLIT_CLEANUP   (1 << 6)

Definition at line 70 of file hash.h.

Referenced by _hash_splitbucket(), hash_xlog_split_cleanup(), and hashbucketcleanup().

◆ LH_BUCKET_PAGE

◆ LH_META_PAGE

◆ LH_OVERFLOW_PAGE

◆ LH_PAGE_HAS_DEAD_TUPLES

#define LH_PAGE_HAS_DEAD_TUPLES   (1 << 7)

◆ LH_PAGE_TYPE

◆ LH_UNUSED_PAGE

#define LH_UNUSED_PAGE   (0)

◆ ROTATE_HIGH_AND_LOW_32BITS

#define ROTATE_HIGH_AND_LOW_32BITS (   v)
Value:
((((v) << 1) & UINT64CONST(0xfffffffefffffffe)) | \
(((v) >> 31) & UINT64CONST(0x100000001)))

Definition at line 48 of file hash.h.

Referenced by hash_range_extended(), and JsonbHashScalarValueExtended().

◆ SETBIT

#define SETBIT (   A,
 
)    ((A)[(N)/BITS_PER_MAP] |= (1<<((N)%BITS_PER_MAP)))

Definition at line 323 of file hash.h.

Typedef Documentation

◆ Bucket

Definition at line 34 of file hash.h.

◆ HashMetaPage

Definition at line 272 of file hash.h.

◆ HashMetaPageData

◆ HashPageOpaque

Definition at line 96 of file hash.h.

◆ HashPageOpaqueData

◆ HashScanOpaque

Definition at line 200 of file hash.h.

◆ HashScanOpaqueData

◆ HashScanPosData

◆ HashScanPosItem

◆ HSpool

Definition at line 443 of file hash.h.

Function Documentation

◆ _h_indexbuild()

void _h_indexbuild ( HSpool hspool,
Relation  heapRel 
)

Definition at line 116 of file hashsort.c.

References _hash_doinsert(), _hash_get_indextuple_hashkey(), _hash_hashkey2bucket(), Assert, HSpool::high_mask, HSpool::index, HSpool::low_mask, HSpool::max_buckets, HSpool::sortstate, tuplesort_getindextuple(), and tuplesort_performsort().

Referenced by hashbuild().

117 {
118  IndexTuple itup;
119 #ifdef USE_ASSERT_CHECKING
120  uint32 hashkey = 0;
121 #endif
122 
124 
125  while ((itup = tuplesort_getindextuple(hspool->sortstate, true)) != NULL)
126  {
127  /*
128  * Technically, it isn't critical that hash keys be found in sorted
129  * order, since this sorting is only used to increase locality of
130  * access as a performance optimization. It still seems like a good
131  * idea to test tuplesort.c's handling of hash index tuple sorts
132  * through an assertion, though.
133  */
134 #ifdef USE_ASSERT_CHECKING
135  uint32 lasthashkey = hashkey;
136 
138  hspool->max_buckets, hspool->high_mask,
139  hspool->low_mask);
140  Assert(hashkey >= lasthashkey);
141 #endif
142 
143  _hash_doinsert(hspool->index, itup, heapRel);
144  }
145 }
IndexTuple tuplesort_getindextuple(Tuplesortstate *state, bool forward)
Definition: tuplesort.c:2215
void _hash_doinsert(Relation rel, IndexTuple itup, Relation heapRel)
Definition: hashinsert.c:37
void tuplesort_performsort(Tuplesortstate *state)
Definition: tuplesort.c:1790
Bucket _hash_hashkey2bucket(uint32 hashkey, uint32 maxbucket, uint32 highmask, uint32 lowmask)
Definition: hashutil.c:125
Tuplesortstate * sortstate
Definition: hashsort.c:38
uint32 low_mask
Definition: hashsort.c:47
uint32 high_mask
Definition: hashsort.c:46
uint32 _hash_get_indextuple_hashkey(IndexTuple itup)
Definition: hashutil.c:299
unsigned int uint32
Definition: c.h:325
uint32 max_buckets
Definition: hashsort.c:48
Relation index
Definition: hashsort.c:39
#define Assert(condition)
Definition: c.h:699

◆ _h_spool()

void _h_spool ( HSpool hspool,
ItemPointer  self,
Datum values,
bool isnull 
)

Definition at line 105 of file hashsort.c.

References HSpool::index, HSpool::sortstate, and tuplesort_putindextuplevalues().

Referenced by hashbuildCallback().

106 {
108  self, values, isnull);
109 }
Tuplesortstate * sortstate
Definition: hashsort.c:38
Relation index
Definition: hashsort.c:39
void tuplesort_putindextuplevalues(Tuplesortstate *state, Relation rel, ItemPointer self, Datum *values, bool *isnull)
Definition: tuplesort.c:1477
static Datum values[MAXATTR]
Definition: bootstrap.c:164

◆ _h_spooldestroy()

void _h_spooldestroy ( HSpool hspool)

Definition at line 95 of file hashsort.c.

References pfree(), HSpool::sortstate, and tuplesort_end().

Referenced by hashbuild().

96 {
97  tuplesort_end(hspool->sortstate);
98  pfree(hspool);
99 }
Tuplesortstate * sortstate
Definition: hashsort.c:38
void pfree(void *pointer)
Definition: mcxt.c:1031
void tuplesort_end(Tuplesortstate *state)
Definition: tuplesort.c:1235

◆ _h_spoolinit()

HSpool* _h_spoolinit ( Relation  heap,
Relation  index,
uint32  num_buckets 
)

Definition at line 56 of file hashsort.c.

References _hash_log2(), HSpool::high_mask, HSpool::index, HSpool::low_mask, maintenance_work_mem, HSpool::max_buckets, palloc0(), HSpool::sortstate, and tuplesort_begin_index_hash().

Referenced by hashbuild().

57 {
58  HSpool *hspool = (HSpool *) palloc0(sizeof(HSpool));
59 
60  hspool->index = index;
61 
62  /*
63  * Determine the bitmask for hash code values. Since there are currently
64  * num_buckets buckets in the index, the appropriate mask can be computed
65  * as follows.
66  *
67  * NOTE : This hash mask calculation should be in sync with similar
68  * calculation in _hash_init_metabuffer.
69  */
70  hspool->high_mask = (((uint32) 1) << _hash_log2(num_buckets + 1)) - 1;
71  hspool->low_mask = (hspool->high_mask >> 1);
72  hspool->max_buckets = num_buckets - 1;
73 
74  /*
75  * We size the sort area as maintenance_work_mem rather than work_mem to
76  * speed index creation. This should be OK since a single backend can't
77  * run multiple index creations in parallel.
78  */
80  index,
81  hspool->high_mask,
82  hspool->low_mask,
83  hspool->max_buckets,
85  NULL,
86  false);
87 
88  return hspool;
89 }
Tuplesortstate * tuplesort_begin_index_hash(Relation heapRel, Relation indexRel, uint32 high_mask, uint32 low_mask, uint32 max_buckets, int workMem, SortCoordinate coordinate, bool randomAccess)
Definition: tuplesort.c:1053
Tuplesortstate * sortstate
Definition: hashsort.c:38
uint32 low_mask
Definition: hashsort.c:47
uint32 high_mask
Definition: hashsort.c:46
unsigned int uint32
Definition: c.h:325
void * palloc0(Size size)
Definition: mcxt.c:955
uint32 max_buckets
Definition: hashsort.c:48
Relation index
Definition: hashsort.c:39
int maintenance_work_mem
Definition: globals.c:121
uint32 _hash_log2(uint32 num)
Definition: hashutil.c:141

◆ _hash_addovflpage()

Buffer _hash_addovflpage ( Relation  rel,
Buffer  metabuf,
Buffer  buf,
bool  retain_pin 
)

Definition at line 111 of file hashovfl.c.

References _hash_checkpage(), _hash_firstfreebit(), _hash_getbuf(), _hash_getinitbuf(), _hash_getnewbuf(), _hash_initbitmapbuffer(), _hash_relbuf(), ALL_SET, Assert, bit(), bitno_to_blkno(), BITS_PER_MAP, BlockNumberIsValid, xl_hash_add_ovfl_page::bmpage_found, BMPG_MASK, BMPG_SHIFT, BMPGSZ_BIT, xl_hash_add_ovfl_page::bmsize, BUFFER_LOCK_EXCLUSIVE, BUFFER_LOCK_UNLOCK, BufferGetBlockNumber(), BufferGetPage, BufferIsValid, END_CRIT_SECTION, ereport, errcode(), errmsg(), ERROR, HASH_MAX_BITMAPS, HASH_WRITE, HashMetaPageData::hashm_bmsize, HashMetaPageData::hashm_firstfree, HashMetaPageData::hashm_mapp, HashMetaPageData::hashm_nmaps, 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, HashPageGetBitmap, HashPageGetMeta, i, InvalidBlockNumber, InvalidBuffer, LH_BITMAP_PAGE, LH_BUCKET_PAGE, LH_META_PAGE, LH_OVERFLOW_PAGE, LH_PAGE_TYPE, LockBuffer(), MAIN_FORKNUM, MarkBufferDirty(), PageGetSpecialPointer, PageSetLSN, REGBUF_STANDARD, REGBUF_WILL_INIT, RelationGetRelationName, RelationNeedsWAL, SETBIT, SizeOfHashAddOvflPage, START_CRIT_SECTION, XLOG_HASH_ADD_OVFL_PAGE, XLogBeginInsert(), XLogInsert(), XLogRegisterBufData(), XLogRegisterBuffer(), and XLogRegisterData().

Referenced by _hash_doinsert(), and _hash_splitbucket().

112 {
113  Buffer ovflbuf;
114  Page page;
115  Page ovflpage;
116  HashPageOpaque pageopaque;
117  HashPageOpaque ovflopaque;
118  HashMetaPage metap;
119  Buffer mapbuf = InvalidBuffer;
120  Buffer newmapbuf = InvalidBuffer;
121  BlockNumber blkno;
122  uint32 orig_firstfree;
123  uint32 splitnum;
124  uint32 *freep = NULL;
125  uint32 max_ovflpg;
126  uint32 bit;
127  uint32 bitmap_page_bit;
128  uint32 first_page;
129  uint32 last_bit;
130  uint32 last_page;
131  uint32 i,
132  j;
133  bool page_found = false;
134 
135  /*
136  * Write-lock the tail page. Here, we need to maintain locking order such
137  * that, first acquire the lock on tail page of bucket, then on meta page
138  * to find and lock the bitmap page and if it is found, then lock on meta
139  * page is released, then finally acquire the lock on new overflow buffer.
140  * We need this locking order to avoid deadlock with backends that are
141  * doing inserts.
142  *
143  * Note: We could have avoided locking many buffers here if we made two
144  * WAL records for acquiring an overflow page (one to allocate an overflow
145  * page and another to add it to overflow bucket chain). However, doing
146  * so can leak an overflow page, if the system crashes after allocation.
147  * Needless to say, it is better to have a single record from a
148  * performance point of view as well.
149  */
151 
152  /* probably redundant... */
154 
155  /* loop to find current tail page, in case someone else inserted too */
156  for (;;)
157  {
158  BlockNumber nextblkno;
159 
160  page = BufferGetPage(buf);
161  pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
162  nextblkno = pageopaque->hasho_nextblkno;
163 
164  if (!BlockNumberIsValid(nextblkno))
165  break;
166 
167  /* we assume we do not need to write the unmodified page */
168  if (retain_pin)
169  {
170  /* pin will be retained only for the primary bucket page */
171  Assert((pageopaque->hasho_flag & LH_PAGE_TYPE) == LH_BUCKET_PAGE);
173  }
174  else
175  _hash_relbuf(rel, buf);
176 
177  retain_pin = false;
178 
179  buf = _hash_getbuf(rel, nextblkno, HASH_WRITE, LH_OVERFLOW_PAGE);
180  }
181 
182  /* Get exclusive lock on the meta page */
184 
185  _hash_checkpage(rel, metabuf, LH_META_PAGE);
186  metap = HashPageGetMeta(BufferGetPage(metabuf));
187 
188  /* start search at hashm_firstfree */
189  orig_firstfree = metap->hashm_firstfree;
190  first_page = orig_firstfree >> BMPG_SHIFT(metap);
191  bit = orig_firstfree & BMPG_MASK(metap);
192  i = first_page;
193  j = bit / BITS_PER_MAP;
194  bit &= ~(BITS_PER_MAP - 1);
195 
196  /* outer loop iterates once per bitmap page */
197  for (;;)
198  {
199  BlockNumber mapblkno;
200  Page mappage;
201  uint32 last_inpage;
202 
203  /* want to end search with the last existing overflow page */
204  splitnum = metap->hashm_ovflpoint;
205  max_ovflpg = metap->hashm_spares[splitnum] - 1;
206  last_page = max_ovflpg >> BMPG_SHIFT(metap);
207  last_bit = max_ovflpg & BMPG_MASK(metap);
208 
209  if (i > last_page)
210  break;
211 
212  Assert(i < metap->hashm_nmaps);
213  mapblkno = metap->hashm_mapp[i];
214 
215  if (i == last_page)
216  last_inpage = last_bit;
217  else
218  last_inpage = BMPGSZ_BIT(metap) - 1;
219 
220  /* Release exclusive lock on metapage while reading bitmap page */
221  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
222 
223  mapbuf = _hash_getbuf(rel, mapblkno, HASH_WRITE, LH_BITMAP_PAGE);
224  mappage = BufferGetPage(mapbuf);
225  freep = HashPageGetBitmap(mappage);
226 
227  for (; bit <= last_inpage; j++, bit += BITS_PER_MAP)
228  {
229  if (freep[j] != ALL_SET)
230  {
231  page_found = true;
232 
233  /* Reacquire exclusive lock on the meta page */
235 
236  /* convert bit to bit number within page */
237  bit += _hash_firstfreebit(freep[j]);
238  bitmap_page_bit = bit;
239 
240  /* convert bit to absolute bit number */
241  bit += (i << BMPG_SHIFT(metap));
242  /* Calculate address of the recycled overflow page */
243  blkno = bitno_to_blkno(metap, bit);
244 
245  /* Fetch and init the recycled page */
246  ovflbuf = _hash_getinitbuf(rel, blkno);
247 
248  goto found;
249  }
250  }
251 
252  /* No free space here, try to advance to next map page */
253  _hash_relbuf(rel, mapbuf);
254  mapbuf = InvalidBuffer;
255  i++;
256  j = 0; /* scan from start of next map page */
257  bit = 0;
258 
259  /* Reacquire exclusive lock on the meta page */
261  }
262 
263  /*
264  * No free pages --- have to extend the relation to add an overflow page.
265  * First, check to see if we have to add a new bitmap page too.
266  */
267  if (last_bit == (uint32) (BMPGSZ_BIT(metap) - 1))
268  {
269  /*
270  * We create the new bitmap page with all pages marked "in use".
271  * Actually two pages in the new bitmap's range will exist
272  * immediately: the bitmap page itself, and the following page which
273  * is the one we return to the caller. Both of these are correctly
274  * marked "in use". Subsequent pages do not exist yet, but it is
275  * convenient to pre-mark them as "in use" too.
276  */
277  bit = metap->hashm_spares[splitnum];
278 
279  /* metapage already has a write lock */
280  if (metap->hashm_nmaps >= HASH_MAX_BITMAPS)
281  ereport(ERROR,
282  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
283  errmsg("out of overflow pages in hash index \"%s\"",
284  RelationGetRelationName(rel))));
285 
286  newmapbuf = _hash_getnewbuf(rel, bitno_to_blkno(metap, bit), MAIN_FORKNUM);
287  }
288  else
289  {
290  /*
291  * Nothing to do here; since the page will be past the last used page,
292  * we know its bitmap bit was preinitialized to "in use".
293  */
294  }
295 
296  /* Calculate address of the new overflow page */
297  bit = BufferIsValid(newmapbuf) ?
298  metap->hashm_spares[splitnum] + 1 : metap->hashm_spares[splitnum];
299  blkno = bitno_to_blkno(metap, bit);
300 
301  /*
302  * Fetch the page with _hash_getnewbuf to ensure smgr's idea of the
303  * relation length stays in sync with ours. XXX It's annoying to do this
304  * with metapage write lock held; would be better to use a lock that
305  * doesn't block incoming searches.
306  *
307  * It is okay to hold two buffer locks here (one on tail page of bucket
308  * and other on new overflow page) since there cannot be anyone else
309  * contending for access to ovflbuf.
310  */
311  ovflbuf = _hash_getnewbuf(rel, blkno, MAIN_FORKNUM);
312 
313 found:
314 
315  /*
316  * Do the update. No ereport(ERROR) until changes are logged. We want to
317  * log the changes for bitmap page and overflow page together to avoid
318  * loss of pages in case the new page is added.
319  */
321 
322  if (page_found)
323  {
324  Assert(BufferIsValid(mapbuf));
325 
326  /* mark page "in use" in the bitmap */
327  SETBIT(freep, bitmap_page_bit);
328  MarkBufferDirty(mapbuf);
329  }
330  else
331  {
332  /* update the count to indicate new overflow page is added */
333  metap->hashm_spares[splitnum]++;
334 
335  if (BufferIsValid(newmapbuf))
336  {
337  _hash_initbitmapbuffer(newmapbuf, metap->hashm_bmsize, false);
338  MarkBufferDirty(newmapbuf);
339 
340  /* add the new bitmap page to the metapage's list of bitmaps */
341  metap->hashm_mapp[metap->hashm_nmaps] = BufferGetBlockNumber(newmapbuf);
342  metap->hashm_nmaps++;
343  metap->hashm_spares[splitnum]++;
344  }
345 
346  MarkBufferDirty(metabuf);
347 
348  /*
349  * for new overflow page, we don't need to explicitly set the bit in
350  * bitmap page, as by default that will be set to "in use".
351  */
352  }
353 
354  /*
355  * Adjust hashm_firstfree to avoid redundant searches. But don't risk
356  * changing it if someone moved it while we were searching bitmap pages.
357  */
358  if (metap->hashm_firstfree == orig_firstfree)
359  {
360  metap->hashm_firstfree = bit + 1;
361  MarkBufferDirty(metabuf);
362  }
363 
364  /* initialize new overflow page */
365  ovflpage = BufferGetPage(ovflbuf);
366  ovflopaque = (HashPageOpaque) PageGetSpecialPointer(ovflpage);
368  ovflopaque->hasho_nextblkno = InvalidBlockNumber;
369  ovflopaque->hasho_bucket = pageopaque->hasho_bucket;
370  ovflopaque->hasho_flag = LH_OVERFLOW_PAGE;
371  ovflopaque->hasho_page_id = HASHO_PAGE_ID;
372 
373  MarkBufferDirty(ovflbuf);
374 
375  /* logically chain overflow page to previous page */
376  pageopaque->hasho_nextblkno = BufferGetBlockNumber(ovflbuf);
377 
379 
380  /* XLOG stuff */
381  if (RelationNeedsWAL(rel))
382  {
383  XLogRecPtr recptr;
384  xl_hash_add_ovfl_page xlrec;
385 
386  xlrec.bmpage_found = page_found;
387  xlrec.bmsize = metap->hashm_bmsize;
388 
389  XLogBeginInsert();
390  XLogRegisterData((char *) &xlrec, SizeOfHashAddOvflPage);
391 
393  XLogRegisterBufData(0, (char *) &pageopaque->hasho_bucket, sizeof(Bucket));
394 
396 
397  if (BufferIsValid(mapbuf))
398  {
400  XLogRegisterBufData(2, (char *) &bitmap_page_bit, sizeof(uint32));
401  }
402 
403  if (BufferIsValid(newmapbuf))
404  XLogRegisterBuffer(3, newmapbuf, REGBUF_WILL_INIT);
405 
406  XLogRegisterBuffer(4, metabuf, REGBUF_STANDARD);
407  XLogRegisterBufData(4, (char *) &metap->hashm_firstfree, sizeof(uint32));
408 
409  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_ADD_OVFL_PAGE);
410 
411  PageSetLSN(BufferGetPage(ovflbuf), recptr);
412  PageSetLSN(BufferGetPage(buf), recptr);
413 
414  if (BufferIsValid(mapbuf))
415  PageSetLSN(BufferGetPage(mapbuf), recptr);
416 
417  if (BufferIsValid(newmapbuf))
418  PageSetLSN(BufferGetPage(newmapbuf), recptr);
419 
420  PageSetLSN(BufferGetPage(metabuf), recptr);
421  }
422 
424 
425  if (retain_pin)
427  else
428  _hash_relbuf(rel, buf);
429 
430  if (BufferIsValid(mapbuf))
431  _hash_relbuf(rel, mapbuf);
432 
433  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
434 
435  if (BufferIsValid(newmapbuf))
436  _hash_relbuf(rel, newmapbuf);
437 
438  return ovflbuf;
439 }
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 SETBIT(x, i)
Definition: blutils.c:33
#define HashPageGetBitmap(page)
Definition: hash.h:306
#define LH_BITMAP_PAGE
Definition: hash.h:66
static BlockNumber bitno_to_blkno(HashMetaPage metap, uint32 ovflbitnum)
Definition: hashovfl.c:34
#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
#define InvalidBuffer
Definition: buf.h:25
#define REGBUF_WILL_INIT
Definition: xloginsert.h:32
#define ALL_SET
Definition: hash.h:292
#define START_CRIT_SECTION()
Definition: miscadmin.h:131
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:207
Buffer _hash_getinitbuf(Relation rel, BlockNumber blkno)
Definition: hashpage.c:144
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:79
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:89
#define LH_PAGE_TYPE
Definition: hash.h:73
static uint32 _hash_firstfreebit(uint32 map)
Definition: hashovfl.c:447
uint32 Bucket
Definition: hash.h:34
BlockNumber hasho_prevblkno
Definition: hash.h:89
#define ERROR
Definition: elog.h:43
#define XLOG_HASH_ADD_OVFL_PAGE
Definition: hash_xlog.h:30
uint32 hashm_nmaps
Definition: hash.h:265
static char * buf
Definition: pg_test_fsync.c:67
#define HASH_WRITE
Definition: hash.h:330
#define BMPG_MASK(metap)
Definition: hash.h:304
#define REGBUF_STANDARD
Definition: xloginsert.h:34
#define RelationGetRelationName(relation)
Definition: rel.h:441
unsigned int uint32
Definition: c.h:325
#define BITS_PER_MAP
Definition: hash.h:319
#define BMPG_SHIFT(metap)
Definition: hash.h:303
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define ereport(elevel, rest)
Definition: elog.h:122
#define HASH_MAX_BITMAPS
Definition: hash.h:235
uint32 hashm_ovflpoint
Definition: hash.h:262
void XLogRegisterData(char *data, int len)
Definition: xloginsert.c:323
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:415
void _hash_checkpage(Relation rel, Buffer buf, int flags)
Definition: hashutil.c:225
#define LH_OVERFLOW_PAGE
Definition: hash.h:64
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
uint32 hashm_firstfree
Definition: hash.h:264
uint32 hashm_spares[HASH_MAX_SPLITPOINTS]
Definition: hash.h:267
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:275
void _hash_initbitmapbuffer(Buffer buf, uint16 bmsize, bool initpage)
Definition: hashovfl.c:740
#define BlockNumberIsValid(blockNumber)
Definition: block.h:70
#define LH_BUCKET_PAGE
Definition: hash.h:65
Datum bit(PG_FUNCTION_ARGS)
Definition: varbit.c:362
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:699
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 BufferIsValid(bufnum)
Definition: bufmgr.h:114
#define RelationNeedsWAL(relation)
Definition: rel.h:510
#define SizeOfHashAddOvflPage
Definition: hash_xlog.h:99
uint16 hasho_flag
Definition: hash.h:92
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:2605
#define HashPageGetMeta(page)
Definition: hash.h:313
int errmsg(const char *fmt,...)
Definition: elog.c:797
int i
BlockNumber hasho_nextblkno
Definition: hash.h:90
uint16 hashm_bmsize
Definition: hash.h:256
void XLogBeginInsert(void)
Definition: xloginsert.c:120
#define PageSetLSN(page, lsn)
Definition: bufpage.h:364
BlockNumber hashm_mapp[HASH_MAX_BITMAPS]
Definition: hash.h:269
int Buffer
Definition: buf.h:23
#define BMPGSZ_BIT(metap)
Definition: hash.h:302
Pointer Page
Definition: bufpage.h:74

◆ _hash_binsearch()

OffsetNumber _hash_binsearch ( Page  page,
uint32  hash_value 
)

Definition at line 358 of file hashutil.c.

References _hash_get_indextuple_hashkey(), Assert, FirstOffsetNumber, lower(), OffsetNumberIsValid, PageGetItem, PageGetItemId, PageGetMaxOffsetNumber, and upper().

Referenced by _hash_pgaddmultitup(), _hash_pgaddtup(), and _hash_readpage().

359 {
362 
363  /* Loop invariant: lower <= desired place <= upper */
364  upper = PageGetMaxOffsetNumber(page) + 1;
365  lower = FirstOffsetNumber;
366 
367  while (upper > lower)
368  {
369  OffsetNumber off;
370  IndexTuple itup;
371  uint32 hashkey;
372 
373  off = (upper + lower) / 2;
375 
376  itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, off));
377  hashkey = _hash_get_indextuple_hashkey(itup);
378  if (hashkey < hash_value)
379  lower = off + 1;
380  else
381  upper = off;
382  }
383 
384  return lower;
385 }
Datum lower(PG_FUNCTION_ARGS)
Definition: oracle_compat.c:44
Datum upper(PG_FUNCTION_ARGS)
Definition: oracle_compat.c:75
#define PageGetMaxOffsetNumber(page)
Definition: bufpage.h:353
uint16 OffsetNumber
Definition: off.h:24
uint32 _hash_get_indextuple_hashkey(IndexTuple itup)
Definition: hashutil.c:299
#define FirstOffsetNumber
Definition: off.h:27
IndexTupleData * IndexTuple
Definition: itup.h:53
unsigned int uint32
Definition: c.h:325
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:231
#define Assert(condition)
Definition: c.h:699
#define OffsetNumberIsValid(offsetNumber)
Definition: off.h:40
#define PageGetItem(page, itemId)
Definition: bufpage.h:336

◆ _hash_binsearch_last()

OffsetNumber _hash_binsearch_last ( Page  page,
uint32  hash_value 
)

Definition at line 396 of file hashutil.c.

References _hash_get_indextuple_hashkey(), Assert, FirstOffsetNumber, lower(), OffsetNumberIsValid, PageGetItem, PageGetItemId, PageGetMaxOffsetNumber, and upper().

Referenced by _hash_readpage().

397 {
400 
401  /* Loop invariant: lower <= desired place <= upper */
402  upper = PageGetMaxOffsetNumber(page);
403  lower = FirstOffsetNumber - 1;
404 
405  while (upper > lower)
406  {
407  IndexTuple itup;
408  OffsetNumber off;
409  uint32 hashkey;
410 
411  off = (upper + lower + 1) / 2;
413 
414  itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, off));
415  hashkey = _hash_get_indextuple_hashkey(itup);
416  if (hashkey > hash_value)
417  upper = off - 1;
418  else
419  lower = off;
420  }
421 
422  return lower;
423 }
Datum lower(PG_FUNCTION_ARGS)
Definition: oracle_compat.c:44
Datum upper(PG_FUNCTION_ARGS)
Definition: oracle_compat.c:75
#define PageGetMaxOffsetNumber(page)
Definition: bufpage.h:353
uint16 OffsetNumber
Definition: off.h:24
uint32 _hash_get_indextuple_hashkey(IndexTuple itup)
Definition: hashutil.c:299
#define FirstOffsetNumber
Definition: off.h:27
IndexTupleData * IndexTuple
Definition: itup.h:53
unsigned int uint32
Definition: c.h:325
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:231
#define Assert(condition)
Definition: c.h:699
#define OffsetNumberIsValid(offsetNumber)
Definition: off.h:40
#define PageGetItem(page, itemId)
Definition: bufpage.h:336

◆ _hash_checkpage()

void _hash_checkpage ( Relation  rel,
Buffer  buf,
int  flags 
)

Definition at line 225 of file hashutil.c.

References BufferGetBlockNumber(), BufferGetPage, ereport, errcode(), errhint(), errmsg(), ERROR, HASH_MAGIC, HASH_VERSION, HashMetaPageData::hashm_magic, HashMetaPageData::hashm_version, HashPageOpaqueData::hasho_flag, HashPageGetMeta, LH_META_PAGE, MAXALIGN, PageGetSpecialPointer, PageGetSpecialSize, PageIsNew, and RelationGetRelationName.

Referenced by _hash_addovflpage(), _hash_expandtable(), _hash_freeovflpage(), _hash_getbuf(), _hash_getbuf_with_condlock_cleanup(), _hash_getbuf_with_strategy(), _hash_pgaddmultitup(), _hash_pgaddtup(), _hash_readpage(), and hashbulkdelete().

226 {
227  Page page = BufferGetPage(buf);
228 
229  /*
230  * ReadBuffer verifies that every newly-read page passes
231  * PageHeaderIsValid, which means it either contains a reasonably sane
232  * page header or is all-zero. We have to defend against the all-zero
233  * case, however.
234  */
235  if (PageIsNew(page))
236  ereport(ERROR,
237  (errcode(ERRCODE_INDEX_CORRUPTED),
238  errmsg("index \"%s\" contains unexpected zero page at block %u",
241  errhint("Please REINDEX it.")));
242 
243  /*
244  * Additionally check that the special area looks sane.
245  */
246  if (PageGetSpecialSize(page) != MAXALIGN(sizeof(HashPageOpaqueData)))
247  ereport(ERROR,
248  (errcode(ERRCODE_INDEX_CORRUPTED),
249  errmsg("index \"%s\" contains corrupted page at block %u",
252  errhint("Please REINDEX it.")));
253 
254  if (flags)
255  {
257 
258  if ((opaque->hasho_flag & flags) == 0)
259  ereport(ERROR,
260  (errcode(ERRCODE_INDEX_CORRUPTED),
261  errmsg("index \"%s\" contains corrupted page at block %u",
264  errhint("Please REINDEX it.")));
265  }
266 
267  /*
268  * When checking the metapage, also verify magic number and version.
269  */
270  if (flags == LH_META_PAGE)
271  {
272  HashMetaPage metap = HashPageGetMeta(page);
273 
274  if (metap->hashm_magic != HASH_MAGIC)
275  ereport(ERROR,
276  (errcode(ERRCODE_INDEX_CORRUPTED),
277  errmsg("index \"%s\" is not a hash index",
278  RelationGetRelationName(rel))));
279 
280  if (metap->hashm_version != HASH_VERSION)
281  ereport(ERROR,
282  (errcode(ERRCODE_INDEX_CORRUPTED),
283  errmsg("index \"%s\" has wrong hash version",
285  errhint("Please REINDEX it.")));
286  }
287 }
int errhint(const char *fmt,...)
Definition: elog.c:987
#define LH_META_PAGE
Definition: hash.h:67
uint32 hashm_magic
Definition: hash.h:251
int errcode(int sqlerrcode)
Definition: elog.c:575
#define HASH_VERSION
Definition: hash.h:209
#define HASH_MAGIC
Definition: hash.h:208
#define ERROR
Definition: elog.h:43
uint32 hashm_version
Definition: hash.h:252
static char * buf
Definition: pg_test_fsync.c:67
#define RelationGetRelationName(relation)
Definition: rel.h:441
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define ereport(elevel, rest)
Definition: elog.h:122
#define PageGetSpecialPointer(page)
Definition: bufpage.h:322
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:96
#define MAXALIGN(LEN)
Definition: c.h:652
#define PageGetSpecialSize(page)
Definition: bufpage.h:296
uint16 hasho_flag
Definition: hash.h:92
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:2605
#define PageIsNew(page)
Definition: bufpage.h:225
#define HashPageGetMeta(page)
Definition: hash.h:313
int errmsg(const char *fmt,...)
Definition: elog.c:797
Pointer Page
Definition: bufpage.h:74

◆ _hash_checkqual()

bool _hash_checkqual ( IndexScanDesc  scan,
IndexTuple  itup 
)

Definition at line 31 of file hashutil.c.

References DatumGetBool, FunctionCall2Coll(), index_getattr, IndexScanDescData::indexRelation, IndexScanDescData::keyData, IndexScanDescData::numberOfKeys, RelationGetDescr, ScanKeyData::sk_argument, ScanKeyData::sk_attno, ScanKeyData::sk_collation, ScanKeyData::sk_flags, ScanKeyData::sk_func, SK_ISNULL, and test().

Referenced by _hash_load_qualified_items().

32 {
33  /*
34  * Currently, we can't check any of the scan conditions since we do not
35  * have the original index entry value to supply to the sk_func. Always
36  * return true; we expect that hashgettuple already set the recheck flag
37  * to make the main indexscan code do it.
38  */
39 #ifdef NOT_USED
40  TupleDesc tupdesc = RelationGetDescr(scan->indexRelation);
41  ScanKey key = scan->keyData;
42  int scanKeySize = scan->numberOfKeys;
43 
44  while (scanKeySize > 0)
45  {
46  Datum datum;
47  bool isNull;
48  Datum test;
49 
50  datum = index_getattr(itup,
51  key->sk_attno,
52  tupdesc,
53  &isNull);
54 
55  /* assume sk_func is strict */
56  if (isNull)
57  return false;
58  if (key->sk_flags & SK_ISNULL)
59  return false;
60 
61  test = FunctionCall2Coll(&key->sk_func, key->sk_collation,
62  datum, key->sk_argument);
63 
64  if (!DatumGetBool(test))
65  return false;
66 
67  key++;
68  scanKeySize--;
69  }
70 #endif
71 
72  return true;
73 }
static void test(void)
#define RelationGetDescr(relation)
Definition: rel.h:433
Datum FunctionCall2Coll(FmgrInfo *flinfo, Oid collation, Datum arg1, Datum arg2)
Definition: fmgr.c:1133
Relation indexRelation
Definition: relscan.h:91
FmgrInfo sk_func
Definition: skey.h:71
#define DatumGetBool(X)
Definition: postgres.h:378
#define SK_ISNULL
Definition: skey.h:115
uintptr_t Datum
Definition: postgres.h:367
int sk_flags
Definition: skey.h:66
#define index_getattr(tup, attnum, tupleDesc, isnull)
Definition: itup.h:100
ScanKey keyData
Definition: relscan.h:95
Oid sk_collation
Definition: skey.h:70
Datum sk_argument
Definition: skey.h:72
AttrNumber sk_attno
Definition: skey.h:67

◆ _hash_convert_tuple()

bool _hash_convert_tuple ( Relation  index,
Datum user_values,
bool user_isnull,
Datum index_values,
bool index_isnull 
)

Definition at line 326 of file hashutil.c.

References _hash_datum2hashkey(), and UInt32GetDatum.

Referenced by hashbuildCallback(), and hashinsert().

329 {
330  uint32 hashkey;
331 
332  /*
333  * We do not insert null values into hash indexes. This is okay because
334  * the only supported search operator is '=', and we assume it is strict.
335  */
336  if (user_isnull[0])
337  return false;
338 
339  hashkey = _hash_datum2hashkey(index, user_values[0]);
340  index_values[0] = UInt32GetDatum(hashkey);
341  index_isnull[0] = false;
342  return true;
343 }
uint32 _hash_datum2hashkey(Relation rel, Datum key)
Definition: hashutil.c:82
unsigned int uint32
Definition: c.h:325
#define UInt32GetDatum(X)
Definition: postgres.h:478

◆ _hash_datum2hashkey()

uint32 _hash_datum2hashkey ( Relation  rel,
Datum  key 
)

Definition at line 82 of file hashutil.c.

References DatumGetUInt32, FunctionCall1Coll(), HASHSTANDARD_PROC, index_getprocinfo(), and RelationData::rd_indcollation.

Referenced by _hash_convert_tuple(), and _hash_first().

83 {
84  FmgrInfo *procinfo;
85  Oid collation;
86 
87  /* XXX assumes index has only one attribute */
88  procinfo = index_getprocinfo(rel, 1, HASHSTANDARD_PROC);
89  collation = rel->rd_indcollation[0];
90 
91  return DatumGetUInt32(FunctionCall1Coll(procinfo, collation, key));
92 }
#define DatumGetUInt32(X)
Definition: postgres.h:471
Definition: fmgr.h:56
FmgrInfo * index_getprocinfo(Relation irel, AttrNumber attnum, uint16 procnum)
Definition: indexam.c:855
unsigned int Oid
Definition: postgres_ext.h:31
Oid * rd_indcollation
Definition: rel.h:165
#define HASHSTANDARD_PROC
Definition: hash.h:351
Datum FunctionCall1Coll(FmgrInfo *flinfo, Oid collation, Datum arg1)
Definition: fmgr.c:1113

◆ _hash_datum2hashkey_type()

uint32 _hash_datum2hashkey_type ( Relation  rel,
Datum  key,
Oid  keytype 
)

Definition at line 102 of file hashutil.c.

References DatumGetUInt32, elog, ERROR, get_opfamily_proc(), HASHSTANDARD_PROC, OidFunctionCall1Coll(), RelationData::rd_indcollation, RelationData::rd_opfamily, RegProcedureIsValid, and RelationGetRelationName.

Referenced by _hash_first().

103 {
104  RegProcedure hash_proc;
105  Oid collation;
106 
107  /* XXX assumes index has only one attribute */
108  hash_proc = get_opfamily_proc(rel->rd_opfamily[0],
109  keytype,
110  keytype,
112  if (!RegProcedureIsValid(hash_proc))
113  elog(ERROR, "missing support function %d(%u,%u) for index \"%s\"",
114  HASHSTANDARD_PROC, keytype, keytype,
116  collation = rel->rd_indcollation[0];
117 
118  return DatumGetUInt32(OidFunctionCall1Coll(hash_proc, collation, key));
119 }
#define DatumGetUInt32(X)
Definition: postgres.h:471
regproc RegProcedure
Definition: c.h:472
unsigned int Oid
Definition: postgres_ext.h:31
Oid * rd_indcollation
Definition: rel.h:165
#define ERROR
Definition: elog.h:43
#define RegProcedureIsValid(p)
Definition: c.h:607
#define RelationGetRelationName(relation)
Definition: rel.h:441
Oid * rd_opfamily
Definition: rel.h:154
#define HASHSTANDARD_PROC
Definition: hash.h:351
Datum OidFunctionCall1Coll(Oid functionId, Oid collation, Datum arg1)
Definition: fmgr.c:1405
Oid get_opfamily_proc(Oid opfamily, Oid lefttype, Oid righttype, int16 procnum)
Definition: lsyscache.c:744
#define elog
Definition: elog.h:219

◆ _hash_doinsert()

void _hash_doinsert ( Relation  rel,
IndexTuple  itup,
Relation  heapRel 
)

Definition at line 37 of file hashinsert.c.

References _hash_addovflpage(), _hash_dropbuf(), _hash_expandtable(), _hash_finish_split(), _hash_get_indextuple_hashkey(), _hash_getbucketbuf_from_hashkey(), _hash_getbuf(), _hash_pgaddtup(), _hash_relbuf(), _hash_vacuum_one_page(), Assert, BlockNumberIsValid, buf, BUFFER_LOCK_EXCLUSIVE, BUFFER_LOCK_UNLOCK, BufferGetPage, CheckForSerializableConflictIn(), END_CRIT_SECTION, ereport, errcode(), errhint(), errmsg(), ERROR, H_BUCKET_BEING_SPLIT, H_HAS_DEAD_TUPLES, HASH_METAPAGE, HASH_NOLOCK, HASH_WRITE, HashMetaPageData::hashm_ffactor, HashMetaPageData::hashm_highmask, HashMetaPageData::hashm_lowmask, HashMetaPageData::hashm_maxbucket, HashMetaPageData::hashm_ntuples, HashMaxItemSize, HashPageOpaqueData::hasho_bucket, HashPageOpaqueData::hasho_flag, HashPageOpaqueData::hasho_nextblkno, HashPageGetMeta, IndexTupleSize, InvalidBuffer, IsBufferCleanupOK(), LH_META_PAGE, LH_OVERFLOW_PAGE, LH_PAGE_TYPE, LockBuffer(), MarkBufferDirty(), MAXALIGN, xl_hash_insert::offnum, PageGetFreeSpace(), PageGetSpecialPointer, PageSetLSN, RelationData::rd_node, REGBUF_STANDARD, RelationNeedsWAL, SizeOfHashInsert, START_CRIT_SECTION, XLOG_HASH_INSERT, XLogBeginInsert(), XLogInsert(), XLogRegisterBufData(), XLogRegisterBuffer(), and XLogRegisterData().

Referenced by _h_indexbuild(), hashbuildCallback(), and hashinsert().

38 {
40  Buffer bucket_buf;
41  Buffer metabuf;
42  HashMetaPage metap;
43  HashMetaPage usedmetap = NULL;
44  Page metapage;
45  Page page;
46  HashPageOpaque pageopaque;
47  Size itemsz;
48  bool do_expand;
49  uint32 hashkey;
50  Bucket bucket;
51  OffsetNumber itup_off;
52 
53  /*
54  * Get the hash key for the item (it's stored in the index tuple itself).
55  */
56  hashkey = _hash_get_indextuple_hashkey(itup);
57 
58  /* compute item size too */
59  itemsz = IndexTupleSize(itup);
60  itemsz = MAXALIGN(itemsz); /* be safe, PageAddItem will do this but we
61  * need to be consistent */
62 
63 restart_insert:
64 
65  /*
66  * Read the metapage. We don't lock it yet; HashMaxItemSize() will
67  * examine pd_pagesize_version, but that can't change so we can examine it
68  * without a lock.
69  */
71  metapage = BufferGetPage(metabuf);
72 
73  /*
74  * Check whether the item can fit on a hash page at all. (Eventually, we
75  * ought to try to apply TOAST methods if not.) Note that at this point,
76  * itemsz doesn't include the ItemId.
77  *
78  * XXX this is useless code if we are only storing hash keys.
79  */
80  if (itemsz > HashMaxItemSize(metapage))
81  ereport(ERROR,
82  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
83  errmsg("index row size %zu exceeds hash maximum %zu",
84  itemsz, HashMaxItemSize(metapage)),
85  errhint("Values larger than a buffer page cannot be indexed.")));
86 
87  /* Lock the primary bucket page for the target bucket. */
88  buf = _hash_getbucketbuf_from_hashkey(rel, hashkey, HASH_WRITE,
89  &usedmetap);
90  Assert(usedmetap != NULL);
91 
92  CheckForSerializableConflictIn(rel, NULL, buf);
93 
94  /* remember the primary bucket buffer to release the pin on it at end. */
95  bucket_buf = buf;
96 
97  page = BufferGetPage(buf);
98  pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
99  bucket = pageopaque->hasho_bucket;
100 
101  /*
102  * If this bucket is in the process of being split, try to finish the
103  * split before inserting, because that might create room for the
104  * insertion to proceed without allocating an additional overflow page.
105  * It's only interesting to finish the split if we're trying to insert
106  * into the bucket from which we're removing tuples (the "old" bucket),
107  * not if we're trying to insert into the bucket into which tuples are
108  * being moved (the "new" bucket).
109  */
110  if (H_BUCKET_BEING_SPLIT(pageopaque) && IsBufferCleanupOK(buf))
111  {
112  /* release the lock on bucket buffer, before completing the split. */
114 
115  _hash_finish_split(rel, metabuf, buf, bucket,
116  usedmetap->hashm_maxbucket,
117  usedmetap->hashm_highmask,
118  usedmetap->hashm_lowmask);
119 
120  /* release the pin on old and meta buffer. retry for insert. */
121  _hash_dropbuf(rel, buf);
122  _hash_dropbuf(rel, metabuf);
123  goto restart_insert;
124  }
125 
126  /* Do the insertion */
127  while (PageGetFreeSpace(page) < itemsz)
128  {
129  BlockNumber nextblkno;
130 
131  /*
132  * Check if current page has any DEAD tuples. If yes, delete these
133  * tuples and see if we can get a space for the new item to be
134  * inserted before moving to the next page in the bucket chain.
135  */
136  if (H_HAS_DEAD_TUPLES(pageopaque))
137  {
138 
139  if (IsBufferCleanupOK(buf))
140  {
141  _hash_vacuum_one_page(rel, metabuf, buf, heapRel->rd_node);
142 
143  if (PageGetFreeSpace(page) >= itemsz)
144  break; /* OK, now we have enough space */
145  }
146  }
147 
148  /*
149  * no space on this page; check for an overflow page
150  */
151  nextblkno = pageopaque->hasho_nextblkno;
152 
153  if (BlockNumberIsValid(nextblkno))
154  {
155  /*
156  * ovfl page exists; go get it. if it doesn't have room, we'll
157  * find out next pass through the loop test above. we always
158  * release both the lock and pin if this is an overflow page, but
159  * only the lock if this is the primary bucket page, since the pin
160  * on the primary bucket must be retained throughout the scan.
161  */
162  if (buf != bucket_buf)
163  _hash_relbuf(rel, buf);
164  else
166  buf = _hash_getbuf(rel, nextblkno, HASH_WRITE, LH_OVERFLOW_PAGE);
167  page = BufferGetPage(buf);
168  }
169  else
170  {
171  /*
172  * we're at the end of the bucket chain and we haven't found a
173  * page with enough room. allocate a new overflow page.
174  */
175 
176  /* release our write lock without modifying buffer */
178 
179  /* chain to a new overflow page */
180  buf = _hash_addovflpage(rel, metabuf, buf, (buf == bucket_buf) ? true : false);
181  page = BufferGetPage(buf);
182 
183  /* should fit now, given test above */
184  Assert(PageGetFreeSpace(page) >= itemsz);
185  }
186  pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
187  Assert((pageopaque->hasho_flag & LH_PAGE_TYPE) == LH_OVERFLOW_PAGE);
188  Assert(pageopaque->hasho_bucket == bucket);
189  }
190 
191  /*
192  * Write-lock the metapage so we can increment the tuple count. After
193  * incrementing it, check to see if it's time for a split.
194  */
196 
197  /* Do the update. No ereport(ERROR) until changes are logged */
199 
200  /* found page with enough space, so add the item here */
201  itup_off = _hash_pgaddtup(rel, buf, itemsz, itup);
202  MarkBufferDirty(buf);
203 
204  /* metapage operations */
205  metap = HashPageGetMeta(metapage);
206  metap->hashm_ntuples += 1;
207 
208  /* Make sure this stays in sync with _hash_expandtable() */
209  do_expand = metap->hashm_ntuples >
210  (double) metap->hashm_ffactor * (metap->hashm_maxbucket + 1);
211 
212  MarkBufferDirty(metabuf);
213 
214  /* XLOG stuff */
215  if (RelationNeedsWAL(rel))
216  {
217  xl_hash_insert xlrec;
218  XLogRecPtr recptr;
219 
220  xlrec.offnum = itup_off;
221 
222  XLogBeginInsert();
223  XLogRegisterData((char *) &xlrec, SizeOfHashInsert);
224 
225  XLogRegisterBuffer(1, metabuf, REGBUF_STANDARD);
226 
228  XLogRegisterBufData(0, (char *) itup, IndexTupleSize(itup));
229 
230  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_INSERT);
231 
232  PageSetLSN(BufferGetPage(buf), recptr);
233  PageSetLSN(BufferGetPage(metabuf), recptr);
234  }
235 
237 
238  /* drop lock on metapage, but keep pin */
239  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
240 
241  /*
242  * Release the modified page and ensure to release the pin on primary
243  * page.
244  */
245  _hash_relbuf(rel, buf);
246  if (buf != bucket_buf)
247  _hash_dropbuf(rel, bucket_buf);
248 
249  /* Attempt to split if a split is needed */
250  if (do_expand)
251  _hash_expandtable(rel, metabuf);
252 
253  /* Finally drop our pin on the metapage */
254  _hash_dropbuf(rel, metabuf);
255 }
void XLogRegisterBufData(uint8 block_id, char *data, int len)
Definition: xloginsert.c:361
#define HashMaxItemSize(page)
Definition: hash.h:277
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:87
#define XLOG_HASH_INSERT
Definition: hash_xlog.h:29
int errhint(const char *fmt,...)
Definition: elog.c:987
#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:254
uint32 hashm_highmask
Definition: hash.h:260
#define InvalidBuffer
Definition: buf.h:25
#define START_CRIT_SECTION()
Definition: miscadmin.h:131
int errcode(int sqlerrcode)
Definition: elog.c:575
Buffer _hash_getbucketbuf_from_hashkey(Relation rel, uint32 hashkey, int access, HashMetaPage *cachedmetap)
Definition: hashpage.c:1567
uint32 BlockNumber
Definition: block.h:31
void _hash_dropbuf(Relation rel, Buffer buf)
Definition: hashpage.c:286
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:79
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:89
uint32 hashm_lowmask
Definition: hash.h:261
void CheckForSerializableConflictIn(Relation relation, HeapTuple tuple, Buffer buffer)
Definition: predicate.c:4280
Size PageGetFreeSpace(Page page)
Definition: bufpage.c:578
uint16 OffsetNumber
Definition: off.h:24
#define LH_PAGE_TYPE
Definition: hash.h:73
uint32 Bucket
Definition: hash.h:34
#define SizeOfHashInsert
Definition: hash_xlog.h:80
#define ERROR
Definition: elog.h:43
uint32 _hash_get_indextuple_hashkey(IndexTuple itup)
Definition: hashutil.c:299
void _hash_finish_split(Relation rel, Buffer metabuf, Buffer obuf, Bucket obucket, uint32 maxbucket, uint32 highmask, uint32 lowmask)
Definition: hashpage.c:1363
static char * buf
Definition: pg_test_fsync.c:67
#define HASH_WRITE
Definition: hash.h:330
#define HASH_NOLOCK
Definition: hash.h:331
#define REGBUF_STANDARD
Definition: xloginsert.h:34
unsigned int uint32
Definition: c.h:325
void _hash_expandtable(Relation rel, Buffer metabuf)
Definition: hashpage.c:626
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define ereport(elevel, rest)
Definition: elog.h:122
bool IsBufferCleanupOK(Buffer buffer)
Definition: bufmgr.c:3774
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:206
double hashm_ntuples
Definition: hash.h:253
#define LH_OVERFLOW_PAGE
Definition: hash.h:64
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:275
#define BlockNumberIsValid(blockNumber)
Definition: block.h:70
#define H_BUCKET_BEING_SPLIT(opaque)
Definition: hash.h:99
RelFileNode rd_node
Definition: rel.h:55
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:699
Bucket hasho_bucket
Definition: hash.h:91
size_t Size
Definition: c.h:433
#define PageGetSpecialPointer(page)
Definition: bufpage.h:322
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:96
OffsetNumber offnum
Definition: hash_xlog.h:77
Buffer _hash_addovflpage(Relation rel, Buffer metabuf, Buffer buf, bool retain_pin)
Definition: hashovfl.c:111
#define MAXALIGN(LEN)
Definition: c.h:652
#define RelationNeedsWAL(relation)
Definition: rel.h:510
uint32 hashm_maxbucket
Definition: hash.h:259
OffsetNumber _hash_pgaddtup(Relation rel, Buffer buf, Size itemsize, IndexTuple itup)
Definition: hashinsert.c:269
static void _hash_vacuum_one_page(Relation rel, Buffer metabuf, Buffer buf, RelFileNode hnode)
Definition: hashinsert.c:339
uint16 hasho_flag
Definition: hash.h:92
#define HashPageGetMeta(page)
Definition: hash.h:313
int errmsg(const char *fmt,...)
Definition: elog.c:797
BlockNumber hasho_nextblkno
Definition: hash.h:90
void XLogBeginInsert(void)
Definition: xloginsert.c:120
#define PageSetLSN(page, lsn)
Definition: bufpage.h:364
int Buffer
Definition: buf.h:23
#define H_HAS_DEAD_TUPLES(opaque)
Definition: hash.h:101
Pointer Page
Definition: bufpage.h:74
#define IndexTupleSize(itup)
Definition: itup.h:71

◆ _hash_dropbuf()

void _hash_dropbuf ( Relation  rel,
Buffer  buf 
)

Definition at line 286 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().

287 {
289 }
void ReleaseBuffer(Buffer buffer)
Definition: bufmgr.c:3309
static char * buf
Definition: pg_test_fsync.c:67

◆ _hash_dropscanbuf()

void _hash_dropscanbuf ( Relation  rel,
HashScanOpaque  so 
)

Definition at line 298 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().

299 {
300  /* release pin we hold on primary bucket page */
301  if (BufferIsValid(so->hashso_bucket_buf) &&
302  so->hashso_bucket_buf != so->currPos.buf)
305 
306  /* release pin we hold on primary bucket page of bucket being split */
311 
312  /* release any pin we still hold */
313  if (BufferIsValid(so->currPos.buf))
314  _hash_dropbuf(rel, so->currPos.buf);
315  so->currPos.buf = InvalidBuffer;
316 
317  /* reset split scan */
318  so->hashso_buc_populated = false;
319  so->hashso_buc_split = false;
320 }
#define InvalidBuffer
Definition: buf.h:25
void _hash_dropbuf(Relation rel, Buffer buf)
Definition: hashpage.c:286
Buffer hashso_bucket_buf
Definition: hash.h:172
bool hashso_buc_populated
Definition: hash.h:182
Buffer buf
Definition: hash.h:119
#define BufferIsValid(bufnum)
Definition: bufmgr.h:114
HashScanPosData currPos
Definition: hash.h:197
bool hashso_buc_split
Definition: hash.h:188
Buffer hashso_split_bucket_buf
Definition: hash.h:179

◆ _hash_expandtable()

void _hash_expandtable ( Relation  rel,
Buffer  metabuf 
)

Definition at line 626 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().

627 {
628  HashMetaPage metap;
629  Bucket old_bucket;
630  Bucket new_bucket;
631  uint32 spare_ndx;
632  BlockNumber start_oblkno;
633  BlockNumber start_nblkno;
634  Buffer buf_nblkno;
635  Buffer buf_oblkno;
636  Page opage;
637  Page npage;
638  HashPageOpaque oopaque;
639  HashPageOpaque nopaque;
640  uint32 maxbucket;
641  uint32 highmask;
642  uint32 lowmask;
643  bool metap_update_masks = false;
644  bool metap_update_splitpoint = false;
645 
646 restart_expand:
647 
648  /*
649  * Write-lock the meta page. It used to be necessary to acquire a
650  * heavyweight lock to begin a split, but that is no longer required.
651  */
653 
654  _hash_checkpage(rel, metabuf, LH_META_PAGE);
655  metap = HashPageGetMeta(BufferGetPage(metabuf));
656 
657  /*
658  * Check to see if split is still needed; someone else might have already
659  * done one while we waited for the lock.
660  *
661  * Make sure this stays in sync with _hash_doinsert()
662  */
663  if (metap->hashm_ntuples <=
664  (double) metap->hashm_ffactor * (metap->hashm_maxbucket + 1))
665  goto fail;
666 
667  /*
668  * Can't split anymore if maxbucket has reached its maximum possible
669  * value.
670  *
671  * Ideally we'd allow bucket numbers up to UINT_MAX-1 (no higher because
672  * the calculation maxbucket+1 mustn't overflow). Currently we restrict
673  * to half that because of overflow looping in _hash_log2() and
674  * insufficient space in hashm_spares[]. It's moot anyway because an
675  * index with 2^32 buckets would certainly overflow BlockNumber and hence
676  * _hash_alloc_buckets() would fail, but if we supported buckets smaller
677  * than a disk block then this would be an independent constraint.
678  *
679  * If you change this, see also the maximum initial number of buckets in
680  * _hash_init().
681  */
682  if (metap->hashm_maxbucket >= (uint32) 0x7FFFFFFE)
683  goto fail;
684 
685  /*
686  * Determine which bucket is to be split, and attempt to take cleanup lock
687  * on the old bucket. If we can't get the lock, give up.
688  *
689  * The cleanup lock protects us not only against other backends, but
690  * against our own backend as well.
691  *
692  * The cleanup lock is mainly to protect the split from concurrent
693  * inserts. See src/backend/access/hash/README, Lock Definitions for
694  * further details. Due to this locking restriction, if there is any
695  * pending scan, the split will give up which is not good, but harmless.
696  */
697  new_bucket = metap->hashm_maxbucket + 1;
698 
699  old_bucket = (new_bucket & metap->hashm_lowmask);
700 
701  start_oblkno = BUCKET_TO_BLKNO(metap, old_bucket);
702 
703  buf_oblkno = _hash_getbuf_with_condlock_cleanup(rel, start_oblkno, LH_BUCKET_PAGE);
704  if (!buf_oblkno)
705  goto fail;
706 
707  opage = BufferGetPage(buf_oblkno);
708  oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
709 
710  /*
711  * We want to finish the split from a bucket as there is no apparent
712  * benefit by not doing so and it will make the code complicated to finish
713  * the split that involves multiple buckets considering the case where new
714  * split also fails. We don't need to consider the new bucket for
715  * completing the split here as it is not possible that a re-split of new
716  * bucket starts when there is still a pending split from old bucket.
717  */
718  if (H_BUCKET_BEING_SPLIT(oopaque))
719  {
720  /*
721  * Copy bucket mapping info now; refer the comment in code below where
722  * we copy this information before calling _hash_splitbucket to see
723  * why this is okay.
724  */
725  maxbucket = metap->hashm_maxbucket;
726  highmask = metap->hashm_highmask;
727  lowmask = metap->hashm_lowmask;
728 
729  /*
730  * Release the lock on metapage and old_bucket, before completing the
731  * split.
732  */
733  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
734  LockBuffer(buf_oblkno, BUFFER_LOCK_UNLOCK);
735 
736  _hash_finish_split(rel, metabuf, buf_oblkno, old_bucket, maxbucket,
737  highmask, lowmask);
738 
739  /* release the pin on old buffer and retry for expand. */
740  _hash_dropbuf(rel, buf_oblkno);
741 
742  goto restart_expand;
743  }
744 
745  /*
746  * Clean the tuples remained from the previous split. This operation
747  * requires cleanup lock and we already have one on the old bucket, so
748  * let's do it. We also don't want to allow further splits from the bucket
749  * till the garbage of previous split is cleaned. This has two
750  * advantages; first, it helps in avoiding the bloat due to garbage and
751  * second is, during cleanup of bucket, we are always sure that the
752  * garbage tuples belong to most recently split bucket. On the contrary,
753  * if we allow cleanup of bucket after meta page is updated to indicate
754  * the new split and before the actual split, the cleanup operation won't
755  * be able to decide whether the tuple has been moved to the newly created
756  * bucket and ended up deleting such tuples.
757  */
758  if (H_NEEDS_SPLIT_CLEANUP(oopaque))
759  {
760  /*
761  * Copy bucket mapping info now; refer to the comment in code below
762  * where we copy this information before calling _hash_splitbucket to
763  * see why this is okay.
764  */
765  maxbucket = metap->hashm_maxbucket;
766  highmask = metap->hashm_highmask;
767  lowmask = metap->hashm_lowmask;
768 
769  /* Release the metapage lock. */
770  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
771 
772  hashbucketcleanup(rel, old_bucket, buf_oblkno, start_oblkno, NULL,
773  maxbucket, highmask, lowmask, NULL, NULL, true,
774  NULL, NULL);
775 
776  _hash_dropbuf(rel, buf_oblkno);
777 
778  goto restart_expand;
779  }
780 
781  /*
782  * There shouldn't be any active scan on new bucket.
783  *
784  * Note: it is safe to compute the new bucket's blkno here, even though we
785  * may still need to update the BUCKET_TO_BLKNO mapping. This is because
786  * the current value of hashm_spares[hashm_ovflpoint] correctly shows
787  * where we are going to put a new splitpoint's worth of buckets.
788  */
789  start_nblkno = BUCKET_TO_BLKNO(metap, new_bucket);
790 
791  /*
792  * If the split point is increasing we need to allocate a new batch of
793  * bucket pages.
794  */
795  spare_ndx = _hash_spareindex(new_bucket + 1);
796  if (spare_ndx > metap->hashm_ovflpoint)
797  {
798  uint32 buckets_to_add;
799 
800  Assert(spare_ndx == metap->hashm_ovflpoint + 1);
801 
802  /*
803  * We treat allocation of buckets as a separate WAL-logged action.
804  * Even if we fail after this operation, won't leak bucket pages;
805  * rather, the next split will consume this space. In any case, even
806  * without failure we don't use all the space in one split operation.
807  */
808  buckets_to_add = _hash_get_totalbuckets(spare_ndx) - new_bucket;
809  if (!_hash_alloc_buckets(rel, start_nblkno, buckets_to_add))
810  {
811  /* can't split due to BlockNumber overflow */
812  _hash_relbuf(rel, buf_oblkno);
813  goto fail;
814  }
815  }
816 
817  /*
818  * Physically allocate the new bucket's primary page. We want to do this
819  * before changing the metapage's mapping info, in case we can't get the
820  * disk space. Ideally, we don't need to check for cleanup lock on new
821  * bucket as no other backend could find this bucket unless meta page is
822  * updated. However, it is good to be consistent with old bucket locking.
823  */
824  buf_nblkno = _hash_getnewbuf(rel, start_nblkno, MAIN_FORKNUM);
825  if (!IsBufferCleanupOK(buf_nblkno))
826  {
827  _hash_relbuf(rel, buf_oblkno);
828  _hash_relbuf(rel, buf_nblkno);
829  goto fail;
830  }
831 
832  /*
833  * Since we are scribbling on the pages in the shared buffers, establish a
834  * critical section. Any failure in this next code leaves us with a big
835  * problem: the metapage is effectively corrupt but could get written back
836  * to disk.
837  */
839 
840  /*
841  * Okay to proceed with split. Update the metapage bucket mapping info.
842  */
843  metap->hashm_maxbucket = new_bucket;
844 
845  if (new_bucket > metap->hashm_highmask)
846  {
847  /* Starting a new doubling */
848  metap->hashm_lowmask = metap->hashm_highmask;
849  metap->hashm_highmask = new_bucket | metap->hashm_lowmask;
850  metap_update_masks = true;
851  }
852 
853  /*
854  * If the split point is increasing we need to adjust the hashm_spares[]
855  * array and hashm_ovflpoint so that future overflow pages will be created
856  * beyond this new batch of bucket pages.
857  */
858  if (spare_ndx > metap->hashm_ovflpoint)
859  {
860  metap->hashm_spares[spare_ndx] = metap->hashm_spares[metap->hashm_ovflpoint];
861  metap->hashm_ovflpoint = spare_ndx;
862  metap_update_splitpoint = true;
863  }
864 
865  MarkBufferDirty(metabuf);
866 
867  /*
868  * Copy bucket mapping info now; this saves re-accessing the meta page
869  * inside _hash_splitbucket's inner loop. Note that once we drop the
870  * split lock, other splits could begin, so these values might be out of
871  * date before _hash_splitbucket finishes. That's okay, since all it
872  * needs is to tell which of these two buckets to map hashkeys into.
873  */
874  maxbucket = metap->hashm_maxbucket;
875  highmask = metap->hashm_highmask;
876  lowmask = metap->hashm_lowmask;
877 
878  opage = BufferGetPage(buf_oblkno);
879  oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
880 
881  /*
882  * Mark the old bucket to indicate that split is in progress. (At
883  * operation end, we will clear the split-in-progress flag.) Also, for a
884  * primary bucket page, hasho_prevblkno stores the number of buckets that
885  * existed as of the last split, so we must update that value here.
886  */
887  oopaque->hasho_flag |= LH_BUCKET_BEING_SPLIT;
888  oopaque->hasho_prevblkno = maxbucket;
889 
890  MarkBufferDirty(buf_oblkno);
891 
892  npage = BufferGetPage(buf_nblkno);
893 
894  /*
895  * initialize the new bucket's primary page and mark it to indicate that
896  * split is in progress.
897  */
898  nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
899  nopaque->hasho_prevblkno = maxbucket;
901  nopaque->hasho_bucket = new_bucket;
903  nopaque->hasho_page_id = HASHO_PAGE_ID;
904 
905  MarkBufferDirty(buf_nblkno);
906 
907  /* XLOG stuff */
908  if (RelationNeedsWAL(rel))
909  {
911  XLogRecPtr recptr;
912 
913  xlrec.new_bucket = maxbucket;
914  xlrec.old_bucket_flag = oopaque->hasho_flag;
915  xlrec.new_bucket_flag = nopaque->hasho_flag;
916  xlrec.flags = 0;
917 
918  XLogBeginInsert();
919 
920  XLogRegisterBuffer(0, buf_oblkno, REGBUF_STANDARD);
921  XLogRegisterBuffer(1, buf_nblkno, REGBUF_WILL_INIT);
922  XLogRegisterBuffer(2, metabuf, REGBUF_STANDARD);
923 
924  if (metap_update_masks)
925  {
927  XLogRegisterBufData(2, (char *) &metap->hashm_lowmask, sizeof(uint32));
928  XLogRegisterBufData(2, (char *) &metap->hashm_highmask, sizeof(uint32));
929  }
930 
931  if (metap_update_splitpoint)
932  {
934  XLogRegisterBufData(2, (char *) &metap->hashm_ovflpoint,
935  sizeof(uint32));
937  (char *) &metap->hashm_spares[metap->hashm_ovflpoint],
938  sizeof(uint32));
939  }
940 
941  XLogRegisterData((char *) &xlrec, SizeOfHashSplitAllocPage);
942 
943  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_ALLOCATE_PAGE);
944 
945  PageSetLSN(BufferGetPage(buf_oblkno), recptr);
946  PageSetLSN(BufferGetPage(buf_nblkno), recptr);
947  PageSetLSN(BufferGetPage(metabuf), recptr);
948  }
949 
951 
952  /* drop lock, but keep pin */
953  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
954 
955  /* Relocate records to the new bucket */
956  _hash_splitbucket(rel, metabuf,
957  old_bucket, new_bucket,
958  buf_oblkno, buf_nblkno, NULL,
959  maxbucket, highmask, lowmask);
960 
961  /* all done, now release the pins on primary buckets. */
962  _hash_dropbuf(rel, buf_oblkno);
963  _hash_dropbuf(rel, buf_nblkno);
964 
965  return;
966 
967  /* Here if decide not to split or fail to acquire old bucket lock */
968 fail:
969 
970  /* We didn't write the metapage, so just drop lock */
971  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
972 }
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:254
uint32 hashm_highmask
Definition: hash.h:260
#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:673
uint32 BlockNumber
Definition: block.h:31
Buffer _hash_getnewbuf(Relation rel, BlockNumber blkno, ForkNumber forkNum)
Definition: hashpage.c:207
void _hash_dropbuf(Relation rel, Buffer buf)
Definition: hashpage.c:286
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:89
uint32 hashm_lowmask
Definition: hash.h:261
#define BUCKET_TO_BLKNO(metap, B)
Definition: hash.h:38
static bool _hash_alloc_buckets(Relation rel, BlockNumber firstblock, uint32 nblocks)
Definition: hashpage.c:1000
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:1363
#define REGBUF_STANDARD
Definition: xloginsert.h:34
#define LH_BUCKET_BEING_SPLIT
Definition: hash.h:69
unsigned int uint32
Definition: c.h:325
Buffer _hash_getbuf_with_condlock_cleanup(Relation rel, BlockNumber blkno, int flags)
Definition: hashpage.c:105
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
bool IsBufferCleanupOK(Buffer buffer)
Definition: bufmgr.c:3774
uint32 hashm_ovflpoint
Definition: hash.h:262
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:253
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
uint32 hashm_spares[HASH_MAX_SPLITPOINTS]
Definition: hash.h:267
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:275
#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:1080
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:699
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:510
uint32 hashm_maxbucket
Definition: hash.h:259
uint16 hasho_flag
Definition: hash.h:92
#define XLOG_HASH_SPLIT_ALLOCATE_PAGE
Definition: hash_xlog.h:31
#define HashPageGetMeta(page)
Definition: hash.h:313
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

◆ _hash_finish_split()

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

Definition at line 1363 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().

1365 {
1366  HASHCTL hash_ctl;
1367  HTAB *tidhtab;
1368  Buffer bucket_nbuf = InvalidBuffer;
1369  Buffer nbuf;
1370  Page npage;
1371  BlockNumber nblkno;
1372  BlockNumber bucket_nblkno;
1373  HashPageOpaque npageopaque;
1374  Bucket nbucket;
1375  bool found;
1376 
1377  /* Initialize hash tables used to track TIDs */
1378  memset(&hash_ctl, 0, sizeof(hash_ctl));
1379  hash_ctl.keysize = sizeof(ItemPointerData);
1380  hash_ctl.entrysize = sizeof(ItemPointerData);
1381  hash_ctl.hcxt = CurrentMemoryContext;
1382 
1383  tidhtab =
1384  hash_create("bucket ctids",
1385  256, /* arbitrary initial size */
1386  &hash_ctl,
1388 
1389  bucket_nblkno = nblkno = _hash_get_newblock_from_oldbucket(rel, obucket);
1390 
1391  /*
1392  * Scan the new bucket and build hash table of TIDs
1393  */
1394  for (;;)
1395  {
1396  OffsetNumber noffnum;
1397  OffsetNumber nmaxoffnum;
1398 
1399  nbuf = _hash_getbuf(rel, nblkno, HASH_READ,
1401 
1402  /* remember the primary bucket buffer to acquire cleanup lock on it. */
1403  if (nblkno == bucket_nblkno)
1404  bucket_nbuf = nbuf;
1405 
1406  npage = BufferGetPage(nbuf);
1407  npageopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
1408 
1409  /* Scan each tuple in new page */
1410  nmaxoffnum = PageGetMaxOffsetNumber(npage);
1411  for (noffnum = FirstOffsetNumber;
1412  noffnum <= nmaxoffnum;
1413  noffnum = OffsetNumberNext(noffnum))
1414  {
1415  IndexTuple itup;
1416 
1417  /* Fetch the item's TID and insert it in hash table. */
1418  itup = (IndexTuple) PageGetItem(npage,
1419  PageGetItemId(npage, noffnum));
1420 
1421  (void) hash_search(tidhtab, &itup->t_tid, HASH_ENTER, &found);
1422 
1423  Assert(!found);
1424  }
1425 
1426  nblkno = npageopaque->hasho_nextblkno;
1427 
1428  /*
1429  * release our write lock without modifying buffer and ensure to
1430  * retain the pin on primary bucket.
1431  */
1432  if (nbuf == bucket_nbuf)
1434  else
1435  _hash_relbuf(rel, nbuf);
1436 
1437  /* Exit loop if no more overflow pages in new bucket */
1438  if (!BlockNumberIsValid(nblkno))
1439  break;
1440  }
1441 
1442  /*
1443  * Conditionally get the cleanup lock on old and new buckets to perform
1444  * the split operation. If we don't get the cleanup locks, silently give
1445  * up and next insertion on old bucket will try again to complete the
1446  * split.
1447  */
1449  {
1450  hash_destroy(tidhtab);
1451  return;
1452  }
1453  if (!ConditionalLockBufferForCleanup(bucket_nbuf))
1454  {
1456  hash_destroy(tidhtab);
1457  return;
1458  }
1459 
1460  npage = BufferGetPage(bucket_nbuf);
1461  npageopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
1462  nbucket = npageopaque->hasho_bucket;
1463 
1464  _hash_splitbucket(rel, metabuf, obucket,
1465  nbucket, obuf, bucket_nbuf, tidhtab,
1466  maxbucket, highmask, lowmask);
1467 
1468  _hash_dropbuf(rel, bucket_nbuf);
1469  hash_destroy(tidhtab);
1470 }
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:87
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:286
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:79
#define PageGetMaxOffsetNumber(page)
Definition: bufpage.h:353
uint16 OffsetNumber
Definition: off.h:24
#define HASH_READ
Definition: hash.h:329
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:38
#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:275
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:1080
#define Assert(condition)
Definition: c.h:699
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

◆ _hash_first()

bool _hash_first ( IndexScanDesc  scan,
ScanDirection  dir 
)

Definition at line 292 of file hashsearch.c.

References _hash_datum2hashkey(), _hash_datum2hashkey_type(), _hash_dropbuf(), _hash_get_oldblock_from_newbucket(), _hash_getbucketbuf_from_hashkey(), _hash_getbuf(), _hash_readnext(), _hash_readpage(), Assert, BlockNumberIsValid, buf, HashScanPosData::buf, BUFFER_LOCK_SHARE, BUFFER_LOCK_UNLOCK, BufferGetBlockNumber(), BufferGetPage, BufferIsInvalid, cur, HashScanOpaqueData::currPos, ereport, errcode(), errmsg(), ERROR, H_BUCKET_BEING_POPULATED, HASH_READ, HashScanOpaqueData::hashso_buc_populated, HashScanOpaqueData::hashso_buc_split, HashScanOpaqueData::hashso_bucket_buf, HashScanOpaqueData::hashso_sk_hash, HashScanOpaqueData::hashso_split_bucket_buf, HTEqualStrategyNumber, IndexScanDescData::indexRelation, InvalidBuffer, InvalidOid, HashScanPosData::itemIndex, HashScanPosData::items, IndexScanDescData::keyData, LH_BUCKET_PAGE, LockBuffer(), IndexScanDescData::numberOfKeys, IndexScanDescData::opaque, PageGetSpecialPointer, pgstat_count_index_scan, PredicateLockPage(), RelationData::rd_opcintype, ScanDirectionIsBackward, SK_ISNULL, HeapTupleData::t_self, TestForOldSnapshot(), IndexScanDescData::xs_ctup, and IndexScanDescData::xs_snapshot.

Referenced by hashgetbitmap(), and hashgettuple().

293 {
294  Relation rel = scan->indexRelation;
295  HashScanOpaque so = (HashScanOpaque) scan->opaque;
296  ScanKey cur;
297  uint32 hashkey;
298  Bucket bucket;
299  Buffer buf;
300  Page page;
301  HashPageOpaque opaque;
302  HashScanPosItem *currItem;
303 
305 
306  /*
307  * We do not support hash scans with no index qualification, because we
308  * would have to read the whole index rather than just one bucket. That
309  * creates a whole raft of problems, since we haven't got a practical way
310  * to lock all the buckets against splits or compactions.
311  */
312  if (scan->numberOfKeys < 1)
313  ereport(ERROR,
314  (errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
315  errmsg("hash indexes do not support whole-index scans")));
316 
317  /* There may be more than one index qual, but we hash only the first */
318  cur = &scan->keyData[0];
319 
320  /* We support only single-column hash indexes */
321  Assert(cur->sk_attno == 1);
322  /* And there's only one operator strategy, too */
323  Assert(cur->sk_strategy == HTEqualStrategyNumber);
324 
325  /*
326  * If the constant in the index qual is NULL, assume it cannot match any
327  * items in the index.
328  */
329  if (cur->sk_flags & SK_ISNULL)
330  return false;
331 
332  /*
333  * Okay to compute the hash key. We want to do this before acquiring any
334  * locks, in case a user-defined hash function happens to be slow.
335  *
336  * If scankey operator is not a cross-type comparison, we can use the
337  * cached hash function; otherwise gotta look it up in the catalogs.
338  *
339  * We support the convention that sk_subtype == InvalidOid means the
340  * opclass input type; this is a hack to simplify life for ScanKeyInit().
341  */
342  if (cur->sk_subtype == rel->rd_opcintype[0] ||
343  cur->sk_subtype == InvalidOid)
344  hashkey = _hash_datum2hashkey(rel, cur->sk_argument);
345  else
346  hashkey = _hash_datum2hashkey_type(rel, cur->sk_argument,
347  cur->sk_subtype);
348 
349  so->hashso_sk_hash = hashkey;
350 
351  buf = _hash_getbucketbuf_from_hashkey(rel, hashkey, HASH_READ, NULL);
353  page = BufferGetPage(buf);
354  TestForOldSnapshot(scan->xs_snapshot, rel, page);
355  opaque = (HashPageOpaque) PageGetSpecialPointer(page);
356  bucket = opaque->hasho_bucket;
357 
358  so->hashso_bucket_buf = buf;
359 
360  /*
361  * If a bucket split is in progress, then while scanning the bucket being
362  * populated, we need to skip tuples that were copied from bucket being
363  * split. We also need to maintain a pin on the bucket being split to
364  * ensure that split-cleanup work done by vacuum doesn't remove tuples
365  * from it till this scan is done. We need to maintain a pin on the
366  * bucket being populated to ensure that vacuum doesn't squeeze that
367  * bucket till this scan is complete; otherwise, the ordering of tuples
368  * can't be maintained during forward and backward scans. Here, we have
369  * to be cautious about locking order: first, acquire the lock on bucket
370  * being split; then, release the lock on it but not the pin; then,
371  * acquire a lock on bucket being populated and again re-verify whether
372  * the bucket split is still in progress. Acquiring the lock on bucket
373  * being split first ensures that the vacuum waits for this scan to
374  * finish.
375  */
376  if (H_BUCKET_BEING_POPULATED(opaque))
377  {
378  BlockNumber old_blkno;
379  Buffer old_buf;
380 
381  old_blkno = _hash_get_oldblock_from_newbucket(rel, bucket);
382 
383  /*
384  * release the lock on new bucket and re-acquire it after acquiring
385  * the lock on old bucket.
386  */
388 
389  old_buf = _hash_getbuf(rel, old_blkno, HASH_READ, LH_BUCKET_PAGE);
390  TestForOldSnapshot(scan->xs_snapshot, rel, BufferGetPage(old_buf));
391 
392  /*
393  * remember the split bucket buffer so as to use it later for
394  * scanning.
395  */
396  so->hashso_split_bucket_buf = old_buf;
397  LockBuffer(old_buf, BUFFER_LOCK_UNLOCK);
398 
400  page = BufferGetPage(buf);
401  opaque = (HashPageOpaque) PageGetSpecialPointer(page);
402  Assert(opaque->hasho_bucket == bucket);
403 
404  if (H_BUCKET_BEING_POPULATED(opaque))
405  so->hashso_buc_populated = true;
406  else
407  {
410  }
411  }
412 
413  /* If a backwards scan is requested, move to the end of the chain */
414  if (ScanDirectionIsBackward(dir))
415  {
416  /*
417  * Backward scans that start during split needs to start from end of
418  * bucket being split.
419  */
420  while (BlockNumberIsValid(opaque->hasho_nextblkno) ||
422  _hash_readnext(scan, &buf, &page, &opaque);
423  }
424 
425  /* remember which buffer we have pinned, if any */
427  so->currPos.buf = buf;
428 
429  /* Now find all the tuples satisfying the qualification from a page */
430  if (!_hash_readpage(scan, &buf, dir))
431  return false;
432 
433  /* OK, itemIndex says what to return */
434  currItem = &so->currPos.items[so->currPos.itemIndex];
435  scan->xs_ctup.t_self = currItem->heapTid;
436 
437  /* if we're here, _hash_readpage found a valid tuples */
438  return true;
439 }
HashScanOpaqueData * HashScanOpaque
Definition: hash.h:200
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:87
void PredicateLockPage(Relation relation, BlockNumber blkno, Snapshot snapshot)
Definition: predicate.c:2475
static void TestForOldSnapshot(Snapshot snapshot, Relation relation, Page page)
Definition: bufmgr.h:265
static void _hash_readnext(IndexScanDesc scan, Buffer *bufp, Page *pagep, HashPageOpaque *opaquep)
Definition: hashsearch.c:133
static bool _hash_readpage(IndexScanDesc scan, Buffer *bufP, ScanDirection dir)
Definition: hashsearch.c:452
#define HTEqualStrategyNumber
Definition: hash.h:336
Snapshot xs_snapshot
Definition: relscan.h:92
#define InvalidBuffer
Definition: buf.h:25
struct cursor * cur
Definition: ecpg.c:28
int errcode(int sqlerrcode)
Definition: elog.c:575
Buffer _hash_getbucketbuf_from_hashkey(Relation rel, uint32 hashkey, int access, HashMetaPage *cachedmetap)
Definition: hashpage.c:1567
uint32 BlockNumber
Definition: block.h:31
uint32 _hash_datum2hashkey(Relation rel, Datum key)
Definition: hashutil.c:82
void _hash_dropbuf(Relation rel, Buffer buf)
Definition: hashpage.c:286
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:79
Relation indexRelation
Definition: relscan.h:91
#define HASH_READ
Definition: hash.h:329
uint32 Bucket
Definition: hash.h:34
#define ScanDirectionIsBackward(direction)
Definition: sdir.h:41
#define ERROR
Definition: elog.h:43
ItemPointerData t_self
Definition: htup.h:65
#define BufferIsInvalid(buffer)
Definition: buf.h:31
static char * buf
Definition: pg_test_fsync.c:67
#define pgstat_count_index_scan(rel)
Definition: pgstat.h:1291
unsigned int uint32
Definition: c.h:325
#define SK_ISNULL
Definition: skey.h:115
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define ereport(elevel, rest)
Definition: elog.h:122
Buffer hashso_bucket_buf
Definition: hash.h:172
#define H_BUCKET_BEING_POPULATED(opaque)
Definition: hash.h:100
bool hashso_buc_populated
Definition: hash.h:182
Buffer buf
Definition: hash.h:119
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
uint32 hashso_sk_hash
Definition: hash.h:169
#define InvalidOid
Definition: postgres_ext.h:36
#define BlockNumberIsValid(blockNumber)
Definition: block.h:70
#define LH_BUCKET_PAGE
Definition: hash.h:65
#define Assert(condition)
Definition: c.h:699
HeapTupleData xs_ctup
Definition: relscan.h:121
#define PageGetSpecialPointer(page)
Definition: bufpage.h:322
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:96
ScanKey keyData
Definition: relscan.h:95
BlockNumber _hash_get_oldblock_from_newbucket(Relation rel, Bucket new_bucket)
Definition: hashutil.c:430
HashScanPosData currPos
Definition: hash.h:197
bool hashso_buc_split
Definition: hash.h:188
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:2605
int errmsg(const char *fmt,...)
Definition: elog.c:797
Buffer hashso_split_bucket_buf
Definition: hash.h:179
#define BUFFER_LOCK_SHARE
Definition: bufmgr.h:88
uint32 _hash_datum2hashkey_type(Relation rel, Datum key, Oid keytype)
Definition: hashutil.c:102
int itemIndex
Definition: hash.h:133
Oid * rd_opcintype
Definition: rel.h:155
int Buffer
Definition: buf.h:23
Pointer Page
Definition: bufpage.h:74
HashScanPosItem items[MaxIndexTuplesPerPage]
Definition: hash.h:135

◆ _hash_freeovflpage()

BlockNumber _hash_freeovflpage ( Relation  rel,
Buffer  bucketbuf,
Buffer  ovflbuf,
Buffer  wbuf,
IndexTuple itups,
OffsetNumber itup_offsets,
Size tups_size,
uint16  nitups,
BufferAccessStrategy  bstrategy 
)

Definition at line 489 of file hashovfl.c.

References _hash_checkpage(), _hash_getbuf(), _hash_getbuf_with_strategy(), _hash_ovflblkno_to_bitno(), _hash_pageinit(), _hash_pgaddmultitup(), _hash_relbuf(), Assert, BlockNumberIsValid, BMPG_MASK, BMPG_SHIFT, BUFFER_LOCK_EXCLUSIVE, BUFFER_LOCK_UNLOCK, BufferGetBlockNumber(), BufferGetPage, BufferGetPageSize, BufferIsValid, CLRBIT, elog, END_CRIT_SECTION, ERROR, HASH_METAPAGE, HASH_READ, HASH_WRITE, HASH_XLOG_FREE_OVFL_BUFS, HashMetaPageData::hashm_firstfree, HashMetaPageData::hashm_mapp, HashMetaPageData::hashm_nmaps, HashPageOpaqueData::hasho_bucket, HashPageOpaqueData::hasho_flag, HashPageOpaqueData::hasho_nextblkno, HashPageOpaqueData::hasho_page_id, HASHO_PAGE_ID, HashPageOpaqueData::hasho_prevblkno, HashPageGetBitmap, HashPageGetMeta, i, InvalidBlockNumber, InvalidBuffer, xl_hash_squeeze_page::is_prev_bucket_same_wrt, xl_hash_squeeze_page::is_prim_bucket_same_wrt, ISSET, LH_BITMAP_PAGE, LH_BUCKET_PAGE, LH_META_PAGE, LH_OVERFLOW_PAGE, LH_UNUSED_PAGE, LockBuffer(), MarkBufferDirty(), xl_hash_squeeze_page::nextblkno, xl_hash_squeeze_page::ntups, PageGetSpecialPointer, PageSetLSN, PG_USED_FOR_ASSERTS_ONLY, xl_hash_squeeze_page::prevblkno, REGBUF_NO_IMAGE, REGBUF_STANDARD, RelationNeedsWAL, SizeOfHashSqueezePage, START_CRIT_SECTION, XLOG_HASH_SQUEEZE_PAGE, XLogBeginInsert(), XLogEnsureRecordSpace(), XLogInsert(), XLogRegisterBufData(), XLogRegisterBuffer(), and XLogRegisterData().

Referenced by _hash_squeezebucket().

493 {
494  HashMetaPage metap;
495  Buffer metabuf;
496  Buffer mapbuf;
497  BlockNumber ovflblkno;
498  BlockNumber prevblkno;
499  BlockNumber blkno;
500  BlockNumber nextblkno;
501  BlockNumber writeblkno;
502  HashPageOpaque ovflopaque;
503  Page ovflpage;
504  Page mappage;
505  uint32 *freep;
506  uint32 ovflbitno;
507  int32 bitmappage,
508  bitmapbit;
510  Buffer prevbuf = InvalidBuffer;
511  Buffer nextbuf = InvalidBuffer;
512  bool update_metap = false;
513 
514  /* Get information from the doomed page */
515  _hash_checkpage(rel, ovflbuf, LH_OVERFLOW_PAGE);
516  ovflblkno = BufferGetBlockNumber(ovflbuf);
517  ovflpage = BufferGetPage(ovflbuf);
518  ovflopaque = (HashPageOpaque) PageGetSpecialPointer(ovflpage);
519  nextblkno = ovflopaque->hasho_nextblkno;
520  prevblkno = ovflopaque->hasho_prevblkno;
521  writeblkno = BufferGetBlockNumber(wbuf);
522  bucket = ovflopaque->hasho_bucket;
523 
524  /*
525  * Fix up the bucket chain. this is a doubly-linked list, so we must fix
526  * up the bucket chain members behind and ahead of the overflow page being
527  * deleted. Concurrency issues are avoided by using lock chaining as
528  * described atop hashbucketcleanup.
529  */
530  if (BlockNumberIsValid(prevblkno))
531  {
532  if (prevblkno == writeblkno)
533  prevbuf = wbuf;
534  else
535  prevbuf = _hash_getbuf_with_strategy(rel,
536  prevblkno,
537  HASH_WRITE,
539  bstrategy);
540  }
541  if (BlockNumberIsValid(nextblkno))
542  nextbuf = _hash_getbuf_with_strategy(rel,
543  nextblkno,
544  HASH_WRITE,
546  bstrategy);
547 
548  /* Note: bstrategy is intentionally not used for metapage and bitmap */
549 
550  /* Read the metapage so we can determine which bitmap page to use */
552  metap = HashPageGetMeta(BufferGetPage(metabuf));
553 
554  /* Identify which bit to set */
555  ovflbitno = _hash_ovflblkno_to_bitno(metap, ovflblkno);
556 
557  bitmappage = ovflbitno >> BMPG_SHIFT(metap);
558  bitmapbit = ovflbitno & BMPG_MASK(metap);
559 
560  if (bitmappage >= metap->hashm_nmaps)
561  elog(ERROR, "invalid overflow bit number %u", ovflbitno);
562  blkno = metap->hashm_mapp[bitmappage];
563 
564  /* Release metapage lock while we access the bitmap page */
565  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
566 
567  /* read the bitmap page to clear the bitmap bit */
568  mapbuf = _hash_getbuf(rel, blkno, HASH_WRITE, LH_BITMAP_PAGE);
569  mappage = BufferGetPage(mapbuf);
570  freep = HashPageGetBitmap(mappage);
571  Assert(ISSET(freep, bitmapbit));
572 
573  /* Get write-lock on metapage to update firstfree */
575 
576  /* This operation needs to log multiple tuples, prepare WAL for that */
577  if (RelationNeedsWAL(rel))
579 
581 
582  /*
583  * we have to insert tuples on the "write" page, being careful to preserve
584  * hashkey ordering. (If we insert many tuples into the same "write" page
585  * it would be worth qsort'ing them).
586  */
587  if (nitups > 0)
588  {
589  _hash_pgaddmultitup(rel, wbuf, itups, itup_offsets, nitups);
590  MarkBufferDirty(wbuf);
591  }
592 
593  /*
594  * Reinitialize the freed overflow page. Just zeroing the page won't
595  * work, because WAL replay routines expect pages to be initialized. See
596  * explanation of RBM_NORMAL mode atop XLogReadBufferExtended. We are
597  * careful to make the special space valid here so that tools like
598  * pageinspect won't get confused.
599  */
600  _hash_pageinit(ovflpage, BufferGetPageSize(ovflbuf));
601 
602  ovflopaque = (HashPageOpaque) PageGetSpecialPointer(ovflpage);
603 
604  ovflopaque->hasho_prevblkno = InvalidBlockNumber;
605  ovflopaque->hasho_nextblkno = InvalidBlockNumber;
606  ovflopaque->hasho_bucket = -1;
607  ovflopaque->hasho_flag = LH_UNUSED_PAGE;
608  ovflopaque->hasho_page_id = HASHO_PAGE_ID;
609 
610  MarkBufferDirty(ovflbuf);
611 
612  if (BufferIsValid(prevbuf))
613  {
614  Page prevpage = BufferGetPage(prevbuf);
615  HashPageOpaque prevopaque = (HashPageOpaque) PageGetSpecialPointer(prevpage);
616 
617  Assert(prevopaque->hasho_bucket == bucket);
618  prevopaque->hasho_nextblkno = nextblkno;
619  MarkBufferDirty(prevbuf);
620  }
621  if (BufferIsValid(nextbuf))
622  {
623  Page nextpage = BufferGetPage(nextbuf);
624  HashPageOpaque nextopaque = (HashPageOpaque) PageGetSpecialPointer(nextpage);
625 
626  Assert(nextopaque->hasho_bucket == bucket);
627  nextopaque->hasho_prevblkno = prevblkno;
628  MarkBufferDirty(nextbuf);
629  }
630 
631  /* Clear the bitmap bit to indicate that this overflow page is free */
632  CLRBIT(freep, bitmapbit);
633  MarkBufferDirty(mapbuf);
634 
635  /* if this is now the first free page, update hashm_firstfree */
636  if (ovflbitno < metap->hashm_firstfree)
637  {
638  metap->hashm_firstfree = ovflbitno;
639  update_metap = true;
640  MarkBufferDirty(metabuf);
641  }
642 
643  /* XLOG stuff */
644  if (RelationNeedsWAL(rel))
645  {
646  xl_hash_squeeze_page xlrec;
647  XLogRecPtr recptr;
648  int i;
649 
650  xlrec.prevblkno = prevblkno;
651  xlrec.nextblkno = nextblkno;
652  xlrec.ntups = nitups;
653  xlrec.is_prim_bucket_same_wrt = (wbuf == bucketbuf);
654  xlrec.is_prev_bucket_same_wrt = (wbuf == prevbuf);
655 
656  XLogBeginInsert();
657  XLogRegisterData((char *) &xlrec, SizeOfHashSqueezePage);
658 
659  /*
660  * bucket buffer needs to be registered to ensure that we can acquire
661  * a cleanup lock on it during replay.
662  */
663  if (!xlrec.is_prim_bucket_same_wrt)
665 
667  if (xlrec.ntups > 0)
668  {
669  XLogRegisterBufData(1, (char *) itup_offsets,
670  nitups * sizeof(OffsetNumber));
671  for (i = 0; i < nitups; i++)
672  XLogRegisterBufData(1, (char *) itups[i], tups_size[i]);
673  }
674 
675  XLogRegisterBuffer(2, ovflbuf, REGBUF_STANDARD);
676 
677  /*
678  * If prevpage and the writepage (block in which we are moving tuples
679  * from overflow) are same, then no need to separately register
680  * prevpage. During replay, we can directly update the nextblock in
681  * writepage.
682  */
683  if (BufferIsValid(prevbuf) && !xlrec.is_prev_bucket_same_wrt)
684  XLogRegisterBuffer(3, prevbuf, REGBUF_STANDARD);
685 
686  if (BufferIsValid(nextbuf))
687  XLogRegisterBuffer(4, nextbuf, REGBUF_STANDARD);
688 
690  XLogRegisterBufData(5, (char *) &bitmapbit, sizeof(uint32));
691 
692  if (update_metap)
693  {
694  XLogRegisterBuffer(6, metabuf, REGBUF_STANDARD);
695  XLogRegisterBufData(6, (char *) &metap->hashm_firstfree, sizeof(uint32));
696  }
697 
698  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SQUEEZE_PAGE);
699 
700  PageSetLSN(BufferGetPage(wbuf), recptr);
701  PageSetLSN(BufferGetPage(ovflbuf), recptr);
702 
703  if (BufferIsValid(prevbuf) && !xlrec.is_prev_bucket_same_wrt)
704  PageSetLSN(BufferGetPage(prevbuf), recptr);
705  if (BufferIsValid(nextbuf))
706  PageSetLSN(BufferGetPage(nextbuf), recptr);
707 
708  PageSetLSN(BufferGetPage(mapbuf), recptr);
709 
710  if (update_metap)
711  PageSetLSN(BufferGetPage(metabuf), recptr);
712  }
713 
715 
716  /* release previous bucket if it is not same as write bucket */
717  if (BufferIsValid(prevbuf) && prevblkno != writeblkno)
718  _hash_relbuf(rel, prevbuf);
719 
720  if (BufferIsValid(ovflbuf))
721  _hash_relbuf(rel, ovflbuf);
722 
723  if (BufferIsValid(nextbuf))
724  _hash_relbuf(rel, nextbuf);
725 
726  _hash_relbuf(rel, mapbuf);
727  _hash_relbuf(rel, metabuf);
728 
729  return nextblkno;
730 }
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
void _hash_pgaddmultitup(Relation rel, Buffer buf, IndexTuple *itups, OffsetNumber *itup_offsets, uint16 nitups)
Definition: hashinsert.c:300
#define HashPageGetBitmap(page)
Definition: hash.h:306
#define LH_BITMAP_PAGE
Definition: hash.h:66
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:608
#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
Buffer _hash_getbuf_with_strategy(Relation rel, BlockNumber blkno, int access, int flags, BufferAccessStrategy bstrategy)
Definition: hashpage.c:248
#define END_CRIT_SECTION()
Definition: miscadmin.h:133
#define InvalidBuffer
Definition: buf.h:25
#define START_CRIT_SECTION()
Definition: miscadmin.h:131
uint32 BlockNumber
Definition: block.h:31
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:79
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:89
signed int int32
Definition: c.h:313
BlockNumber prevblkno
Definition: hash_xlog.h:174
#define LH_UNUSED_PAGE
Definition: hash.h:63
bool is_prim_bucket_same_wrt
Definition: hash_xlog.h:177
uint16 OffsetNumber
Definition: off.h:24
#define HASH_READ
Definition: hash.h:329
uint32 Bucket
Definition: hash.h:34
#define SizeOfHashSqueezePage
Definition: hash_xlog.h:186
BlockNumber hasho_prevblkno
Definition: hash.h:89
#define ERROR
Definition: elog.h:43
BlockNumber nextblkno
Definition: hash_xlog.h:175
uint32 hashm_nmaps
Definition: hash.h:265
#define HASH_WRITE
Definition: hash.h:330
#define BMPG_MASK(metap)
Definition: hash.h:304
#define REGBUF_STANDARD
Definition: xloginsert.h:34
unsigned int uint32
Definition: c.h:325
#define BMPG_SHIFT(metap)
Definition: hash.h:303
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define HASH_XLOG_FREE_OVFL_BUFS
Definition: hash_xlog.h:22
#define ISSET(A, N)
Definition: hash.h:324
void XLogRegisterData(char *data, int len)
Definition: xloginsert.c:323
XLogRecPtr XLogInsert(RmgrId rmid, uint8 info)
Definition: xloginsert.c:415
void _hash_checkpage(Relation rel, Buffer buf, int flags)
Definition: hashutil.c:225
#define CLRBIT(x, i)
Definition: blutils.c:32
#define HASH_METAPAGE
Definition: hash.h:206
#define BufferGetPageSize(buffer)
Definition: bufmgr.h:147
#define LH_OVERFLOW_PAGE
Definition: hash.h:64
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
uint32 hashm_firstfree
Definition: hash.h:264
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:275
#define BlockNumberIsValid(blockNumber)
Definition: block.h:70
#define LH_BUCKET_PAGE
Definition: hash.h:65
uint64 XLogRecPtr
Definition: xlogdefs.h:21
#define Assert(condition)
Definition: c.h:699
Bucket hasho_bucket
Definition: hash.h:91
bool is_prev_bucket_same_wrt
Definition: hash_xlog.h:180
#define PageGetSpecialPointer(page)
Definition: bufpage.h:322
#define REGBUF_NO_IMAGE
Definition: xloginsert.h:31
#define InvalidBlockNumber
Definition: block.h:33
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:96
void XLogEnsureRecordSpace(int max_block_id, int ndatas)
Definition: xloginsert.c:146
#define HASHO_PAGE_ID
Definition: hash.h:109
#define BufferIsValid(bufnum)
Definition: bufmgr.h:114
#define RelationNeedsWAL(relation)
Definition: rel.h:510
uint32 _hash_ovflblkno_to_bitno(HashMetaPage metap, BlockNumber ovflblkno)
Definition: hashovfl.c:61
uint16 hasho_flag
Definition: hash.h:92
BlockNumber BufferGetBlockNumber(Buffer buffer)
Definition: bufmgr.c:2605
#define HashPageGetMeta(page)
Definition: hash.h:313
int i
#define XLOG_HASH_SQUEEZE_PAGE
Definition: hash_xlog.h:36
BlockNumber hasho_nextblkno
Definition: hash.h:90
#define elog
Definition: elog.h:219
#define PG_USED_FOR_ASSERTS_ONLY
Definition: c.h:123
void XLogBeginInsert(void)
Definition: xloginsert.c:120
#define PageSetLSN(page, lsn)
Definition: bufpage.h:364
BlockNumber hashm_mapp[HASH_MAX_BITMAPS]
Definition: hash.h:269
int Buffer
Definition: buf.h:23
Pointer Page
Definition: bufpage.h:74

◆ _hash_get_indextuple_hashkey()

uint32 _hash_get_indextuple_hashkey ( IndexTuple  itup)

Definition at line 299 of file hashutil.c.

References IndexInfoFindDataOffset, and IndexTupleData::t_info.

Referenced by _h_indexbuild(), _hash_binsearch(), _hash_binsearch_last(), _hash_doinsert(), _hash_load_qualified_items(), _hash_pgaddmultitup(), _hash_pgaddtup(), _hash_splitbucket(), hash_page_items(), and hashbucketcleanup().

300 {
301  char *attp;
302 
303  /*
304  * We assume the hash key is the first attribute and can't be null, so
305  * this can be done crudely but very very cheaply ...
306  */
307  attp = (char *) itup + IndexInfoFindDataOffset(itup->t_info);
308  return *((uint32 *) attp);
309 }
#define IndexInfoFindDataOffset(t_info)
Definition: itup.h:80
unsigned int uint32
Definition: c.h:325
unsigned short t_info
Definition: itup.h:49

◆ _hash_get_newblock_from_oldbucket()

BlockNumber _hash_get_newblock_from_oldbucket ( Relation  rel,
Bucket  old_bucket 
)

Definition at line 469 of file hashutil.c.

References _hash_get_newbucket_from_oldbucket(), _hash_getbuf(), _hash_relbuf(), BUCKET_TO_BLKNO, BufferGetPage, HASH_METAPAGE, HASH_READ, HashMetaPageData::hashm_lowmask, HashMetaPageData::hashm_maxbucket, HashPageGetMeta, and LH_META_PAGE.

Referenced by _hash_finish_split().

470 {
471  Bucket new_bucket;
472  Buffer metabuf;
473  HashMetaPage metap;
474  BlockNumber blkno;
475 
477  metap = HashPageGetMeta(BufferGetPage(metabuf));
478 
479  new_bucket = _hash_get_newbucket_from_oldbucket(rel, old_bucket,
480  metap->hashm_lowmask,
481  metap->hashm_maxbucket);
482  blkno = BUCKET_TO_BLKNO(metap, new_bucket);
483 
484  _hash_relbuf(rel, metabuf);
485 
486  return blkno;
487 }
#define LH_META_PAGE
Definition: hash.h:67
uint32 BlockNumber
Definition: block.h:31
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:79
uint32 hashm_lowmask
Definition: hash.h:261
#define BUCKET_TO_BLKNO(metap, B)
Definition: hash.h:38
#define HASH_READ
Definition: hash.h:329
uint32 Bucket
Definition: hash.h:34
Bucket _hash_get_newbucket_from_oldbucket(Relation rel, Bucket old_bucket, uint32 lowmask, uint32 maxbucket)
Definition: hashutil.c:502
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define HASH_METAPAGE
Definition: hash.h:206
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:275
uint32 hashm_maxbucket
Definition: hash.h:259
#define HashPageGetMeta(page)
Definition: hash.h:313
int Buffer
Definition: buf.h:23

◆ _hash_get_newbucket_from_oldbucket()

Bucket _hash_get_newbucket_from_oldbucket ( Relation  rel,
Bucket  old_bucket,
uint32  lowmask,
uint32  maxbucket 
)

Definition at line 502 of file hashutil.c.

References CALC_NEW_BUCKET.

Referenced by _hash_get_newblock_from_oldbucket(), and hashbucketcleanup().

504 {
505  Bucket new_bucket;
506 
507  new_bucket = CALC_NEW_BUCKET(old_bucket, lowmask);
508  if (new_bucket > maxbucket)
509  {
510  lowmask = lowmask >> 1;
511  new_bucket = CALC_NEW_BUCKET(old_bucket, lowmask);
512  }
513 
514  return new_bucket;
515 }
#define CALC_NEW_BUCKET(old_bucket, lowmask)
Definition: hashutil.c:24
uint32 Bucket
Definition: hash.h:34

◆ _hash_get_oldblock_from_newbucket()

BlockNumber _hash_get_oldblock_from_newbucket ( Relation  rel,
Bucket  new_bucket 
)

Definition at line 430 of file hashutil.c.

References _hash_getbuf(), _hash_relbuf(), BUCKET_TO_BLKNO, BufferGetPage, fls(), HASH_METAPAGE, HASH_READ, HashPageGetMeta, and LH_META_PAGE.

Referenced by _hash_first().

431 {
432  Bucket old_bucket;
433  uint32 mask;
434  Buffer metabuf;
435  HashMetaPage metap;
436  BlockNumber blkno;
437 
438  /*
439  * To get the old bucket from the current bucket, we need a mask to modulo
440  * into lower half of table. This mask is stored in meta page as
441  * hashm_lowmask, but here we can't rely on the same, because we need a
442  * value of lowmask that was prevalent at the time when bucket split was
443  * started. Masking the most significant bit of new bucket would give us
444  * old bucket.
445  */
446  mask = (((uint32) 1) << (fls(new_bucket) - 1)) - 1;
447  old_bucket = new_bucket & mask;
448 
450  metap = HashPageGetMeta(BufferGetPage(metabuf));
451 
452  blkno = BUCKET_TO_BLKNO(metap, old_bucket);
453 
454  _hash_relbuf(rel, metabuf);
455 
456  return blkno;
457 }
#define LH_META_PAGE
Definition: hash.h:67
uint32 BlockNumber
Definition: block.h:31
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:79
#define BUCKET_TO_BLKNO(metap, B)
Definition: hash.h:38
int fls(int mask)
Definition: fls.c:55
#define HASH_READ
Definition: hash.h:329
uint32 Bucket
Definition: hash.h:34
unsigned int uint32
Definition: c.h:325
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define HASH_METAPAGE
Definition: hash.h:206
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:275
#define HashPageGetMeta(page)
Definition: hash.h:313
int Buffer
Definition: buf.h:23

◆ _hash_get_totalbuckets()

uint32 _hash_get_totalbuckets ( uint32  splitpoint_phase)

Definition at line 189 of file hashutil.c.

References HASH_SPLITPOINT_GROUPS_WITH_ONE_PHASE, HASH_SPLITPOINT_PHASE_BITS, and HASH_SPLITPOINT_PHASE_MASK.

Referenced by _hash_expandtable(), _hash_init_metabuffer(), _hash_ovflblkno_to_bitno(), and bitno_to_blkno().

190 {
191  uint32 splitpoint_group;
192  uint32 total_buckets;
193  uint32 phases_within_splitpoint_group;
194 
195  if (splitpoint_phase < HASH_SPLITPOINT_GROUPS_WITH_ONE_PHASE)
196  return (1 << splitpoint_phase);
197 
198  /* get splitpoint's group */
199  splitpoint_group = HASH_SPLITPOINT_GROUPS_WITH_ONE_PHASE;
200  splitpoint_group +=
201  ((splitpoint_phase - HASH_SPLITPOINT_GROUPS_WITH_ONE_PHASE) >>
203 
204  /* account for buckets before splitpoint_group */
205  total_buckets = (1 << (splitpoint_group - 1));
206 
207  /* account for buckets within splitpoint_group */
208  phases_within_splitpoint_group =
209  (((splitpoint_phase - HASH_SPLITPOINT_GROUPS_WITH_ONE_PHASE) &
210  HASH_SPLITPOINT_PHASE_MASK) + 1); /* from 0-based to 1-based */
211  total_buckets +=
212  (((1 << (splitpoint_group - 1)) >> HASH_SPLITPOINT_PHASE_BITS) *
213  phases_within_splitpoint_group);
214 
215  return total_buckets;
216 }
#define HASH_SPLITPOINT_PHASE_MASK
Definition: hash.h:239
#define HASH_SPLITPOINT_PHASE_BITS
Definition: hash.h:237
unsigned int uint32
Definition: c.h:325
#define HASH_SPLITPOINT_GROUPS_WITH_ONE_PHASE
Definition: hash.h:240

◆ _hash_getbucketbuf_from_hashkey()

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

Definition at line 1567 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().

1569 {
1570  HashMetaPage metap;
1571  Buffer buf;
1572  Buffer metabuf = InvalidBuffer;
1573  Page page;
1574  Bucket bucket;
1575  BlockNumber blkno;
1576  HashPageOpaque opaque;
1577 
1578  /* We read from target bucket buffer, hence locking is must. */
1579  Assert(access == HASH_READ || access == HASH_WRITE);
1580 
1581  metap = _hash_getcachedmetap(rel, &metabuf, false);
1582  Assert(metap != NULL);
1583 
1584  /*
1585  * Loop until we get a lock on the correct target bucket.
1586  */
1587  for (;;)
1588  {
1589  /*
1590  * Compute the target bucket number, and convert to block number.
1591  */
1592  bucket = _hash_hashkey2bucket(hashkey,
1593  metap->hashm_maxbucket,
1594  metap->hashm_highmask,
1595  metap->hashm_lowmask);
1596 
1597  blkno = BUCKET_TO_BLKNO(metap, bucket);
1598 
1599  /* Fetch the primary bucket page for the bucket */
1600  buf = _hash_getbuf(rel, blkno, access, LH_BUCKET_PAGE);
1601  page = BufferGetPage(buf);
1602  opaque = (HashPageOpaque) PageGetSpecialPointer(page);
1603  Assert(opaque->hasho_bucket == bucket);
1605 
1606  /*
1607  * If this bucket hasn't been split, we're done.
1608  */
1609  if (opaque->hasho_prevblkno <= metap->hashm_maxbucket)
1610  break;
1611 
1612  /* Drop lock on this buffer, update cached metapage, and retry. */
1613  _hash_relbuf(rel, buf);
1614  metap = _hash_getcachedmetap(rel, &metabuf, true);
1615  Assert(metap != NULL);
1616  }
1617 
1618  if (BufferIsValid(metabuf))
1619  _hash_dropbuf(rel, metabuf);
1620 
1621  if (cachedmetap)
1622  *cachedmetap = metap;
1623 
1624  return buf;
1625 }
Bucket _hash_hashkey2bucket(uint32 hashkey, uint32 maxbucket, uint32 highmask, uint32 lowmask)
Definition: hashutil.c:125
uint32 hashm_highmask
Definition: hash.h:260
#define InvalidBuffer
Definition: buf.h:25
uint32 BlockNumber
Definition: block.h:31
void _hash_dropbuf(Relation rel, Buffer buf)
Definition: hashpage.c:286
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:79
uint32 hashm_lowmask
Definition: hash.h:261
#define BUCKET_TO_BLKNO(metap, B)
Definition: hash.h:38
#define HASH_READ
Definition: hash.h:329
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:330
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
HashMetaPage _hash_getcachedmetap(Relation rel, Buffer *metabuf, bool force_refresh)
Definition: hashpage.c:1509
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:275
#define LH_BUCKET_PAGE
Definition: hash.h:65
#define Assert(condition)
Definition: c.h:699
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:259
int Buffer
Definition: buf.h:23
Pointer Page
Definition: bufpage.h:74

◆ _hash_getbuf()

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

Definition at line 79 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().

80 {
81  Buffer buf;
82 
83  if (blkno == P_NEW)
84  elog(ERROR, "hash AM does not use P_NEW");
85 
86  buf = ReadBuffer(rel, blkno);
87 
88  if (access != HASH_NOLOCK)
89  LockBuffer(buf, access);
90 
91  /* ref count and lock type are correct */
92 
93  _hash_checkpage(rel, buf, flags);
94 
95  return buf;
96 }
#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:331
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

◆ _hash_getbuf_with_condlock_cleanup()

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

Definition at line 105 of file hashpage.c.

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

Referenced by _hash_expandtable().

106 {
107  Buffer buf;
108 
109  if (blkno == P_NEW)
110  elog(ERROR, "hash AM does not use P_NEW");
111 
112  buf = ReadBuffer(rel, blkno);
113 
115  {
116  ReleaseBuffer(buf);
117  return InvalidBuffer;
118  }
119 
120  /* ref count and lock type are correct */
121 
122  _hash_checkpage(rel, buf, flags);
123 
124  return buf;
125 }
#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

◆ _hash_getbuf_with_strategy()

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

Definition at line 248 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().

251 {
252  Buffer buf;
253 
254  if (blkno == P_NEW)
255  elog(ERROR, "hash AM does not use P_NEW");
256 
257  buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_NORMAL, bstrategy);
258 
259  if (access != HASH_NOLOCK)
260  LockBuffer(buf, access);
261 
262  /* ref count and lock type are correct */
263 
264  _hash_checkpage(rel, buf, flags);
265 
266  return buf;
267 }
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:331
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

◆ _hash_getcachedmetap()

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

Definition at line 1509 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().

1510 {
1511  Page page;
1512 
1513  Assert(metabuf);
1514  if (force_refresh || rel->rd_amcache == NULL)
1515  {
1516  char *cache = NULL;
1517 
1518  /*
1519  * It's important that we don't set rd_amcache to an invalid value.
1520  * Either MemoryContextAlloc or _hash_getbuf could fail, so don't
1521  * install a pointer to the newly-allocated storage in the actual
1522  * relcache entry until both have succeeeded.
1523  */
1524  if (rel->rd_amcache == NULL)
1525  cache = MemoryContextAlloc(rel->rd_indexcxt,
1526  sizeof(HashMetaPageData));
1527 
1528  /* Read the metapage. */
1529  if (BufferIsValid(*metabuf))
1530  LockBuffer(*metabuf, BUFFER_LOCK_SHARE);
1531  else
1532  *metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ,
1533  LH_META_PAGE);
1534  page = BufferGetPage(*metabuf);
1535 
1536  /* Populate the cache. */
1537  if (rel->rd_amcache == NULL)
1538  rel->rd_amcache = cache;
1539  memcpy(rel->rd_amcache, HashPageGetMeta(page),
1540  sizeof(HashMetaPageData));
1541 
1542  /* Release metapage lock, but keep the pin. */
1543  LockBuffer(*metabuf, BUFFER_LOCK_UNLOCK);
1544  }
1545 
1546  return (HashMetaPage) rel->rd_amcache;
1547 }
#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:79
#define HASH_READ
Definition: hash.h:329
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define HASH_METAPAGE
Definition: hash.h:206
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
#define Assert(condition)
Definition: c.h:699
#define BufferIsValid(bufnum)
Definition: bufmgr.h:114
#define HashPageGetMeta(page)
Definition: hash.h:313
void * MemoryContextAlloc(MemoryContext context, Size size)
Definition: mcxt.c:771
MemoryContext rd_indexcxt
Definition: rel.h:151
#define BUFFER_LOCK_SHARE
Definition: bufmgr.h:88
void * rd_amcache
Definition: rel.h:164
Pointer Page
Definition: bufpage.h:74

◆ _hash_getinitbuf()

Buffer _hash_getinitbuf ( Relation  rel,
BlockNumber  blkno 
)

Definition at line 144 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().

145 {
146  Buffer buf;
147 
148  if (blkno == P_NEW)
149  elog(ERROR, "hash AM does not use P_NEW");
150 
152  NULL);
153 
154  /* ref count and lock type are correct */
155 
156  /* initialize the page */
158 
159  return buf;
160 }
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:608
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

◆ _hash_getnewbuf()

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

Definition at line 207 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().

208 {
209  BlockNumber nblocks = RelationGetNumberOfBlocksInFork(rel, forkNum);
210  Buffer buf;
211 
212  if (blkno == P_NEW)
213  elog(ERROR, "hash AM does not use P_NEW");
214  if (blkno > nblocks)
215  elog(ERROR, "access to noncontiguous page in hash index \"%s\"",
217 
218  /* smgr insists we use P_NEW to extend the relation */
219  if (blkno == nblocks)
220  {
221  buf = ReadBufferExtended(rel, forkNum, P_NEW, RBM_NORMAL, NULL);
222  if (BufferGetBlockNumber(buf) != blkno)
223  elog(ERROR, "unexpected hash relation size: %u, should be %u",
224  BufferGetBlockNumber(buf), blkno);
225  LockBuffer(buf, HASH_WRITE);
226  }
227  else
228  {
229  buf = ReadBufferExtended(rel, forkNum, blkno, RBM_ZERO_AND_LOCK,
230  NULL);
231  }
232 
233  /* ref count and lock type are correct */
234 
235  /* initialize the page */
237 
238  return buf;
239 }
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:608
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:330
#define RelationGetRelationName(relation)
Definition: rel.h:441
#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

◆ _hash_hashkey2bucket()

Bucket _hash_hashkey2bucket ( uint32  hashkey,
uint32  maxbucket,
uint32  highmask,
uint32  lowmask 
)

Definition at line 125 of file hashutil.c.

Referenced by _h_indexbuild(), _hash_getbucketbuf_from_hashkey(), _hash_splitbucket(), comparetup_index_hash(), and hashbucketcleanup().

127 {
128  Bucket bucket;
129 
130  bucket = hashkey & highmask;
131  if (bucket > maxbucket)
132  bucket = bucket & lowmask;
133 
134  return bucket;
135 }
uint32 Bucket
Definition: hash.h:34

◆ _hash_init()

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

Definition at line 336 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().

337 {
338  Buffer metabuf;
339  Buffer buf;
340  Buffer bitmapbuf;
341  Page pg;
342  HashMetaPage metap;
343  RegProcedure procid;
344  int32 data_width;
345  int32 item_width;
346  int32 ffactor;
347  uint32 num_buckets;
348  uint32 i;
349  bool use_wal;
350 
351  /* safety check */
352  if (RelationGetNumberOfBlocksInFork(rel, forkNum) != 0)
353  elog(ERROR, "cannot initialize non-empty hash index \"%s\"",
355 
356  /*
357  * WAL log creation of pages if the relation is persistent, or this is the
358  * init fork. Init forks for unlogged relations always need to be WAL
359  * logged.
360  */
361  use_wal = RelationNeedsWAL(rel) || forkNum == INIT_FORKNUM;
362 
363  /*
364  * Determine the target fill factor (in tuples per bucket) for this index.
365  * The idea is to make the fill factor correspond to pages about as full
366  * as the user-settable fillfactor parameter says. We can compute it
367  * exactly since the index datatype (i.e. uint32 hash key) is fixed-width.
368  */
369  data_width = sizeof(uint32);
370  item_width = MAXALIGN(sizeof(IndexTupleData)) + MAXALIGN(data_width) +
371  sizeof(ItemIdData); /* include the line pointer */
372  ffactor = RelationGetTargetPageUsage(rel, HASH_DEFAULT_FILLFACTOR) / item_width;
373  /* keep to a sane range */
374  if (ffactor < 10)
375  ffactor = 10;
376 
377  procid = index_getprocid(rel, 1, HASHSTANDARD_PROC);
378 
379  /*
380  * We initialize the metapage, the first N bucket pages, and the first
381  * bitmap page in sequence, using _hash_getnewbuf to cause smgrextend()
382  * calls to occur. This ensures that the smgr level has the right idea of
383  * the physical index length.
384  *
385  * Critical section not required, because on error the creation of the
386  * whole relation will be rolled back.
387  */
388  metabuf = _hash_getnewbuf(rel, HASH_METAPAGE, forkNum);
389  _hash_init_metabuffer(metabuf, num_tuples, procid, ffactor, false);
390  MarkBufferDirty(metabuf);
391 
392  pg = BufferGetPage(metabuf);
393  metap = HashPageGetMeta(pg);
394 
395  /* XLOG stuff */
396  if (use_wal)
397  {
399  XLogRecPtr recptr;
400 
401  xlrec.num_tuples = num_tuples;
402  xlrec.procid = metap->hashm_procid;
403  xlrec.ffactor = metap->hashm_ffactor;
404 
405  XLogBeginInsert();
406  XLogRegisterData((char *) &xlrec, SizeOfHashInitMetaPage);
408 
409  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_INIT_META_PAGE);
410 
411  PageSetLSN(BufferGetPage(metabuf), recptr);
412  }
413 
414  num_buckets = metap->hashm_maxbucket + 1;
415 
416  /*
417  * Release buffer lock on the metapage while we initialize buckets.
418  * Otherwise, we'll be in interrupt holdoff and the CHECK_FOR_INTERRUPTS
419  * won't accomplish anything. It's a bad idea to hold buffer locks for
420  * long intervals in any case, since that can block the bgwriter.
421  */
422  LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
423 
424  /*
425  * Initialize and WAL Log the first N buckets
426  */
427  for (i = 0; i < num_buckets; i++)
428  {
429  BlockNumber blkno;
430 
431  /* Allow interrupts, in case N is huge */
433 
434  blkno = BUCKET_TO_BLKNO(metap, i);
435  buf = _hash_getnewbuf(rel, blkno, forkNum);
436  _hash_initbuf(buf, metap->hashm_maxbucket, i, LH_BUCKET_PAGE, false);
437  MarkBufferDirty(buf);
438 
439  if (use_wal)
440  log_newpage(&rel->rd_node,
441  forkNum,
442  blkno,
443  BufferGetPage(buf),
444  true);
445  _hash_relbuf(rel, buf);
446  }
447 
448  /* Now reacquire buffer lock on metapage */
450 
451  /*
452  * Initialize bitmap page
453  */
454  bitmapbuf = _hash_getnewbuf(rel, num_buckets + 1, forkNum);
455  _hash_initbitmapbuffer(bitmapbuf, metap->hashm_bmsize, false);
456  MarkBufferDirty(bitmapbuf);
457 
458  /* add the new bitmap page to the metapage's list of bitmaps */
459  /* metapage already has a write lock */
460  if (metap->hashm_nmaps >= HASH_MAX_BITMAPS)
461  ereport(ERROR,
462  (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
463  errmsg("out of overflow pages in hash index \"%s\"",
464  RelationGetRelationName(rel))));
465 
466  metap->hashm_mapp[metap->hashm_nmaps] = num_buckets + 1;
467 
468  metap->hashm_nmaps++;
469  MarkBufferDirty(metabuf);
470 
471  /* XLOG stuff */
472  if (use_wal)
473  {
475  XLogRecPtr recptr;
476 
477  xlrec.bmsize = metap->hashm_bmsize;
478 
479  XLogBeginInsert();
480  XLogRegisterData((char *) &xlrec, SizeOfHashInitBitmapPage);
481  XLogRegisterBuffer(0, bitmapbuf, REGBUF_WILL_INIT);
482 
483  /*
484  * This is safe only because nobody else can be modifying the index at
485  * this stage; it's only visible to the transaction that is creating
486  * it.
487  */
488  XLogRegisterBuffer(1, metabuf, REGBUF_STANDARD);
489 
490  recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_INIT_BITMAP_PAGE);
491 
492  PageSetLSN(BufferGetPage(bitmapbuf), recptr);
493  PageSetLSN(BufferGetPage(metabuf), recptr);
494  }
495 
496  /* all done */
497  _hash_relbuf(rel, bitmapbuf);
498  _hash_relbuf(rel, metabuf);
499 
500  return num_buckets;
501 }
#define HASH_DEFAULT_FILLFACTOR
Definition: hash.h:286
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:87
RegProcedure hashm_procid
Definition: hash.h:266
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:472
#define XLOG_HASH_INIT_BITMAP_PAGE
Definition: hash_xlog.h:28
uint16 hashm_ffactor
Definition: hash.h:254
#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:207
#define BUFFER_LOCK_EXCLUSIVE
Definition: bufmgr.h:89
signed int int32
Definition: c.h:313
#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:507
#define SizeOfHashInitBitmapPage
Definition: hash_xlog.h:252
uint32 hashm_nmaps
Definition: hash.h:265
static char * buf
Definition: pg_test_fsync.c:67
#define REGBUF_STANDARD
Definition: xloginsert.h:34
#define RelationGetRelationName(relation)
Definition: rel.h:441
unsigned int uint32
Definition: c.h:325
struct ItemIdData ItemIdData
#define XLOG_HASH_INIT_META_PAGE
Definition: hash_xlog.h:27
#define HASHSTANDARD_PROC
Definition: hash.h:351
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define ereport(elevel, rest)
Definition: elog.h:122
#define HASH_MAX_BITMAPS
Definition: hash.h:235
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:206
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:275
void _hash_initbitmapbuffer(Buffer buf, uint16 bmsize, bool initpage)
Definition: hashovfl.c:740
#define LH_BUCKET_PAGE
Definition: hash.h:65
BlockNumber RelationGetNumberOfBlocksInFork(Relation relation, ForkNumber forkNum)
Definition: bufmgr.c:2788
RelFileNode rd_node
Definition: rel.h:55
#define RelationGetTargetPageUsage(relation, defaultff)
Definition: rel.h:291
uint64 XLogRecPtr
Definition: xlogdefs.h:21
void _hash_initbuf(Buffer buf, uint32 max_bucket, uint32 num_bucket, uint32 flag, bool initpage)
Definition: hashpage.c:166
#define MAXALIGN(LEN)
Definition: c.h:652
#define RelationNeedsWAL(relation)
Definition: rel.h:510
uint32 hashm_maxbucket
Definition: hash.h:259
RegProcedure procid
Definition: hash_xlog.h:232
#define HashPageGetMeta(page)
Definition: hash.h:313
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:256
#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:269
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

◆ _hash_init_metabuffer()

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

Definition at line 507 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().

509 {
510  HashMetaPage metap;
511  HashPageOpaque pageopaque;
512  Page page;
513  double dnumbuckets;
514  uint32 num_buckets;
515  uint32 spare_index;
516  uint32 i;
517 
518  /*
519  * Choose the number of initial bucket pages to match the fill factor
520  * given the estimated number of tuples. We round up the result to the
521  * total number of buckets which has to be allocated before using its
522  * _hashm_spare element. However always force at least 2 bucket pages. The
523  * upper limit is determined by considerations explained in
524  * _hash_expandtable().
525  */
526  dnumbuckets = num_tuples / ffactor;
527  if (dnumbuckets <= 2.0)
528  num_buckets = 2;
529  else if (dnumbuckets >= (double) 0x40000000)
530  num_buckets = 0x40000000;
531  else
532  num_buckets = _hash_get_totalbuckets(_hash_spareindex(dnumbuckets));
533 
534  spare_index = _hash_spareindex(num_buckets);
535  Assert(spare_index < HASH_MAX_SPLITPOINTS);
536 
537  page = BufferGetPage(buf);
538  if (initpage)
540 
541  pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
542  pageopaque->hasho_prevblkno = InvalidBlockNumber;
543  pageopaque->hasho_nextblkno = InvalidBlockNumber;
544  pageopaque->hasho_bucket = -1;
545  pageopaque->hasho_flag = LH_META_PAGE;
546  pageopaque->hasho_page_id = HASHO_PAGE_ID;
547 
548  metap = HashPageGetMeta(page);
549 
550  metap->hashm_magic = HASH_MAGIC;
551  metap->hashm_version = HASH_VERSION;
552  metap->hashm_ntuples = 0;
553  metap->hashm_nmaps = 0;
554  metap->hashm_ffactor = ffactor;
555  metap->hashm_bsize = HashGetMaxBitmapSize(page);
556  /* find largest bitmap array size that will fit in page size */
557  for (i = _hash_log2(metap->hashm_bsize); i > 0; --i)
558  {
559  if ((1 << i) <= metap->hashm_bsize)
560  break;
561  }
562  Assert(i > 0);
563  metap->hashm_bmsize = 1 << i;
564  metap->hashm_bmshift = i + BYTE_TO_BIT;
565  Assert((1 << BMPG_SHIFT(metap)) == (BMPG_MASK(metap) + 1));
566 
567  /*
568  * Label the index with its primary hash support function's OID. This is
569  * pretty useless for normal operation (in fact, hashm_procid is not used
570  * anywhere), but it might be handy for forensic purposes so we keep it.
571  */
572  metap->hashm_procid = procid;
573 
574  /*
575  * We initialize the index with N buckets, 0 .. N-1, occupying physical
576  * blocks 1 to N. The first freespace bitmap page is in block N+1.
577  */
578  metap->hashm_maxbucket = num_buckets - 1;
579 
580  /*
581  * Set highmask as next immediate ((2 ^ x) - 1), which should be
582  * sufficient to cover num_buckets.
583  */
584  metap->hashm_highmask = (1 << (_hash_log2(num_buckets + 1))) - 1;
585  metap->hashm_lowmask = (metap->hashm_highmask >> 1);
586 
587  MemSet(metap->hashm_spares, 0, sizeof(metap->hashm_spares));
588  MemSet(metap->hashm_mapp, 0, sizeof(metap->hashm_mapp));
589 
590  /* Set up mapping for one spare page after the initial splitpoints */
591  metap->hashm_spares[spare_index] = 1;
592  metap->hashm_ovflpoint = spare_index;
593  metap->hashm_firstfree = 0;
594 
595  /*
596  * Set pd_lower just past the end of the metadata. This is essential,
597  * because without doing so, metadata will be lost if xlog.c compresses
598  * the page.
599  */
600  ((PageHeader) page)->pd_lower =
601  ((char *) metap + sizeof(HashMetaPageData)) - (char *) page;
602 }
uint16 hashm_bmshift
Definition: hash.h:258
uint16 hasho_page_id
Definition: hash.h:93
RegProcedure hashm_procid
Definition: hash.h:266
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:608
#define LH_META_PAGE
Definition: hash.h:67
#define BYTE_TO_BIT
Definition: hash.h:291
uint32 hashm_magic
Definition: hash.h:251
uint16 hashm_ffactor
Definition: hash.h:254
uint32 hashm_highmask
Definition: hash.h:260
#define MemSet(start, val, len)
Definition: c.h:908
#define HASH_VERSION
Definition: hash.h:209
uint32 hashm_lowmask
Definition: hash.h:261
#define HASH_MAX_SPLITPOINTS
Definition: hash.h:244
#define HASH_MAGIC
Definition: hash.h:208
#define HashGetMaxBitmapSize(page)
Definition: hash.h:309
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:252
uint32 hashm_nmaps
Definition: hash.h:265
static char * buf
Definition: pg_test_fsync.c:67
#define BMPG_MASK(metap)
Definition: hash.h:304
unsigned int uint32
Definition: c.h:325
#define BMPG_SHIFT(metap)
Definition: hash.h:303
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
uint32 hashm_ovflpoint
Definition: hash.h:262
uint16 hashm_bsize
Definition: hash.h:255
double hashm_ntuples
Definition: hash.h:253
#define BufferGetPageSize(buffer)
Definition: bufmgr.h:147
uint32 hashm_firstfree
Definition: hash.h:264
uint32 hashm_spares[HASH_MAX_SPLITPOINTS]
Definition: hash.h:267
PageHeaderData * PageHeader
Definition: bufpage.h:162
#define Assert(condition)
Definition: c.h:699
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:259
uint16 hasho_flag
Definition: hash.h:92
#define HashPageGetMeta(page)
Definition: hash.h:313
uint32 _hash_log2(uint32 num)
Definition: hashutil.c:141
int i
BlockNumber hasho_nextblkno
Definition: hash.h:90
uint16 hashm_bmsize
Definition: hash.h:256
BlockNumber hashm_mapp[HASH_MAX_BITMAPS]
Definition: hash.h:269
Pointer Page
Definition: bufpage.h:74

◆ _hash_initbitmapbuffer()

void _hash_initbitmapbuffer ( Buffer  buf,
uint16  bmsize,
bool  initpage 
)

Definition at line 740 of file hashovfl.c.

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

Referenced by _hash_addovflpage(), _hash_init(), hash_xlog_add_ovfl_page(), and hash_xlog_init_bitmap_page().

741 {
742  Page pg;
743  HashPageOpaque op;
744  uint32 *freep;
745 
746  pg = BufferGetPage(buf);
747 
748  /* initialize the page */
749  if (initpage)
751 
752  /* initialize the page's special space */
756  op->hasho_bucket = -1;
759 
760  /* set all of the bits to 1 */
761  freep = HashPageGetBitmap(pg);
762  MemSet(freep, 0xFF, bmsize);
763 
764  /*
765  * Set pd_lower just past the end of the bitmap page data. We could even
766  * set pd_lower equal to pd_upper, but this is more precise and makes the
767  * page look compressible to xlog.c.
768  */
769  ((PageHeader) pg)->pd_lower = ((char *) freep + bmsize) - (char *) pg;
770 }
uint16 hasho_page_id
Definition: hash.h:93
#define HashPageGetBitmap(page)
Definition: hash.h:306
#define LH_BITMAP_PAGE
Definition: hash.h:66
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:608
#define MemSet(start, val, len)
Definition: c.h:908
BlockNumber hasho_prevblkno
Definition: hash.h:89
static char * buf
Definition: pg_test_fsync.c:67
unsigned int uint32
Definition: c.h:325
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
#define BufferGetPageSize(buffer)
Definition: bufmgr.h:147
PageHeaderData * PageHeader
Definition: bufpage.h:162
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

◆ _hash_initbuf()

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

Definition at line 166 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().

168 {
169  HashPageOpaque pageopaque;
170  Page page;
171 
172  page = BufferGetPage(buf);
173 
174  /* initialize the page */
175  if (initpage)
177 
178  pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
179 
180  /*
181  * Set hasho_prevblkno with current hashm_maxbucket. This value will be
182  * used to validate cached HashMetaPageData. See
183  * _hash_getbucketbuf_from_hashkey().
184  */
185  pageopaque->hasho_prevblkno = max_bucket;
186  pageopaque->hasho_nextblkno = InvalidBlockNumber;
187  pageopaque->hasho_bucket = num_bucket;
188  pageopaque->hasho_flag = flag;
189  pageopaque->hasho_page_id = HASHO_PAGE_ID;
190 }
uint16 hasho_page_id
Definition: hash.h:93
void _hash_pageinit(Page page, Size size)
Definition: hashpage.c:608
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

◆ _hash_kill_items()

void _hash_kill_items ( IndexScanDesc  scan)

Definition at line 544 of file hashutil.c.

References _hash_getbuf(), _hash_relbuf(), Assert, buf, HashScanPosData::buf, BUFFER_LOCK_SHARE, BUFFER_LOCK_UNLOCK, BufferGetPage, HashScanPosData::currPage, HashScanOpaqueData::currPos, HashScanPosData::firstItem, HASH_READ, HashScanPosIsPinned, HashScanPosIsValid, HashScanOpaqueData::hashso_bucket_buf, HashScanPosItem::heapTid, i, HashScanPosItem::indexOffset, IndexScanDescData::indexRelation, ItemIdMarkDead, ItemPointerEquals(), HashScanPosData::items, HashScanOpaqueData::killedItems, LH_OVERFLOW_PAGE, LH_PAGE_HAS_DEAD_TUPLES, LockBuffer(), MarkBufferDirtyHint(), HashScanOpaqueData::numKilled, OffsetNumberNext, IndexScanDescData::opaque, PageGetItem, PageGetItemId, PageGetMaxOffsetNumber, PageGetSpecialPointer, and IndexTupleData::t_tid.

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

545 {
546  HashScanOpaque so = (HashScanOpaque) scan->opaque;
547  Relation rel = scan->indexRelation;
548  BlockNumber blkno;
549  Buffer buf;
550  Page page;
551  HashPageOpaque opaque;
552  OffsetNumber offnum,
553  maxoff;
554  int numKilled = so->numKilled;
555  int i;
556  bool killedsomething = false;
557  bool havePin = false;
558 
559  Assert(so->numKilled > 0);
560  Assert(so->killedItems != NULL);
562 
563  /*
564  * Always reset the scan state, so we don't look for same items on other
565  * pages.
566  */
567  so->numKilled = 0;
568 
569  blkno = so->currPos.currPage;
570  if (HashScanPosIsPinned(so->currPos))
571  {
572  /*
573  * We already have pin on this buffer, so, all we need to do is
574  * acquire lock on it.
575  */
576  havePin = true;
577  buf = so->currPos.buf;
579  }
580  else
581  buf = _hash_getbuf(rel, blkno, HASH_READ, LH_OVERFLOW_PAGE);
582 
583  page = BufferGetPage(buf);
584  opaque = (HashPageOpaque) PageGetSpecialPointer(page);
585  maxoff = PageGetMaxOffsetNumber(page);
586 
587  for (i = 0; i < numKilled; i++)
588  {
589  int itemIndex = so->killedItems[i];
590  HashScanPosItem *currItem = &so->currPos.items[itemIndex];
591 
592  offnum = currItem->indexOffset;
593 
594  Assert(itemIndex >= so->currPos.firstItem &&
595  itemIndex <= so->currPos.lastItem);
596 
597  while (offnum <= maxoff)
598  {
599  ItemId iid = PageGetItemId(page, offnum);
600  IndexTuple ituple = (IndexTuple) PageGetItem(page, iid);
601 
602  if (ItemPointerEquals(&ituple->t_tid, &currItem->heapTid))
603  {
604  /* found the item */
605  ItemIdMarkDead(iid);
606  killedsomething = true;
607  break; /* out of inner search loop */
608  }
609  offnum = OffsetNumberNext(offnum);
610  }
611  }
612 
613  /*
614  * Since this can be redone later if needed, mark as dirty hint. Whenever
615  * we mark anything LP_DEAD, we also set the page's
616  * LH_PAGE_HAS_DEAD_TUPLES flag, which is likewise just a hint.
617  */
618  if (killedsomething)
619  {
620  opaque->hasho_flag |= LH_PAGE_HAS_DEAD_TUPLES;
621  MarkBufferDirtyHint(buf, true);
622  }
623 
624  if (so->hashso_bucket_buf == so->currPos.buf ||
625  havePin)
627  else
628  _hash_relbuf(rel, buf);
629 }
HashScanOpaqueData * HashScanOpaque
Definition: hash.h:200
#define BUFFER_LOCK_UNLOCK
Definition: bufmgr.h:87
void MarkBufferDirtyHint(Buffer buffer, bool buffer_std)
Definition: bufmgr.c:3379
#define ItemIdMarkDead(itemId)
Definition: itemid.h:178
#define HashScanPosIsPinned(scanpos)
Definition: hash.h:138
#define HashScanPosIsValid(scanpos)
Definition: hash.h:145
ItemPointerData t_tid
Definition: itup.h:37
uint32 BlockNumber
Definition: block.h:31
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:79
#define PageGetMaxOffsetNumber(page)
Definition: bufpage.h:353
Relation indexRelation
Definition: relscan.h:91
uint16 OffsetNumber
Definition: off.h:24
#define HASH_READ
Definition: hash.h:329
int * killedItems
Definition: hash.h:190
static char * buf
Definition: pg_test_fsync.c:67
IndexTupleData * IndexTuple
Definition: itup.h:53
BlockNumber currPage
Definition: hash.h:120
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
Buffer hashso_bucket_buf
Definition: hash.h:172
#define PageGetItemId(page, offsetNumber)
Definition: bufpage.h:231
Buffer buf
Definition: hash.h:119
#define LH_OVERFLOW_PAGE
Definition: hash.h:64
void LockBuffer(Buffer buffer, int mode)
Definition: bufmgr.c:3546
void _hash_relbuf(Relation rel, Buffer buf)
Definition: hashpage.c:275
#define Assert(condition)
Definition: c.h:699
#define OffsetNumberNext(offsetNumber)
Definition: off.h:53
#define PageGetSpecialPointer(page)
Definition: bufpage.h:322
HashPageOpaqueData * HashPageOpaque
Definition: hash.h:96
bool ItemPointerEquals(ItemPointer pointer1, ItemPointer pointer2)
Definition: itemptr.c:29
HashScanPosData currPos
Definition: hash.h:197
OffsetNumber indexOffset
Definition: hash.h:114
int i
#define BUFFER_LOCK_SHARE
Definition: bufmgr.h:88
ItemPointerData heapTid
Definition: hash.h:113
int firstItem
Definition: hash.h:131
int Buffer
Definition: buf.h:23
#define LH_PAGE_HAS_DEAD_TUPLES
Definition: hash.h:71
#define PageGetItem(page, itemId)
Definition: bufpage.h:336
Pointer Page
Definition: bufpage.h:74
HashScanPosItem items[MaxIndexTuplesPerPage]
Definition: hash.h:135

◆ _hash_log2()

uint32 _hash_log2 ( uint32  num)

Definition at line 141 of file hashutil.c.

References i.

Referenced by _h_spoolinit(), _hash_init_metabuffer(), and _hash_spareindex().

142 {
143  uint32 i,
144  limit;
145 
146  limit = 1;
147  for (i = 0; limit < num; limit <<= 1, i++)
148  ;
149  return i;
150 }
unsigned int uint32
Definition: c.h:325
int i

◆ _hash_next()

bool _hash_next ( IndexScanDesc  scan,
ScanDirection  dir 
)

Definition at line 48 of file hashsearch.c.

References _hash_dropbuf(), _hash_dropscanbuf(), _hash_getbuf(), _hash_kill_items(), _hash_readpage(), BlockNumberIsValid, buf, BufferGetPage, HashScanOpaqueData::currPos, HashScanPosData::firstItem, HASH_READ, HashScanPosInvalidate, HashScanOpaqueData::hashso_bucket_buf, HashScanOpaqueData::hashso_split_bucket_buf, IndexScanDescData::indexRelation, HashScanPosData::itemIndex, HashScanPosData::items, HashScanPosData::lastItem, LH_BUCKET_PAGE, LH_OVERFLOW_PAGE, HashScanPosData::nextPage, HashScanOpaqueData::numKilled, IndexScanDescData::opaque, HashScanPosData::prevPage, ScanDirectionIsForward, HeapTupleData::t_self, TestForOldSnapshot(), IndexScanDescData::xs_ctup, and IndexScanDescData::xs_snapshot.

Referenced by hashgetbitmap(), and hashgettuple().

49 {
50  Relation rel = scan->indexRelation;
52  HashScanPosItem *currItem;
53  BlockNumber blkno;
54  Buffer buf;
55  bool end_of_scan = false;
56 
57  /*
58  * Advance to the next tuple on the current page; or if done, try to read
59  * data from the next or previous page based on the scan direction. Before
60  * moving to the next or previous page make sure that we deal with all the
61  * killed items.
62  */
63  if (ScanDirectionIsForward(dir))
64  {
65  if (++so->currPos.itemIndex > so->currPos.lastItem)
66  {
67  if (so->numKilled > 0)
68  _hash_kill_items(scan);
69 
70  blkno = so->currPos.nextPage;
71  if (BlockNumberIsValid(blkno))
72  {
75  if (!_hash_readpage(scan, &buf, dir))
76  end_of_scan = true;
77  }
78  else
79  end_of_scan = true;
80  }
81  }
82  else
83  {
84  if (--so->currPos.itemIndex < so->currPos.firstItem)
85  {
86  if (so->numKilled > 0)
87  _hash_kill_items(scan);
88 
89  blkno = so->currPos.prevPage;
90  if (BlockNumberIsValid(blkno))
91  {
92  buf = _hash_getbuf(rel, blkno, HASH_READ,
95 
96  /*
97  * We always maintain the pin on bucket page for whole scan
98  * operation, so releasing the additional pin we have acquired
99  * here.
100  */
101  if (buf == so->hashso_bucket_buf ||
103  _hash_dropbuf(rel, buf);
104 
105  if (!_hash_readpage(scan, &buf, dir))
106  end_of_scan = true;
107  }
108  else
109  end_of_scan = true;
110  }
111  }
112 
113  if (end_of_scan)
114  {
115  _hash_dropscanbuf(rel, so);
117  return false;
118  }
119 
120  /* OK, itemIndex says what to return */
121  currItem = &so->currPos.items[so->currPos.itemIndex];
122  scan->xs_ctup.t_self = currItem->heapTid;
123 
124  return true;
125 }
HashScanOpaqueData * HashScanOpaque
Definition: hash.h:200
static void TestForOldSnapshot(Snapshot snapshot, Relation relation, Page page)
Definition: bufmgr.h:265
void _hash_dropscanbuf(Relation rel, HashScanOpaque so)
Definition: hashpage.c:298
static bool _hash_readpage(IndexScanDesc scan, Buffer *bufP, ScanDirection dir)
Definition: hashsearch.c:452
#define ScanDirectionIsForward(direction)
Definition: sdir.h:55
Snapshot xs_snapshot
Definition: relscan.h:92
uint32 BlockNumber
Definition: block.h:31
void _hash_dropbuf(Relation rel, Buffer buf)
Definition: hashpage.c:286
Buffer _hash_getbuf(Relation rel, BlockNumber blkno, int access, int flags)
Definition: hashpage.c:79
Relation indexRelation
Definition: relscan.h:91
#define HASH_READ
Definition: hash.h:329
ItemPointerData t_self
Definition: htup.h:65
static char * buf
Definition: pg_test_fsync.c:67
#define BufferGetPage(buffer)
Definition: bufmgr.h:160
Buffer hashso_bucket_buf
Definition: hash.h:172
#define HashScanPosInvalidate(scanpos)
Definition: hash.h:152
#define LH_OVERFLOW_PAGE
Definition: hash.h:64
#define BlockNumberIsValid(blockNumber)
Definition: block.h:70
#define LH_BUCKET_PAGE
Definition: hash.h:65
int lastItem
Definition: hash.h:132
HeapTupleData xs_ctup
Definition: relscan.h:121
void _hash_kill_items(IndexScanDesc scan)
Definition: hashutil.c:544
HashScanPosData currPos
Definition: hash.h:197
Buffer hashso_split_bucket_buf
Definition: hash.h:179
BlockNumber nextPage
Definition: hash.h:121
int itemIndex
Definition: hash.h:133
int firstItem
Definition: