/*
* _hash_get_oldblock_from_newbucket() -- get the block number of a bucket
* from which current (new) bucket is being split.
*/
BlockNumber
_hash_get_oldblock_from_newbucket(Relation rel, Bucket new_bucket)
{
Bucket old_bucket;
uint32 mask;
Buffer metabuf;
HashMetaPage metap;
BlockNumber blkno;
/*
* To get the old bucket from the current bucket, we need a mask to modulo
* into lower half of table. This mask is stored in meta page as
* hashm_lowmask, but here we can't rely on the same, because we need a
* value of lowmask that was prevalent at the time when bucket split was
* started. Masking the most significant bit of new bucket would give us
* old bucket.
*/
mask = (((uint32) 1) << (fls(new_bucket) - 1)) - 1;
old_bucket = new_bucket & mask;
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
blkno = BUCKET_TO_BLKNO(metap, old_bucket);
_hash_relbuf(rel, metabuf);
return blkno;
}
/*
* _hash_getbucketbuf_from_hashkey() -- Get the bucket's buffer for the given
* hashkey.
*
* Bucket pages do not move or get removed once they are allocated. This give
* us an opportunity to use the previously saved metapage contents to reach
* the target bucket buffer, instead of reading from the metapage every time.
* This saves one buffer access every time we want to reach the target bucket
* buffer, which is very helpful savings in bufmgr traffic and contention.
*
* The access type parameter (HASH_READ or HASH_WRITE) indicates whether the
* bucket buffer has to be locked for reading or writing.
*
* The out parameter cachedmetap is set with metapage contents used for
* hashkey to bucket buffer mapping. Some callers need this info to reach the
* old bucket in case of bucket split, see _hash_doinsert().
*/
Buffer
_hash_getbucketbuf_from_hashkey(Relation rel, uint32 hashkey, int access,
HashMetaPage *cachedmetap)
{
HashMetaPage metap;
Buffer buf;
Buffer metabuf = InvalidBuffer;
Page page;
Bucket bucket;
BlockNumber blkno;
HashPageOpaque opaque;
/* We read from target bucket buffer, hence locking is must. */
Assert(access == HASH_READ || access == HASH_WRITE);
metap = _hash_getcachedmetap(rel, &metabuf, false);
Assert(metap != NULL);
/*
* Loop until we get a lock on the correct target bucket.
*/
for (;;)
{
/*
* Compute the target bucket number, and convert to block number.
*/
bucket = _hash_hashkey2bucket(hashkey,
metap->hashm_maxbucket,
metap->hashm_highmask,
metap->hashm_lowmask);
blkno = BUCKET_TO_BLKNO(metap, bucket);
/* Fetch the primary bucket page for the bucket */
buf = _hash_getbuf(rel, blkno, access, LH_BUCKET_PAGE);
page = BufferGetPage(buf);
opaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(opaque->hasho_bucket == bucket);
Assert(opaque->hasho_prevblkno != InvalidBlockNumber);
/*
* If this bucket hasn't been split, we're done.
*/
if (opaque->hasho_prevblkno <= metap->hashm_maxbucket)
break;
/* Drop lock on this buffer, update cached metapage, and retry. */
_hash_relbuf(rel, buf);
metap = _hash_getcachedmetap(rel, &metabuf, true);
Assert(metap != NULL);
}
if (BufferIsValid(metabuf))
_hash_dropbuf(rel, metabuf);
if (cachedmetap)
*cachedmetap = metap;
return buf;
}
/*
* _hash_addovflpage
*
* Add an overflow page to the bucket whose last page is pointed to by 'buf'.
*
* On entry, the caller must hold a pin but no lock on 'buf'. The pin is
* dropped before exiting (we assume the caller is not interested in 'buf'
* anymore). The returned overflow page will be pinned and write-locked;
* it is guaranteed to be empty.
*
* The caller must hold a pin, but no lock, on the metapage buffer.
* That buffer is returned in the same state.
*
* The caller must hold at least share lock on the bucket, to ensure that
* no one else tries to compact the bucket meanwhile. This guarantees that
* 'buf' won't stop being part of the bucket while it's unlocked.
*
* NB: since this could be executed concurrently by multiple processes,
* one should not assume that the returned overflow page will be the
* immediate successor of the originally passed 'buf'. Additional overflow
* pages might have been added to the bucket chain in between.
*/
Buffer
_hash_addovflpage(Relation rel, Buffer metabuf, Buffer buf)
{
Buffer ovflbuf;
Page page;
Page ovflpage;
HashPageOpaque pageopaque;
HashPageOpaque ovflopaque;
/* allocate and lock an empty overflow page */
ovflbuf = _hash_getovflpage(rel, metabuf);
/*
* Write-lock the tail page. It is okay to hold two buffer locks here
* since there cannot be anyone else contending for access to ovflbuf.
*/
_hash_chgbufaccess(rel, buf, HASH_NOLOCK, HASH_WRITE);
/* probably redundant... */
_hash_checkpage(rel, buf, LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
/* loop to find current tail page, in case someone else inserted too */
for (;;)
{
BlockNumber nextblkno;
page = BufferGetPage(buf);
pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
nextblkno = pageopaque->hasho_nextblkno;
if (!BlockNumberIsValid(nextblkno))
break;
/* we assume we do not need to write the unmodified page */
_hash_relbuf(rel, buf);
buf = _hash_getbuf(rel, nextblkno, HASH_WRITE, LH_OVERFLOW_PAGE);
}
/* now that we have correct backlink, initialize new overflow page */
ovflpage = BufferGetPage(ovflbuf);
ovflopaque = (HashPageOpaque) PageGetSpecialPointer(ovflpage);
ovflopaque->hasho_prevblkno = BufferGetBlockNumber(buf);
ovflopaque->hasho_nextblkno = InvalidBlockNumber;
ovflopaque->hasho_bucket = pageopaque->hasho_bucket;
ovflopaque->hasho_flag = LH_OVERFLOW_PAGE;
ovflopaque->hasho_page_id = HASHO_PAGE_ID;
MarkBufferDirty(ovflbuf);
/* logically chain overflow page to previous page */
pageopaque->hasho_nextblkno = BufferGetBlockNumber(ovflbuf);
_hash_wrtbuf(rel, buf);
return ovflbuf;
}
/*
* Advance to next page in a bucket, if any.
*/
static void
_hash_readnext(Relation rel,
Buffer *bufp, Page *pagep, HashPageOpaque *opaquep)
{
BlockNumber blkno;
blkno = (*opaquep)->hasho_nextblkno;
_hash_relbuf(rel, *bufp);
*bufp = InvalidBuffer;
if (BlockNumberIsValid(blkno))
{
*bufp = _hash_getbuf(rel, blkno, HASH_READ, LH_OVERFLOW_PAGE);
*pagep = BufferGetPage(*bufp);
*opaquep = (HashPageOpaque) PageGetSpecialPointer(*pagep);
}
}
/*
* Advance to previous page in a bucket, if any.
*/
static void
_hash_readprev(struct relation *rel, buf_id_t * bufp, page_p * pagep, struct hash_page * *opaquep)
{
block_t blkno;
blkno = (*opaquep)->hasho_prevblkno;
_hash_relbuf(rel, *bufp);
*bufp = INVALID_BUF;
/* check for interrupts while we're not holding any buffer lock */
CHECK_FOR_INTERRUPTS();
if (BLK_NR_VALID(blkno)) {
*bufp = _hash_getbuf(rel, blkno, HASH_READ, LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
*pagep = BUF_PAGE(*bufp);
*opaquep = (struct hash_page *) PAGE_SPECIAL_PTR(*pagep);
}
}
/*
* Advance to next page in a bucket, if any.
*/
static void
_hash_readnext(Relation rel,
Buffer *bufp, Page *pagep, HashPageOpaque *opaquep)
{
BlockNumber blkno;
blkno = (*opaquep)->hasho_nextblkno;
_hash_relbuf(rel, *bufp);
*bufp = InvalidBuffer;
/* check for interrupts while we're not holding any buffer lock */
CHECK_FOR_INTERRUPTS();
if (BlockNumberIsValid(blkno))
{
*bufp = _hash_getbuf(rel, blkno, HASH_READ, LH_OVERFLOW_PAGE);
*pagep = BufferGetPage(*bufp);
*opaquep = (HashPageOpaque) PageGetSpecialPointer(*pagep);
}
}
/*
* _hash_get_newblock_from_oldbucket() -- get the block number of a bucket
* that will be generated after split from old bucket.
*
* This is used to find the new bucket from old bucket based on current table
* half. It is mainly required to finish the incomplete splits where we are
* sure that not more than one bucket could have split in progress from old
* bucket.
*/
BlockNumber
_hash_get_newblock_from_oldbucket(Relation rel, Bucket old_bucket)
{
Bucket new_bucket;
Buffer metabuf;
HashMetaPage metap;
BlockNumber blkno;
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
new_bucket = _hash_get_newbucket_from_oldbucket(rel, old_bucket,
metap->hashm_lowmask,
metap->hashm_maxbucket);
blkno = BUCKET_TO_BLKNO(metap, new_bucket);
_hash_relbuf(rel, metabuf);
return blkno;
}
/*
* Advance to previous page in a bucket, if any.
*/
static void
_hash_readprev(Relation rel,
Buffer *bufp, Page *pagep, HashPageOpaque *opaquep)
{
BlockNumber blkno;
MIRROREDLOCK_BUFMGR_MUST_ALREADY_BE_HELD;
blkno = (*opaquep)->hasho_prevblkno;
_hash_relbuf(rel, *bufp);
*bufp = InvalidBuffer;
/* check for interrupts while we're not holding any buffer lock */
CHECK_FOR_INTERRUPTS();
if (BlockNumberIsValid(blkno))
{
*bufp = _hash_getbuf(rel, blkno, HASH_READ);
_hash_checkpage(rel, *bufp, LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
*pagep = BufferGetPage(*bufp);
*opaquep = (HashPageOpaque) PageGetSpecialPointer(*pagep);
}
}
/*
* pgstat_hash_page -- check tuples in a hash page
*/
static void
pgstat_hash_page(pgstattuple_type *stat, Relation rel, BlockNumber blkno,
BufferAccessStrategy bstrategy)
{
Buffer buf;
Page page;
_hash_getlock(rel, blkno, HASH_SHARE);
buf = _hash_getbuf_with_strategy(rel, blkno, HASH_READ, 0, bstrategy);
page = BufferGetPage(buf);
if (PageGetSpecialSize(page) == MAXALIGN(sizeof(HashPageOpaqueData)))
{
HashPageOpaque opaque;
opaque = (HashPageOpaque) PageGetSpecialPointer(page);
switch (opaque->hasho_flag)
{
case LH_UNUSED_PAGE:
stat->free_space += BLCKSZ;
break;
case LH_BUCKET_PAGE:
case LH_OVERFLOW_PAGE:
pgstat_index_page(stat, page, FirstOffsetNumber,
PageGetMaxOffsetNumber(page));
break;
case LH_BITMAP_PAGE:
case LH_META_PAGE:
default:
break;
}
}
else
{
/* maybe corrupted */
}
_hash_relbuf(rel, buf);
_hash_droplock(rel, blkno, HASH_SHARE);
}
/*
* _hash_squeezebucket(rel, bucket)
*
* Try to squeeze the tuples onto pages occurring earlier in the
* bucket chain in an attempt to free overflow pages. When we start
* the "squeezing", the page from which we start taking tuples (the
* "read" page) is the last bucket in the bucket chain and the page
* onto which we start squeezing tuples (the "write" page) is the
* first page in the bucket chain. The read page works backward and
* the write page works forward; the procedure terminates when the
* read page and write page are the same page.
*
* At completion of this procedure, it is guaranteed that all pages in
* the bucket are nonempty, unless the bucket is totally empty (in
* which case all overflow pages will be freed). The original implementation
* required that to be true on entry as well, but it's a lot easier for
* callers to leave empty overflow pages and let this guy clean it up.
*
* Caller must acquire cleanup lock on the primary page of the target
* bucket to exclude any scans that are in progress, which could easily
* be confused into returning the same tuple more than once or some tuples
* not at all by the rearrangement we are performing here. To prevent
* any concurrent scan to cross the squeeze scan we use lock chaining
* similar to hasbucketcleanup. Refer comments atop hashbucketcleanup.
*
* We need to retain a pin on the primary bucket to ensure that no concurrent
* split can start.
*
* Since this function is invoked in VACUUM, we provide an access strategy
* parameter that controls fetches of the bucket pages.
*/
void
_hash_squeezebucket(Relation rel,
Bucket bucket,
BlockNumber bucket_blkno,
Buffer bucket_buf,
BufferAccessStrategy bstrategy)
{
BlockNumber wblkno;
BlockNumber rblkno;
Buffer wbuf;
Buffer rbuf;
Page wpage;
Page rpage;
HashPageOpaque wopaque;
HashPageOpaque ropaque;
/*
* start squeezing into the primary bucket page.
*/
wblkno = bucket_blkno;
wbuf = bucket_buf;
wpage = BufferGetPage(wbuf);
wopaque = (HashPageOpaque) PageGetSpecialPointer(wpage);
/*
* if there aren't any overflow pages, there's nothing to squeeze. caller
* is responsible for releasing the pin on primary bucket page.
*/
if (!BlockNumberIsValid(wopaque->hasho_nextblkno))
{
LockBuffer(wbuf, BUFFER_LOCK_UNLOCK);
return;
}
/*
* Find the last page in the bucket chain by starting at the base bucket
* page and working forward. Note: we assume that a hash bucket chain is
* usually smaller than the buffer ring being used by VACUUM, else using
* the access strategy here would be counterproductive.
*/
rbuf = InvalidBuffer;
ropaque = wopaque;
do
{
rblkno = ropaque->hasho_nextblkno;
if (rbuf != InvalidBuffer)
_hash_relbuf(rel, rbuf);
rbuf = _hash_getbuf_with_strategy(rel,
rblkno,
HASH_WRITE,
LH_OVERFLOW_PAGE,
bstrategy);
rpage = BufferGetPage(rbuf);
ropaque = (HashPageOpaque) PageGetSpecialPointer(rpage);
Assert(ropaque->hasho_bucket == bucket);
} while (BlockNumberIsValid(ropaque->hasho_nextblkno));
/*
* squeeze the tuples.
*/
for (;;)
{
OffsetNumber roffnum;
OffsetNumber maxroffnum;
OffsetNumber deletable[MaxOffsetNumber];
IndexTuple itups[MaxIndexTuplesPerPage];
Size tups_size[MaxIndexTuplesPerPage];
OffsetNumber itup_offsets[MaxIndexTuplesPerPage];
uint16 ndeletable = 0;
uint16 nitups = 0;
Size all_tups_size = 0;
int i;
bool retain_pin = false;
readpage:
/* Scan each tuple in "read" page */
maxroffnum = PageGetMaxOffsetNumber(rpage);
for (roffnum = FirstOffsetNumber;
roffnum <= maxroffnum;
roffnum = OffsetNumberNext(roffnum))
{
IndexTuple itup;
Size itemsz;
/* skip dead tuples */
if (ItemIdIsDead(PageGetItemId(rpage, roffnum)))
continue;
itup = (IndexTuple) PageGetItem(rpage,
PageGetItemId(rpage, roffnum));
itemsz = IndexTupleDSize(*itup);
itemsz = MAXALIGN(itemsz);
/*
* Walk up the bucket chain, looking for a page big enough for
* this item and all other accumulated items. Exit if we reach
* the read page.
*/
while (PageGetFreeSpaceForMultipleTuples(wpage, nitups + 1) < (all_tups_size + itemsz))
{
Buffer next_wbuf = InvalidBuffer;
bool tups_moved = false;
Assert(!PageIsEmpty(wpage));
if (wblkno == bucket_blkno)
retain_pin = true;
wblkno = wopaque->hasho_nextblkno;
Assert(BlockNumberIsValid(wblkno));
/* don't need to move to next page if we reached the read page */
if (wblkno != rblkno)
next_wbuf = _hash_getbuf_with_strategy(rel,
wblkno,
HASH_WRITE,
LH_OVERFLOW_PAGE,
bstrategy);
if (nitups > 0)
{
Assert(nitups == ndeletable);
/*
* This operation needs to log multiple tuples, prepare
* WAL for that.
*/
if (RelationNeedsWAL(rel))
XLogEnsureRecordSpace(0, 3 + nitups);
START_CRIT_SECTION();
/*
* we have to insert tuples on the "write" page, being
* careful to preserve hashkey ordering. (If we insert
* many tuples into the same "write" page it would be
* worth qsort'ing them).
*/
_hash_pgaddmultitup(rel, wbuf, itups, itup_offsets, nitups);
MarkBufferDirty(wbuf);
/* Delete tuples we already moved off read page */
PageIndexMultiDelete(rpage, deletable, ndeletable);
MarkBufferDirty(rbuf);
/* XLOG stuff */
if (RelationNeedsWAL(rel))
{
XLogRecPtr recptr;
xl_hash_move_page_contents xlrec;
xlrec.ntups = nitups;
xlrec.is_prim_bucket_same_wrt = (wbuf == bucket_buf) ? true : false;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHashMovePageContents);
/*
* bucket buffer needs to be registered to ensure that
* we can acquire a cleanup lock on it during replay.
*/
if (!xlrec.is_prim_bucket_same_wrt)
XLogRegisterBuffer(0, bucket_buf, REGBUF_STANDARD | REGBUF_NO_IMAGE);
XLogRegisterBuffer(1, wbuf, REGBUF_STANDARD);
XLogRegisterBufData(1, (char *) itup_offsets,
nitups * sizeof(OffsetNumber));
for (i = 0; i < nitups; i++)
XLogRegisterBufData(1, (char *) itups[i], tups_size[i]);
XLogRegisterBuffer(2, rbuf, REGBUF_STANDARD);
XLogRegisterBufData(2, (char *) deletable,
ndeletable * sizeof(OffsetNumber));
recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_MOVE_PAGE_CONTENTS);
PageSetLSN(BufferGetPage(wbuf), recptr);
PageSetLSN(BufferGetPage(rbuf), recptr);
}
END_CRIT_SECTION();
tups_moved = true;
}
/*
* release the lock on previous page after acquiring the lock
* on next page
*/
if (retain_pin)
LockBuffer(wbuf, BUFFER_LOCK_UNLOCK);
else
_hash_relbuf(rel, wbuf);
/* nothing more to do if we reached the read page */
if (rblkno == wblkno)
{
_hash_relbuf(rel, rbuf);
return;
}
wbuf = next_wbuf;
wpage = BufferGetPage(wbuf);
wopaque = (HashPageOpaque) PageGetSpecialPointer(wpage);
Assert(wopaque->hasho_bucket == bucket);
retain_pin = false;
/* be tidy */
for (i = 0; i < nitups; i++)
pfree(itups[i]);
nitups = 0;
all_tups_size = 0;
ndeletable = 0;
/*
* after moving the tuples, rpage would have been compacted,
* so we need to rescan it.
*/
if (tups_moved)
goto readpage;
}
/* remember tuple for deletion from "read" page */
deletable[ndeletable++] = roffnum;
/*
* we need a copy of index tuples as they can be freed as part of
* overflow page, however we need them to write a WAL record in
* _hash_freeovflpage.
*/
itups[nitups] = CopyIndexTuple(itup);
tups_size[nitups++] = itemsz;
all_tups_size += itemsz;
}
/*
* If we reach here, there are no live tuples on the "read" page ---
* it was empty when we got to it, or we moved them all. So we can
* just free the page without bothering with deleting tuples
* individually. Then advance to the previous "read" page.
*
* Tricky point here: if our read and write pages are adjacent in the
* bucket chain, our write lock on wbuf will conflict with
* _hash_freeovflpage's attempt to update the sibling links of the
* removed page. In that case, we don't need to lock it again.
*/
rblkno = ropaque->hasho_prevblkno;
Assert(BlockNumberIsValid(rblkno));
/* free this overflow page (releases rbuf) */
_hash_freeovflpage(rel, bucket_buf, rbuf, wbuf, itups, itup_offsets,
tups_size, nitups, bstrategy);
/* be tidy */
for (i = 0; i < nitups; i++)
pfree(itups[i]);
/* are we freeing the page adjacent to wbuf? */
if (rblkno == wblkno)
{
/* retain the pin on primary bucket page till end of bucket scan */
if (wblkno == bucket_blkno)
LockBuffer(wbuf, BUFFER_LOCK_UNLOCK);
else
_hash_relbuf(rel, wbuf);
return;
}
rbuf = _hash_getbuf_with_strategy(rel,
rblkno,
HASH_WRITE,
LH_OVERFLOW_PAGE,
bstrategy);
rpage = BufferGetPage(rbuf);
ropaque = (HashPageOpaque) PageGetSpecialPointer(rpage);
Assert(ropaque->hasho_bucket == bucket);
}
/* NOTREACHED */
}
/*
* _hash_first() -- Find the first item in a scan.
*
* Find the first item in the index that
* satisfies the qualification associated with the scan descriptor. On
* success, the page containing the current index tuple is read locked
* and pinned, and the scan's opaque data entry is updated to
* include the buffer.
*/
bool
_hash_first(IndexScanDesc scan, ScanDirection dir)
{
Relation rel = scan->indexRelation;
HashScanOpaque so = (HashScanOpaque) scan->opaque;
uint32 hashkey;
Bucket bucket;
BlockNumber blkno;
Buffer buf;
Buffer metabuf;
Page page;
HashPageOpaque opaque;
HashMetaPage metap;
IndexTuple itup;
ItemPointer current;
OffsetNumber offnum;
MIRROREDLOCK_BUFMGR_MUST_ALREADY_BE_HELD;
pgstat_count_index_scan(rel);
current = &(scan->currentItemData);
ItemPointerSetInvalid(current);
/*
* We do not support hash scans with no index qualification, because we
* would have to read the whole index rather than just one bucket. That
* creates a whole raft of problems, since we haven't got a practical way
* to lock all the buckets against splits or compactions.
*/
if (scan->numberOfKeys < 1)
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("hash indexes do not support whole-index scans")));
/*
* If the constant in the index qual is NULL, assume it cannot match any
* items in the index.
*/
if (scan->keyData[0].sk_flags & SK_ISNULL)
return false;
/*
* Okay to compute the hash key. We want to do this before acquiring any
* locks, in case a user-defined hash function happens to be slow.
*/
hashkey = _hash_datum2hashkey(rel, scan->keyData[0].sk_argument);
/*
* Acquire shared split lock so we can compute the target bucket safely
* (see README).
*/
_hash_getlock(rel, 0, HASH_SHARE);
/* Read the metapage */
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ);
_hash_checkpage(rel, metabuf, LH_META_PAGE);
metap = (HashMetaPage) BufferGetPage(metabuf);
/*
* Compute the target bucket number, and convert to block number.
*/
bucket = _hash_hashkey2bucket(hashkey,
metap->hashm_maxbucket,
metap->hashm_highmask,
metap->hashm_lowmask);
blkno = BUCKET_TO_BLKNO(metap, bucket);
/* done with the metapage */
_hash_relbuf(rel, metabuf);
/*
* Acquire share lock on target bucket; then we can release split lock.
*/
_hash_getlock(rel, blkno, HASH_SHARE);
_hash_droplock(rel, 0, HASH_SHARE);
/* Update scan opaque state to show we have lock on the bucket */
so->hashso_bucket = bucket;
so->hashso_bucket_valid = true;
so->hashso_bucket_blkno = blkno;
/* Fetch the primary bucket page for the bucket */
buf = _hash_getbuf(rel, blkno, HASH_READ);
_hash_checkpage(rel, buf, LH_BUCKET_PAGE);
page = BufferGetPage(buf);
opaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(opaque->hasho_bucket == bucket);
/* If a backwards scan is requested, move to the end of the chain */
if (ScanDirectionIsBackward(dir))
{
while (BlockNumberIsValid(opaque->hasho_nextblkno))
_hash_readnext(rel, &buf, &page, &opaque);
}
/* Now find the first tuple satisfying the qualification */
if (!_hash_step(scan, &buf, dir))
return false;
/* if we're here, _hash_step found a valid tuple */
offnum = ItemPointerGetOffsetNumber(current);
_hash_checkpage(rel, buf, LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
page = BufferGetPage(buf);
itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
scan->xs_ctup.t_self = itup->t_tid;
return true;
}
/*
* Helper function to perform deletion of index entries from a bucket.
*
* This function expects that the caller has acquired a cleanup lock on the
* primary bucket page, and will return with a write lock again held on the
* primary bucket page. The lock won't necessarily be held continuously,
* though, because we'll release it when visiting overflow pages.
*
* It would be very bad if this function cleaned a page while some other
* backend was in the midst of scanning it, because hashgettuple assumes
* that the next valid TID will be greater than or equal to the current
* valid TID. There can't be any concurrent scans in progress when we first
* enter this function because of the cleanup lock we hold on the primary
* bucket page, but as soon as we release that lock, there might be. We
* handle that by conspiring to prevent those scans from passing our cleanup
* scan. To do that, we lock the next page in the bucket chain before
* releasing the lock on the previous page. (This type of lock chaining is
* not ideal, so we might want to look for a better solution at some point.)
*
* We need to retain a pin on the primary bucket to ensure that no concurrent
* split can start.
*/
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)
{
BlockNumber blkno;
Buffer buf;
Bucket new_bucket PG_USED_FOR_ASSERTS_ONLY = InvalidBucket;
bool bucket_dirty = false;
blkno = bucket_blkno;
buf = bucket_buf;
if (split_cleanup)
new_bucket = _hash_get_newbucket_from_oldbucket(rel, cur_bucket,
lowmask, maxbucket);
/* Scan each page in bucket */
for (;;)
{
HashPageOpaque opaque;
OffsetNumber offno;
OffsetNumber maxoffno;
Buffer next_buf;
Page page;
OffsetNumber deletable[MaxOffsetNumber];
int ndeletable = 0;
bool retain_pin = false;
bool clear_dead_marking = false;
vacuum_delay_point();
page = BufferGetPage(buf);
opaque = (HashPageOpaque) PageGetSpecialPointer(page);
/* Scan each tuple in page */
maxoffno = PageGetMaxOffsetNumber(page);
for (offno = FirstOffsetNumber;
offno <= maxoffno;
offno = OffsetNumberNext(offno))
{
ItemPointer htup;
IndexTuple itup;
Bucket bucket;
bool kill_tuple = false;
itup = (IndexTuple) PageGetItem(page,
PageGetItemId(page, offno));
htup = &(itup->t_tid);
/*
* To remove the dead tuples, we strictly want to rely on results
* of callback function. refer btvacuumpage for detailed reason.
*/
if (callback && callback(htup, callback_state))
{
kill_tuple = true;
if (tuples_removed)
*tuples_removed += 1;
}
else if (split_cleanup)
{
/* delete the tuples that are moved by split. */
bucket = _hash_hashkey2bucket(_hash_get_indextuple_hashkey(itup),
maxbucket,
highmask,
lowmask);
/* mark the item for deletion */
if (bucket != cur_bucket)
{
/*
* We expect tuples to either belong to current bucket or
* new_bucket. This is ensured because we don't allow
* further splits from bucket that contains garbage. See
* comments in _hash_expandtable.
*/
Assert(bucket == new_bucket);
kill_tuple = true;
}
}
if (kill_tuple)
{
/* mark the item for deletion */
deletable[ndeletable++] = offno;
}
else
{
/* we're keeping it, so count it */
if (num_index_tuples)
*num_index_tuples += 1;
}
}
/* retain the pin on primary bucket page till end of bucket scan */
if (blkno == bucket_blkno)
retain_pin = true;
else
retain_pin = false;
blkno = opaque->hasho_nextblkno;
/*
* Apply deletions, advance to next page and write page if needed.
*/
if (ndeletable > 0)
{
/* No ereport(ERROR) until changes are logged */
START_CRIT_SECTION();
PageIndexMultiDelete(page, deletable, ndeletable);
bucket_dirty = true;
/*
* Let us mark the page as clean if vacuum removes the DEAD tuples
* from an index page. We do this by clearing
* LH_PAGE_HAS_DEAD_TUPLES flag.
*/
if (tuples_removed && *tuples_removed > 0 &&
H_HAS_DEAD_TUPLES(opaque))
{
opaque->hasho_flag &= ~LH_PAGE_HAS_DEAD_TUPLES;
clear_dead_marking = true;
}
MarkBufferDirty(buf);
/* XLOG stuff */
if (RelationNeedsWAL(rel))
{
xl_hash_delete xlrec;
XLogRecPtr recptr;
xlrec.clear_dead_marking = clear_dead_marking;
xlrec.is_primary_bucket_page = (buf == bucket_buf) ? true : false;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHashDelete);
/*
* bucket buffer needs to be registered to ensure that we can
* acquire a cleanup lock on it during replay.
*/
if (!xlrec.is_primary_bucket_page)
XLogRegisterBuffer(0, bucket_buf, REGBUF_STANDARD | REGBUF_NO_IMAGE);
XLogRegisterBuffer(1, buf, REGBUF_STANDARD);
XLogRegisterBufData(1, (char *) deletable,
ndeletable * sizeof(OffsetNumber));
recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_DELETE);
PageSetLSN(BufferGetPage(buf), recptr);
}
END_CRIT_SECTION();
}
/* bail out if there are no more pages to scan. */
if (!BlockNumberIsValid(blkno))
break;
next_buf = _hash_getbuf_with_strategy(rel, blkno, HASH_WRITE,
LH_OVERFLOW_PAGE,
bstrategy);
/*
* release the lock on previous page after acquiring the lock on next
* page
*/
if (retain_pin)
LockBuffer(buf, BUFFER_LOCK_UNLOCK);
else
_hash_relbuf(rel, buf);
buf = next_buf;
}
/*
* lock the bucket page to clear the garbage flag and squeeze the bucket.
* if the current buffer is same as bucket buffer, then we already have
* lock on bucket page.
*/
if (buf != bucket_buf)
{
_hash_relbuf(rel, buf);
LockBuffer(bucket_buf, BUFFER_LOCK_EXCLUSIVE);
}
/*
* Clear the garbage flag from bucket after deleting the tuples that are
* moved by split. We purposefully clear the flag before squeeze bucket,
* so that after restart, vacuum shouldn't again try to delete the moved
* by split tuples.
*/
if (split_cleanup)
{
HashPageOpaque bucket_opaque;
Page page;
page = BufferGetPage(bucket_buf);
bucket_opaque = (HashPageOpaque) PageGetSpecialPointer(page);
/* No ereport(ERROR) until changes are logged */
START_CRIT_SECTION();
bucket_opaque->hasho_flag &= ~LH_BUCKET_NEEDS_SPLIT_CLEANUP;
MarkBufferDirty(bucket_buf);
/* XLOG stuff */
if (RelationNeedsWAL(rel))
{
XLogRecPtr recptr;
XLogBeginInsert();
XLogRegisterBuffer(0, bucket_buf, REGBUF_STANDARD);
recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_CLEANUP);
PageSetLSN(page, recptr);
}
END_CRIT_SECTION();
}
/*
* If we have deleted anything, try to compact free space. For squeezing
* the bucket, we must have a cleanup lock, else it can impact the
* ordering of tuples for a scan that has started before it.
*/
if (bucket_dirty && IsBufferCleanupOK(bucket_buf))
_hash_squeezebucket(rel, cur_bucket, bucket_blkno, bucket_buf,
bstrategy);
else
LockBuffer(bucket_buf, BUFFER_LOCK_UNLOCK);
}
/*
* Bulk deletion of all index entries pointing to a set of heap tuples.
* The set of target tuples is specified via a callback routine that tells
* whether any given heap tuple (identified by ItemPointer) is being deleted.
*
* This function also deletes the tuples that are moved by split to other
* bucket.
*
* Result: a palloc'd struct containing statistical info for VACUUM displays.
*/
IndexBulkDeleteResult *
hashbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
IndexBulkDeleteCallback callback, void *callback_state)
{
Relation rel = info->index;
double tuples_removed;
double num_index_tuples;
double orig_ntuples;
Bucket orig_maxbucket;
Bucket cur_maxbucket;
Bucket cur_bucket;
Buffer metabuf = InvalidBuffer;
HashMetaPage metap;
HashMetaPage cachedmetap;
tuples_removed = 0;
num_index_tuples = 0;
/*
* We need a copy of the metapage so that we can use its hashm_spares[]
* values to compute bucket page addresses, but a cached copy should be
* good enough. (If not, we'll detect that further down and refresh the
* cache as necessary.)
*/
cachedmetap = _hash_getcachedmetap(rel, &metabuf, false);
Assert(cachedmetap != NULL);
orig_maxbucket = cachedmetap->hashm_maxbucket;
orig_ntuples = cachedmetap->hashm_ntuples;
/* Scan the buckets that we know exist */
cur_bucket = 0;
cur_maxbucket = orig_maxbucket;
loop_top:
while (cur_bucket <= cur_maxbucket)
{
BlockNumber bucket_blkno;
BlockNumber blkno;
Buffer bucket_buf;
Buffer buf;
HashPageOpaque bucket_opaque;
Page page;
bool split_cleanup = false;
/* Get address of bucket's start page */
bucket_blkno = BUCKET_TO_BLKNO(cachedmetap, cur_bucket);
blkno = bucket_blkno;
/*
* We need to acquire a cleanup lock on the primary bucket page to out
* wait concurrent scans before deleting the dead tuples.
*/
buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_NORMAL, info->strategy);
LockBufferForCleanup(buf);
_hash_checkpage(rel, buf, LH_BUCKET_PAGE);
page = BufferGetPage(buf);
bucket_opaque = (HashPageOpaque) PageGetSpecialPointer(page);
/*
* If the bucket contains tuples that are moved by split, then we need
* to delete such tuples. We can't delete such tuples if the split
* operation on bucket is not finished as those are needed by scans.
*/
if (!H_BUCKET_BEING_SPLIT(bucket_opaque) &&
H_NEEDS_SPLIT_CLEANUP(bucket_opaque))
{
split_cleanup = true;
/*
* This bucket might have been split since we last held a lock on
* the metapage. If so, hashm_maxbucket, hashm_highmask and
* hashm_lowmask might be old enough to cause us to fail to remove
* tuples left behind by the most recent split. To prevent that,
* now that the primary page of the target bucket has been locked
* (and thus can't be further split), check whether we need to
* update our cached metapage data.
*/
Assert(bucket_opaque->hasho_prevblkno != InvalidBlockNumber);
if (bucket_opaque->hasho_prevblkno > cachedmetap->hashm_maxbucket)
{
cachedmetap = _hash_getcachedmetap(rel, &metabuf, true);
Assert(cachedmetap != NULL);
}
}
bucket_buf = buf;
hashbucketcleanup(rel, cur_bucket, bucket_buf, blkno, info->strategy,
cachedmetap->hashm_maxbucket,
cachedmetap->hashm_highmask,
cachedmetap->hashm_lowmask, &tuples_removed,
&num_index_tuples, split_cleanup,
callback, callback_state);
_hash_dropbuf(rel, bucket_buf);
/* Advance to next bucket */
cur_bucket++;
}
if (BufferIsInvalid(metabuf))
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_NOLOCK, LH_META_PAGE);
/* Write-lock metapage and check for split since we started */
LockBuffer(metabuf, BUFFER_LOCK_EXCLUSIVE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
if (cur_maxbucket != metap->hashm_maxbucket)
{
/* There's been a split, so process the additional bucket(s) */
LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
cachedmetap = _hash_getcachedmetap(rel, &metabuf, true);
Assert(cachedmetap != NULL);
cur_maxbucket = cachedmetap->hashm_maxbucket;
goto loop_top;
}
/* Okay, we're really done. Update tuple count in metapage. */
START_CRIT_SECTION();
if (orig_maxbucket == metap->hashm_maxbucket &&
orig_ntuples == metap->hashm_ntuples)
{
/*
* No one has split or inserted anything since start of scan, so
* believe our count as gospel.
*/
metap->hashm_ntuples = num_index_tuples;
}
else
{
/*
* Otherwise, our count is untrustworthy since we may have
* double-scanned tuples in split buckets. Proceed by dead-reckoning.
* (Note: we still return estimated_count = false, because using this
* count is better than not updating reltuples at all.)
*/
if (metap->hashm_ntuples > tuples_removed)
metap->hashm_ntuples -= tuples_removed;
else
metap->hashm_ntuples = 0;
num_index_tuples = metap->hashm_ntuples;
}
MarkBufferDirty(metabuf);
/* XLOG stuff */
if (RelationNeedsWAL(rel))
{
xl_hash_update_meta_page xlrec;
XLogRecPtr recptr;
xlrec.ntuples = metap->hashm_ntuples;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHashUpdateMetaPage);
XLogRegisterBuffer(0, metabuf, REGBUF_STANDARD);
recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_UPDATE_META_PAGE);
PageSetLSN(BufferGetPage(metabuf), recptr);
}
END_CRIT_SECTION();
_hash_relbuf(rel, metabuf);
/* return statistics */
if (stats == NULL)
stats = (IndexBulkDeleteResult *) palloc0(sizeof(IndexBulkDeleteResult));
stats->estimated_count = false;
stats->num_index_tuples = num_index_tuples;
stats->tuples_removed += tuples_removed;
/* hashvacuumcleanup will fill in num_pages */
return stats;
}
/*
* _hash_first() -- Find the first item in a scan.
*
* Find the first item in the index that
* satisfies the qualification associated with the scan descriptor. On
* success, the page containing the current index tuple is read locked
* and pinned, and the scan's opaque data entry is updated to
* include the buffer.
*/
bool
_hash_first(IndexScanDesc scan, ScanDirection dir)
{
Relation rel = scan->indexRelation;
HashScanOpaque so = (HashScanOpaque) scan->opaque;
ScanKey cur;
uint32 hashkey;
Bucket bucket;
BlockNumber blkno;
Buffer buf;
Buffer metabuf;
Page page;
HashPageOpaque opaque;
HashMetaPage metap;
IndexTuple itup;
ItemPointer current;
OffsetNumber offnum;
pgstat_count_index_scan(rel);
current = &(so->hashso_curpos);
ItemPointerSetInvalid(current);
/*
* We do not support hash scans with no index qualification, because we
* would have to read the whole index rather than just one bucket. That
* creates a whole raft of problems, since we haven't got a practical way
* to lock all the buckets against splits or compactions.
*/
if (scan->numberOfKeys < 1)
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("hash indexes do not support whole-index scans")));
/* There may be more than one index qual, but we hash only the first */
cur = &scan->keyData[0];
/* We support only single-column hash indexes */
Assert(cur->sk_attno == 1);
/* And there's only one operator strategy, too */
Assert(cur->sk_strategy == HTEqualStrategyNumber);
/*
* If the constant in the index qual is NULL, assume it cannot match any
* items in the index.
*/
if (cur->sk_flags & SK_ISNULL)
return false;
/*
* Okay to compute the hash key. We want to do this before acquiring any
* locks, in case a user-defined hash function happens to be slow.
*
* If scankey operator is not a cross-type comparison, we can use the
* cached hash function; otherwise gotta look it up in the catalogs.
*
* We support the convention that sk_subtype == InvalidOid means the
* opclass input type; this is a hack to simplify life for ScanKeyInit().
*/
if (cur->sk_subtype == rel->rd_opcintype[0] ||
cur->sk_subtype == InvalidOid)
hashkey = _hash_datum2hashkey(rel, cur->sk_argument);
else
hashkey = _hash_datum2hashkey_type(rel, cur->sk_argument,
cur->sk_subtype);
so->hashso_sk_hash = hashkey;
/*
* Acquire shared split lock so we can compute the target bucket safely
* (see README).
*/
_hash_getlock(rel, 0, HASH_SHARE);
/* Read the metapage */
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
/*
* Compute the target bucket number, and convert to block number.
*/
bucket = _hash_hashkey2bucket(hashkey,
metap->hashm_maxbucket,
metap->hashm_highmask,
metap->hashm_lowmask);
blkno = BUCKET_TO_BLKNO(metap, bucket);
/* done with the metapage */
_hash_relbuf(rel, metabuf);
/*
* Acquire share lock on target bucket; then we can release split lock.
*/
_hash_getlock(rel, blkno, HASH_SHARE);
_hash_droplock(rel, 0, HASH_SHARE);
/* Update scan opaque state to show we have lock on the bucket */
so->hashso_bucket = bucket;
so->hashso_bucket_valid = true;
so->hashso_bucket_blkno = blkno;
/* Fetch the primary bucket page for the bucket */
buf = _hash_getbuf(rel, blkno, HASH_READ, LH_BUCKET_PAGE);
page = BufferGetPage(buf);
opaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(opaque->hasho_bucket == bucket);
/* If a backwards scan is requested, move to the end of the chain */
if (ScanDirectionIsBackward(dir))
{
while (BlockNumberIsValid(opaque->hasho_nextblkno))
_hash_readnext(rel, &buf, &page, &opaque);
}
/* Now find the first tuple satisfying the qualification */
if (!_hash_step(scan, &buf, dir))
return false;
/* if we're here, _hash_step found a valid tuple */
offnum = ItemPointerGetOffsetNumber(current);
_hash_checkpage(rel, buf, LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
page = BufferGetPage(buf);
itup = (IndexTuple) PageGetItem(page, PageGetItemId(page, offnum));
so->hashso_heappos = itup->t_tid;
return true;
}
/*
* _hash_finish_split() -- Finish the previously interrupted split operation
*
* To complete the split operation, we form the hash table of TIDs in new
* bucket which is then used by split operation to skip tuples that are
* already moved before the split operation was previously interrupted.
*
* The caller must hold a pin, but no lock, on the metapage and old bucket's
* primary page buffer. The buffers are returned in the same state. (The
* metapage is only touched if it becomes necessary to add or remove overflow
* pages.)
*/
void
_hash_finish_split(Relation rel, Buffer metabuf, Buffer obuf, Bucket obucket,
uint32 maxbucket, uint32 highmask, uint32 lowmask)
{
HASHCTL hash_ctl;
HTAB *tidhtab;
Buffer bucket_nbuf = InvalidBuffer;
Buffer nbuf;
Page npage;
BlockNumber nblkno;
BlockNumber bucket_nblkno;
HashPageOpaque npageopaque;
Bucket nbucket;
bool found;
/* Initialize hash tables used to track TIDs */
memset(&hash_ctl, 0, sizeof(hash_ctl));
hash_ctl.keysize = sizeof(ItemPointerData);
hash_ctl.entrysize = sizeof(ItemPointerData);
hash_ctl.hcxt = CurrentMemoryContext;
tidhtab =
hash_create("bucket ctids",
256, /* arbitrary initial size */
&hash_ctl,
HASH_ELEM | HASH_BLOBS | HASH_CONTEXT);
bucket_nblkno = nblkno = _hash_get_newblock_from_oldbucket(rel, obucket);
/*
* Scan the new bucket and build hash table of TIDs
*/
for (;;)
{
OffsetNumber noffnum;
OffsetNumber nmaxoffnum;
nbuf = _hash_getbuf(rel, nblkno, HASH_READ,
LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
/* remember the primary bucket buffer to acquire cleanup lock on it. */
if (nblkno == bucket_nblkno)
bucket_nbuf = nbuf;
npage = BufferGetPage(nbuf);
npageopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
/* Scan each tuple in new page */
nmaxoffnum = PageGetMaxOffsetNumber(npage);
for (noffnum = FirstOffsetNumber;
noffnum <= nmaxoffnum;
noffnum = OffsetNumberNext(noffnum))
{
IndexTuple itup;
/* Fetch the item's TID and insert it in hash table. */
itup = (IndexTuple) PageGetItem(npage,
PageGetItemId(npage, noffnum));
(void) hash_search(tidhtab, &itup->t_tid, HASH_ENTER, &found);
Assert(!found);
}
nblkno = npageopaque->hasho_nextblkno;
/*
* release our write lock without modifying buffer and ensure to
* retain the pin on primary bucket.
*/
if (nbuf == bucket_nbuf)
LockBuffer(nbuf, BUFFER_LOCK_UNLOCK);
else
_hash_relbuf(rel, nbuf);
/* Exit loop if no more overflow pages in new bucket */
if (!BlockNumberIsValid(nblkno))
break;
}
/*
* Conditionally get the cleanup lock on old and new buckets to perform
* the split operation. If we don't get the cleanup locks, silently give
* up and next insertion on old bucket will try again to complete the
* split.
*/
if (!ConditionalLockBufferForCleanup(obuf))
{
hash_destroy(tidhtab);
return;
}
if (!ConditionalLockBufferForCleanup(bucket_nbuf))
{
LockBuffer(obuf, BUFFER_LOCK_UNLOCK);
hash_destroy(tidhtab);
return;
}
npage = BufferGetPage(bucket_nbuf);
npageopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
nbucket = npageopaque->hasho_bucket;
_hash_splitbucket(rel, metabuf, obucket,
nbucket, obuf, bucket_nbuf, tidhtab,
maxbucket, highmask, lowmask);
_hash_dropbuf(rel, bucket_nbuf);
hash_destroy(tidhtab);
}
/*
* _hash_init() -- Initialize the metadata page of a hash index,
* the initial buckets, and the initial bitmap page.
*
* The initial number of buckets is dependent on num_tuples, an estimate
* of the number of tuples to be loaded into the index initially. The
* chosen number of buckets is returned.
*
* We are fairly cavalier about locking here, since we know that no one else
* could be accessing this index. In particular the rule about not holding
* multiple buffer locks is ignored.
*/
uint32
_hash_init(Relation rel, double num_tuples, ForkNumber forkNum)
{
Buffer metabuf;
Buffer buf;
Buffer bitmapbuf;
Page pg;
HashMetaPage metap;
RegProcedure procid;
int32 data_width;
int32 item_width;
int32 ffactor;
uint32 num_buckets;
uint32 i;
bool use_wal;
/* safety check */
if (RelationGetNumberOfBlocksInFork(rel, forkNum) != 0)
elog(ERROR, "cannot initialize non-empty hash index \"%s\"",
RelationGetRelationName(rel));
/*
* WAL log creation of pages if the relation is persistent, or this is the
* init fork. Init forks for unlogged relations always need to be WAL
* logged.
*/
use_wal = RelationNeedsWAL(rel) || forkNum == INIT_FORKNUM;
/*
* Determine the target fill factor (in tuples per bucket) for this index.
* The idea is to make the fill factor correspond to pages about as full
* as the user-settable fillfactor parameter says. We can compute it
* exactly since the index datatype (i.e. uint32 hash key) is fixed-width.
*/
data_width = sizeof(uint32);
item_width = MAXALIGN(sizeof(IndexTupleData)) + MAXALIGN(data_width) +
sizeof(ItemIdData); /* include the line pointer */
ffactor = RelationGetTargetPageUsage(rel, HASH_DEFAULT_FILLFACTOR) / item_width;
/* keep to a sane range */
if (ffactor < 10)
ffactor = 10;
procid = index_getprocid(rel, 1, HASHSTANDARD_PROC);
/*
* We initialize the metapage, the first N bucket pages, and the first
* bitmap page in sequence, using _hash_getnewbuf to cause smgrextend()
* calls to occur. This ensures that the smgr level has the right idea of
* the physical index length.
*
* Critical section not required, because on error the creation of the
* whole relation will be rolled back.
*/
metabuf = _hash_getnewbuf(rel, HASH_METAPAGE, forkNum);
_hash_init_metabuffer(metabuf, num_tuples, procid, ffactor, false);
MarkBufferDirty(metabuf);
pg = BufferGetPage(metabuf);
metap = HashPageGetMeta(pg);
/* XLOG stuff */
if (use_wal)
{
xl_hash_init_meta_page xlrec;
XLogRecPtr recptr;
xlrec.num_tuples = num_tuples;
xlrec.procid = metap->hashm_procid;
xlrec.ffactor = metap->hashm_ffactor;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHashInitMetaPage);
XLogRegisterBuffer(0, metabuf, REGBUF_WILL_INIT);
recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_INIT_META_PAGE);
PageSetLSN(BufferGetPage(metabuf), recptr);
}
num_buckets = metap->hashm_maxbucket + 1;
/*
* Release buffer lock on the metapage while we initialize buckets.
* Otherwise, we'll be in interrupt holdoff and the CHECK_FOR_INTERRUPTS
* won't accomplish anything. It's a bad idea to hold buffer locks for
* long intervals in any case, since that can block the bgwriter.
*/
LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
/*
* Initialize and WAL Log the first N buckets
*/
for (i = 0; i < num_buckets; i++)
{
BlockNumber blkno;
/* Allow interrupts, in case N is huge */
CHECK_FOR_INTERRUPTS();
blkno = BUCKET_TO_BLKNO(metap, i);
buf = _hash_getnewbuf(rel, blkno, forkNum);
_hash_initbuf(buf, metap->hashm_maxbucket, i, LH_BUCKET_PAGE, false);
MarkBufferDirty(buf);
if (use_wal)
log_newpage(&rel->rd_node,
forkNum,
blkno,
BufferGetPage(buf),
true);
_hash_relbuf(rel, buf);
}
/* Now reacquire buffer lock on metapage */
LockBuffer(metabuf, BUFFER_LOCK_EXCLUSIVE);
/*
* Initialize bitmap page
*/
bitmapbuf = _hash_getnewbuf(rel, num_buckets + 1, forkNum);
_hash_initbitmapbuffer(bitmapbuf, metap->hashm_bmsize, false);
MarkBufferDirty(bitmapbuf);
/* add the new bitmap page to the metapage's list of bitmaps */
/* metapage already has a write lock */
if (metap->hashm_nmaps >= HASH_MAX_BITMAPS)
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("out of overflow pages in hash index \"%s\"",
RelationGetRelationName(rel))));
metap->hashm_mapp[metap->hashm_nmaps] = num_buckets + 1;
metap->hashm_nmaps++;
MarkBufferDirty(metabuf);
/* XLOG stuff */
if (use_wal)
{
xl_hash_init_bitmap_page xlrec;
XLogRecPtr recptr;
xlrec.bmsize = metap->hashm_bmsize;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHashInitBitmapPage);
XLogRegisterBuffer(0, bitmapbuf, REGBUF_WILL_INIT);
/*
* This is safe only because nobody else can be modifying the index at
* this stage; it's only visible to the transaction that is creating
* it.
*/
XLogRegisterBuffer(1, metabuf, REGBUF_STANDARD);
recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_INIT_BITMAP_PAGE);
PageSetLSN(BufferGetPage(bitmapbuf), recptr);
PageSetLSN(BufferGetPage(metabuf), recptr);
}
/* all done */
_hash_relbuf(rel, bitmapbuf);
_hash_relbuf(rel, metabuf);
return num_buckets;
}
/* ------------------------------------------------
* hash_bitmap_info()
*
* Get bitmap information for a particular overflow page
*
* Usage: SELECT * FROM hash_bitmap_info('con_hash_index'::regclass, 5);
* ------------------------------------------------
*/
Datum
hash_bitmap_info(PG_FUNCTION_ARGS)
{
Oid indexRelid = PG_GETARG_OID(0);
uint64 ovflblkno = PG_GETARG_INT64(1);
HashMetaPage metap;
Buffer metabuf,
mapbuf;
BlockNumber bitmapblkno;
Page mappage;
bool bit = false;
TupleDesc tupleDesc;
Relation indexRel;
uint32 ovflbitno;
int32 bitmappage,
bitmapbit;
HeapTuple tuple;
int i,
j;
Datum values[3];
bool nulls[3];
uint32 *freep;
if (!superuser())
ereport(ERROR,
(errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
(errmsg("must be superuser to use raw page functions"))));
indexRel = index_open(indexRelid, AccessShareLock);
if (!IS_HASH(indexRel))
elog(ERROR, "relation \"%s\" is not a hash index",
RelationGetRelationName(indexRel));
if (RELATION_IS_OTHER_TEMP(indexRel))
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("cannot access temporary tables of other sessions")));
if (ovflblkno >= RelationGetNumberOfBlocks(indexRel))
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("block number " UINT64_FORMAT " is out of range for relation \"%s\"",
ovflblkno, RelationGetRelationName(indexRel))));
/* Read the metapage so we can determine which bitmap page to use */
metabuf = _hash_getbuf(indexRel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
/*
* Reject attempt to read the bit for a metapage or bitmap page; this is
* only meaningful for overflow pages.
*/
if (ovflblkno == 0)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("invalid overflow block number %u",
(BlockNumber) ovflblkno)));
for (i = 0; i < metap->hashm_nmaps; i++)
if (metap->hashm_mapp[i] == ovflblkno)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("invalid overflow block number %u",
(BlockNumber) ovflblkno)));
/*
* Identify overflow bit number. This will error out for primary bucket
* pages, and we've already rejected the metapage and bitmap pages above.
*/
ovflbitno = _hash_ovflblkno_to_bitno(metap, (BlockNumber) ovflblkno);
bitmappage = ovflbitno >> BMPG_SHIFT(metap);
bitmapbit = ovflbitno & BMPG_MASK(metap);
if (bitmappage >= metap->hashm_nmaps)
ereport(ERROR,
(errcode(ERRCODE_INVALID_PARAMETER_VALUE),
errmsg("invalid overflow block number %u",
(BlockNumber) ovflblkno)));
bitmapblkno = metap->hashm_mapp[bitmappage];
_hash_relbuf(indexRel, metabuf);
/* Check the status of bitmap bit for overflow page */
mapbuf = _hash_getbuf(indexRel, bitmapblkno, HASH_READ, LH_BITMAP_PAGE);
mappage = BufferGetPage(mapbuf);
freep = HashPageGetBitmap(mappage);
bit = ISSET(freep, bitmapbit) != 0;
_hash_relbuf(indexRel, mapbuf);
index_close(indexRel, AccessShareLock);
/* Build a tuple descriptor for our result type */
if (get_call_result_type(fcinfo, NULL, &tupleDesc) != TYPEFUNC_COMPOSITE)
elog(ERROR, "return type must be a row type");
tupleDesc = BlessTupleDesc(tupleDesc);
MemSet(nulls, 0, sizeof(nulls));
j = 0;
values[j++] = Int64GetDatum((int64) bitmapblkno);
values[j++] = Int32GetDatum(bitmapbit);
values[j++] = BoolGetDatum(bit);
tuple = heap_form_tuple(tupleDesc, values, nulls);
PG_RETURN_DATUM(HeapTupleGetDatum(tuple));
}
/*
* _hash_freeovflpage() -
*
* Remove this overflow page from its bucket's chain, and mark the page as
* free. On entry, ovflbuf is write-locked; it is released before exiting.
*
* Since this function is invoked in VACUUM, we provide an access strategy
* parameter that controls fetches of the bucket pages.
*
* Returns the block number of the page that followed the given page
* in the bucket, or InvalidBlockNumber if no following page.
*
* NB: caller must not hold lock on metapage, nor on either page that's
* adjacent in the bucket chain. The caller had better hold exclusive lock
* on the bucket, too.
*/
BlockNumber
_hash_freeovflpage(Relation rel, Buffer ovflbuf,
BufferAccessStrategy bstrategy)
{
HashMetaPage metap;
Buffer metabuf;
Buffer mapbuf;
BlockNumber ovflblkno;
BlockNumber prevblkno;
BlockNumber blkno;
BlockNumber nextblkno;
HashPageOpaque ovflopaque;
Page ovflpage;
Page mappage;
uint32 *freep;
uint32 ovflbitno;
int32 bitmappage,
bitmapbit;
Bucket bucket PG_USED_FOR_ASSERTS_ONLY;
/* Get information from the doomed page */
_hash_checkpage(rel, ovflbuf, LH_OVERFLOW_PAGE);
ovflblkno = BufferGetBlockNumber(ovflbuf);
ovflpage = BufferGetPage(ovflbuf);
ovflopaque = (HashPageOpaque) PageGetSpecialPointer(ovflpage);
nextblkno = ovflopaque->hasho_nextblkno;
prevblkno = ovflopaque->hasho_prevblkno;
bucket = ovflopaque->hasho_bucket;
/*
* Zero the page for debugging's sake; then write and release it. (Note:
* if we failed to zero the page here, we'd have problems with the Assert
* in _hash_pageinit() when the page is reused.)
*/
MemSet(ovflpage, 0, BufferGetPageSize(ovflbuf));
_hash_wrtbuf(rel, ovflbuf);
/*
* Fix up the bucket chain. this is a doubly-linked list, so we must fix
* up the bucket chain members behind and ahead of the overflow page being
* deleted. No concurrency issues since we hold exclusive lock on the
* entire bucket.
*/
if (BlockNumberIsValid(prevblkno))
{
Buffer prevbuf = _hash_getbuf_with_strategy(rel,
prevblkno,
HASH_WRITE,
LH_BUCKET_PAGE | LH_OVERFLOW_PAGE,
bstrategy);
Page prevpage = BufferGetPage(prevbuf);
HashPageOpaque prevopaque = (HashPageOpaque) PageGetSpecialPointer(prevpage);
Assert(prevopaque->hasho_bucket == bucket);
prevopaque->hasho_nextblkno = nextblkno;
_hash_wrtbuf(rel, prevbuf);
}
if (BlockNumberIsValid(nextblkno))
{
Buffer nextbuf = _hash_getbuf_with_strategy(rel,
nextblkno,
HASH_WRITE,
LH_OVERFLOW_PAGE,
bstrategy);
Page nextpage = BufferGetPage(nextbuf);
HashPageOpaque nextopaque = (HashPageOpaque) PageGetSpecialPointer(nextpage);
Assert(nextopaque->hasho_bucket == bucket);
nextopaque->hasho_prevblkno = prevblkno;
_hash_wrtbuf(rel, nextbuf);
}
/* Note: bstrategy is intentionally not used for metapage and bitmap */
/* Read the metapage so we can determine which bitmap page to use */
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
/* Identify which bit to set */
ovflbitno = blkno_to_bitno(metap, ovflblkno);
bitmappage = ovflbitno >> BMPG_SHIFT(metap);
bitmapbit = ovflbitno & BMPG_MASK(metap);
if (bitmappage >= metap->hashm_nmaps)
elog(ERROR, "invalid overflow bit number %u", ovflbitno);
blkno = metap->hashm_mapp[bitmappage];
/* Release metapage lock while we access the bitmap page */
_hash_chgbufaccess(rel, metabuf, HASH_READ, HASH_NOLOCK);
/* Clear the bitmap bit to indicate that this overflow page is free */
mapbuf = _hash_getbuf(rel, blkno, HASH_WRITE, LH_BITMAP_PAGE);
mappage = BufferGetPage(mapbuf);
freep = HashPageGetBitmap(mappage);
Assert(ISSET(freep, bitmapbit));
CLRBIT(freep, bitmapbit);
_hash_wrtbuf(rel, mapbuf);
/* Get write-lock on metapage to update firstfree */
_hash_chgbufaccess(rel, metabuf, HASH_NOLOCK, HASH_WRITE);
/* if this is now the first free page, update hashm_firstfree */
if (ovflbitno < metap->hashm_firstfree)
{
metap->hashm_firstfree = ovflbitno;
_hash_wrtbuf(rel, metabuf);
}
else
{
/* no need to change metapage */
_hash_relbuf(rel, metabuf);
}
return nextblkno;
}
/*
* Attempt to expand the hash table by creating one new bucket.
*
* This will silently do nothing if we don't get cleanup lock on old or
* new bucket.
*
* Complete the pending splits and remove the tuples from old bucket,
* if there are any left over from the previous split.
*
* The caller must hold a pin, but no lock, on the metapage buffer.
* The buffer is returned in the same state.
*/
void
_hash_expandtable(Relation rel, Buffer metabuf)
{
HashMetaPage metap;
Bucket old_bucket;
Bucket new_bucket;
uint32 spare_ndx;
BlockNumber start_oblkno;
BlockNumber start_nblkno;
Buffer buf_nblkno;
Buffer buf_oblkno;
Page opage;
Page npage;
HashPageOpaque oopaque;
HashPageOpaque nopaque;
uint32 maxbucket;
uint32 highmask;
uint32 lowmask;
bool metap_update_masks = false;
bool metap_update_splitpoint = false;
restart_expand:
/*
* Write-lock the meta page. It used to be necessary to acquire a
* heavyweight lock to begin a split, but that is no longer required.
*/
LockBuffer(metabuf, BUFFER_LOCK_EXCLUSIVE);
_hash_checkpage(rel, metabuf, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
/*
* Check to see if split is still needed; someone else might have already
* done one while we waited for the lock.
*
* Make sure this stays in sync with _hash_doinsert()
*/
if (metap->hashm_ntuples <=
(double) metap->hashm_ffactor * (metap->hashm_maxbucket + 1))
goto fail;
/*
* Can't split anymore if maxbucket has reached its maximum possible
* value.
*
* Ideally we'd allow bucket numbers up to UINT_MAX-1 (no higher because
* the calculation maxbucket+1 mustn't overflow). Currently we restrict
* to half that because of overflow looping in _hash_log2() and
* insufficient space in hashm_spares[]. It's moot anyway because an
* index with 2^32 buckets would certainly overflow BlockNumber and hence
* _hash_alloc_buckets() would fail, but if we supported buckets smaller
* than a disk block then this would be an independent constraint.
*
* If you change this, see also the maximum initial number of buckets in
* _hash_init().
*/
if (metap->hashm_maxbucket >= (uint32) 0x7FFFFFFE)
goto fail;
/*
* Determine which bucket is to be split, and attempt to take cleanup lock
* on the old bucket. If we can't get the lock, give up.
*
* The cleanup lock protects us not only against other backends, but
* against our own backend as well.
*
* The cleanup lock is mainly to protect the split from concurrent
* inserts. See src/backend/access/hash/README, Lock Definitions for
* further details. Due to this locking restriction, if there is any
* pending scan, the split will give up which is not good, but harmless.
*/
new_bucket = metap->hashm_maxbucket + 1;
old_bucket = (new_bucket & metap->hashm_lowmask);
start_oblkno = BUCKET_TO_BLKNO(metap, old_bucket);
buf_oblkno = _hash_getbuf_with_condlock_cleanup(rel, start_oblkno, LH_BUCKET_PAGE);
if (!buf_oblkno)
goto fail;
opage = BufferGetPage(buf_oblkno);
oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
/*
* We want to finish the split from a bucket as there is no apparent
* benefit by not doing so and it will make the code complicated to finish
* the split that involves multiple buckets considering the case where new
* split also fails. We don't need to consider the new bucket for
* completing the split here as it is not possible that a re-split of new
* bucket starts when there is still a pending split from old bucket.
*/
if (H_BUCKET_BEING_SPLIT(oopaque))
{
/*
* Copy bucket mapping info now; refer the comment in code below where
* we copy this information before calling _hash_splitbucket to see
* why this is okay.
*/
maxbucket = metap->hashm_maxbucket;
highmask = metap->hashm_highmask;
lowmask = metap->hashm_lowmask;
/*
* Release the lock on metapage and old_bucket, before completing the
* split.
*/
LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
LockBuffer(buf_oblkno, BUFFER_LOCK_UNLOCK);
_hash_finish_split(rel, metabuf, buf_oblkno, old_bucket, maxbucket,
highmask, lowmask);
/* release the pin on old buffer and retry for expand. */
_hash_dropbuf(rel, buf_oblkno);
goto restart_expand;
}
/*
* Clean the tuples remained from the previous split. This operation
* requires cleanup lock and we already have one on the old bucket, so
* let's do it. We also don't want to allow further splits from the bucket
* till the garbage of previous split is cleaned. This has two
* advantages; first, it helps in avoiding the bloat due to garbage and
* second is, during cleanup of bucket, we are always sure that the
* garbage tuples belong to most recently split bucket. On the contrary,
* if we allow cleanup of bucket after meta page is updated to indicate
* the new split and before the actual split, the cleanup operation won't
* be able to decide whether the tuple has been moved to the newly created
* bucket and ended up deleting such tuples.
*/
if (H_NEEDS_SPLIT_CLEANUP(oopaque))
{
/*
* Copy bucket mapping info now; refer to the comment in code below
* where we copy this information before calling _hash_splitbucket to
* see why this is okay.
*/
maxbucket = metap->hashm_maxbucket;
highmask = metap->hashm_highmask;
lowmask = metap->hashm_lowmask;
/* Release the metapage lock. */
LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
hashbucketcleanup(rel, old_bucket, buf_oblkno, start_oblkno, NULL,
maxbucket, highmask, lowmask, NULL, NULL, true,
NULL, NULL);
_hash_dropbuf(rel, buf_oblkno);
goto restart_expand;
}
/*
* There shouldn't be any active scan on new bucket.
*
* Note: it is safe to compute the new bucket's blkno here, even though we
* may still need to update the BUCKET_TO_BLKNO mapping. This is because
* the current value of hashm_spares[hashm_ovflpoint] correctly shows
* where we are going to put a new splitpoint's worth of buckets.
*/
start_nblkno = BUCKET_TO_BLKNO(metap, new_bucket);
/*
* If the split point is increasing we need to allocate a new batch of
* bucket pages.
*/
spare_ndx = _hash_spareindex(new_bucket + 1);
if (spare_ndx > metap->hashm_ovflpoint)
{
uint32 buckets_to_add;
Assert(spare_ndx == metap->hashm_ovflpoint + 1);
/*
* We treat allocation of buckets as a separate WAL-logged action.
* Even if we fail after this operation, won't leak bucket pages;
* rather, the next split will consume this space. In any case, even
* without failure we don't use all the space in one split operation.
*/
buckets_to_add = _hash_get_totalbuckets(spare_ndx) - new_bucket;
if (!_hash_alloc_buckets(rel, start_nblkno, buckets_to_add))
{
/* can't split due to BlockNumber overflow */
_hash_relbuf(rel, buf_oblkno);
goto fail;
}
}
/*
* Physically allocate the new bucket's primary page. We want to do this
* before changing the metapage's mapping info, in case we can't get the
* disk space. Ideally, we don't need to check for cleanup lock on new
* bucket as no other backend could find this bucket unless meta page is
* updated. However, it is good to be consistent with old bucket locking.
*/
buf_nblkno = _hash_getnewbuf(rel, start_nblkno, MAIN_FORKNUM);
if (!IsBufferCleanupOK(buf_nblkno))
{
_hash_relbuf(rel, buf_oblkno);
_hash_relbuf(rel, buf_nblkno);
goto fail;
}
/*
* Since we are scribbling on the pages in the shared buffers, establish a
* critical section. Any failure in this next code leaves us with a big
* problem: the metapage is effectively corrupt but could get written back
* to disk.
*/
START_CRIT_SECTION();
/*
* Okay to proceed with split. Update the metapage bucket mapping info.
*/
metap->hashm_maxbucket = new_bucket;
if (new_bucket > metap->hashm_highmask)
{
/* Starting a new doubling */
metap->hashm_lowmask = metap->hashm_highmask;
metap->hashm_highmask = new_bucket | metap->hashm_lowmask;
metap_update_masks = true;
}
/*
* If the split point is increasing we need to adjust the hashm_spares[]
* array and hashm_ovflpoint so that future overflow pages will be created
* beyond this new batch of bucket pages.
*/
if (spare_ndx > metap->hashm_ovflpoint)
{
metap->hashm_spares[spare_ndx] = metap->hashm_spares[metap->hashm_ovflpoint];
metap->hashm_ovflpoint = spare_ndx;
metap_update_splitpoint = true;
}
MarkBufferDirty(metabuf);
/*
* Copy bucket mapping info now; this saves re-accessing the meta page
* inside _hash_splitbucket's inner loop. Note that once we drop the
* split lock, other splits could begin, so these values might be out of
* date before _hash_splitbucket finishes. That's okay, since all it
* needs is to tell which of these two buckets to map hashkeys into.
*/
maxbucket = metap->hashm_maxbucket;
highmask = metap->hashm_highmask;
lowmask = metap->hashm_lowmask;
opage = BufferGetPage(buf_oblkno);
oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
/*
* Mark the old bucket to indicate that split is in progress. (At
* operation end, we will clear the split-in-progress flag.) Also, for a
* primary bucket page, hasho_prevblkno stores the number of buckets that
* existed as of the last split, so we must update that value here.
*/
oopaque->hasho_flag |= LH_BUCKET_BEING_SPLIT;
oopaque->hasho_prevblkno = maxbucket;
MarkBufferDirty(buf_oblkno);
npage = BufferGetPage(buf_nblkno);
/*
* initialize the new bucket's primary page and mark it to indicate that
* split is in progress.
*/
nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
nopaque->hasho_prevblkno = maxbucket;
nopaque->hasho_nextblkno = InvalidBlockNumber;
nopaque->hasho_bucket = new_bucket;
nopaque->hasho_flag = LH_BUCKET_PAGE | LH_BUCKET_BEING_POPULATED;
nopaque->hasho_page_id = HASHO_PAGE_ID;
MarkBufferDirty(buf_nblkno);
/* XLOG stuff */
if (RelationNeedsWAL(rel))
{
xl_hash_split_allocate_page xlrec;
XLogRecPtr recptr;
xlrec.new_bucket = maxbucket;
xlrec.old_bucket_flag = oopaque->hasho_flag;
xlrec.new_bucket_flag = nopaque->hasho_flag;
xlrec.flags = 0;
XLogBeginInsert();
XLogRegisterBuffer(0, buf_oblkno, REGBUF_STANDARD);
XLogRegisterBuffer(1, buf_nblkno, REGBUF_WILL_INIT);
XLogRegisterBuffer(2, metabuf, REGBUF_STANDARD);
if (metap_update_masks)
{
xlrec.flags |= XLH_SPLIT_META_UPDATE_MASKS;
XLogRegisterBufData(2, (char *) &metap->hashm_lowmask, sizeof(uint32));
XLogRegisterBufData(2, (char *) &metap->hashm_highmask, sizeof(uint32));
}
if (metap_update_splitpoint)
{
xlrec.flags |= XLH_SPLIT_META_UPDATE_SPLITPOINT;
XLogRegisterBufData(2, (char *) &metap->hashm_ovflpoint,
sizeof(uint32));
XLogRegisterBufData(2,
(char *) &metap->hashm_spares[metap->hashm_ovflpoint],
sizeof(uint32));
}
XLogRegisterData((char *) &xlrec, SizeOfHashSplitAllocPage);
recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SPLIT_ALLOCATE_PAGE);
PageSetLSN(BufferGetPage(buf_oblkno), recptr);
PageSetLSN(BufferGetPage(buf_nblkno), recptr);
PageSetLSN(BufferGetPage(metabuf), recptr);
}
END_CRIT_SECTION();
/* drop lock, but keep pin */
LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
/* Relocate records to the new bucket */
_hash_splitbucket(rel, metabuf,
old_bucket, new_bucket,
buf_oblkno, buf_nblkno, NULL,
maxbucket, highmask, lowmask);
/* all done, now release the pins on primary buckets. */
_hash_dropbuf(rel, buf_oblkno);
_hash_dropbuf(rel, buf_nblkno);
return;
/* Here if decide not to split or fail to acquire old bucket lock */
fail:
/* We didn't write the metapage, so just drop lock */
LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
}
/* ------------------------------------------------------
* pgstathashindex()
*
* Usage: SELECT * FROM pgstathashindex('hashindex');
* ------------------------------------------------------
*/
Datum
pgstathashindex(PG_FUNCTION_ARGS)
{
Oid relid = PG_GETARG_OID(0);
BlockNumber nblocks;
BlockNumber blkno;
Relation rel;
HashIndexStat stats;
BufferAccessStrategy bstrategy;
HeapTuple tuple;
TupleDesc tupleDesc;
Datum values[8];
bool nulls[8];
Buffer metabuf;
HashMetaPage metap;
float8 free_percent;
uint64 total_space;
rel = index_open(relid, AccessShareLock);
/* index_open() checks that it's an index */
if (!IS_HASH(rel))
ereport(ERROR,
(errcode(ERRCODE_WRONG_OBJECT_TYPE),
errmsg("relation \"%s\" is not a HASH index",
RelationGetRelationName(rel))));
/*
* Reject attempts to read non-local temporary relations; we would be
* likely to get wrong data since we have no visibility into the owning
* session's local buffers.
*/
if (RELATION_IS_OTHER_TEMP(rel))
ereport(ERROR,
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
errmsg("cannot access temporary indexes of other sessions")));
/* Get the information we need from the metapage. */
memset(&stats, 0, sizeof(stats));
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
stats.version = metap->hashm_version;
stats.space_per_page = metap->hashm_bsize;
_hash_relbuf(rel, metabuf);
/* Get the current relation length */
nblocks = RelationGetNumberOfBlocks(rel);
/* prepare access strategy for this index */
bstrategy = GetAccessStrategy(BAS_BULKREAD);
/* Start from blkno 1 as 0th block is metapage */
for (blkno = 1; blkno < nblocks; blkno++)
{
Buffer buf;
Page page;
CHECK_FOR_INTERRUPTS();
buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_NORMAL,
bstrategy);
LockBuffer(buf, BUFFER_LOCK_SHARE);
page = (Page) BufferGetPage(buf);
if (PageIsNew(page))
stats.unused_pages++;
else if (PageGetSpecialSize(page) !=
MAXALIGN(sizeof(HashPageOpaqueData)))
ereport(ERROR,
(errcode(ERRCODE_INDEX_CORRUPTED),
errmsg("index \"%s\" contains corrupted page at block %u",
RelationGetRelationName(rel),
BufferGetBlockNumber(buf))));
else
{
HashPageOpaque opaque;
int pagetype;
opaque = (HashPageOpaque) PageGetSpecialPointer(page);
pagetype = opaque->hasho_flag & LH_PAGE_TYPE;
if (pagetype == LH_BUCKET_PAGE)
{
stats.bucket_pages++;
GetHashPageStats(page, &stats);
}
else if (pagetype == LH_OVERFLOW_PAGE)
{
stats.overflow_pages++;
GetHashPageStats(page, &stats);
}
else if (pagetype == LH_BITMAP_PAGE)
stats.bitmap_pages++;
else if (pagetype == LH_UNUSED_PAGE)
stats.unused_pages++;
else
ereport(ERROR,
(errcode(ERRCODE_INDEX_CORRUPTED),
errmsg("unexpected page type 0x%04X in HASH index \"%s\" block %u",
opaque->hasho_flag, RelationGetRelationName(rel),
BufferGetBlockNumber(buf))));
}
UnlockReleaseBuffer(buf);
}
/* Done accessing the index */
index_close(rel, AccessShareLock);
/* Count unused pages as free space. */
stats.free_space += stats.unused_pages * stats.space_per_page;
/*
* Total space available for tuples excludes the metapage and the bitmap
* pages.
*/
total_space = (nblocks - (stats.bitmap_pages + 1)) * stats.space_per_page;
if (total_space == 0)
free_percent = 0.0;
else
free_percent = 100.0 * stats.free_space / total_space;
/*
* Build a tuple descriptor for our result type
*/
if (get_call_result_type(fcinfo, NULL, &tupleDesc) != TYPEFUNC_COMPOSITE)
elog(ERROR, "return type must be a row type");
tupleDesc = BlessTupleDesc(tupleDesc);
/*
* Build and return the tuple
*/
MemSet(nulls, 0, sizeof(nulls));
values[0] = Int32GetDatum(stats.version);
values[1] = Int64GetDatum((int64) stats.bucket_pages);
values[2] = Int64GetDatum((int64) stats.overflow_pages);
values[3] = Int64GetDatum((int64) stats.bitmap_pages);
values[4] = Int64GetDatum((int64) stats.unused_pages);
values[5] = Int64GetDatum(stats.live_items);
values[6] = Int64GetDatum(stats.dead_items);
values[7] = Float8GetDatum(free_percent);
tuple = heap_form_tuple(tupleDesc, values, nulls);
PG_RETURN_DATUM(HeapTupleGetDatum(tuple));
}
/*
* _hash_freeovflpage() -
*
* Remove this overflow page from its bucket's chain, and mark the page as
* free. On entry, ovflbuf is write-locked; it is released before exiting.
*
* Add the tuples (itups) to wbuf in this function. We could do that in the
* caller as well, but the advantage of doing it here is we can easily write
* the WAL for XLOG_HASH_SQUEEZE_PAGE operation. Addition of tuples and
* removal of overflow page has to done as an atomic operation, otherwise
* during replay on standby users might find duplicate records.
*
* Since this function is invoked in VACUUM, we provide an access strategy
* parameter that controls fetches of the bucket pages.
*
* Returns the block number of the page that followed the given page
* in the bucket, or InvalidBlockNumber if no following page.
*
* NB: caller must not hold lock on metapage, nor on page, that's next to
* ovflbuf in the bucket chain. We don't acquire the lock on page that's
* prior to ovflbuf in chain if it is same as wbuf because the caller already
* has a lock on same.
*/
BlockNumber
_hash_freeovflpage(Relation rel, Buffer bucketbuf, Buffer ovflbuf,
Buffer wbuf, IndexTuple *itups, OffsetNumber *itup_offsets,
Size *tups_size, uint16 nitups,
BufferAccessStrategy bstrategy)
{
HashMetaPage metap;
Buffer metabuf;
Buffer mapbuf;
BlockNumber ovflblkno;
BlockNumber prevblkno;
BlockNumber blkno;
BlockNumber nextblkno;
BlockNumber writeblkno;
HashPageOpaque ovflopaque;
Page ovflpage;
Page mappage;
uint32 *freep;
uint32 ovflbitno;
int32 bitmappage,
bitmapbit;
Bucket bucket PG_USED_FOR_ASSERTS_ONLY;
Buffer prevbuf = InvalidBuffer;
Buffer nextbuf = InvalidBuffer;
bool update_metap = false;
/* Get information from the doomed page */
_hash_checkpage(rel, ovflbuf, LH_OVERFLOW_PAGE);
ovflblkno = BufferGetBlockNumber(ovflbuf);
ovflpage = BufferGetPage(ovflbuf);
ovflopaque = (HashPageOpaque) PageGetSpecialPointer(ovflpage);
nextblkno = ovflopaque->hasho_nextblkno;
prevblkno = ovflopaque->hasho_prevblkno;
writeblkno = BufferGetBlockNumber(wbuf);
bucket = ovflopaque->hasho_bucket;
/*
* Fix up the bucket chain. this is a doubly-linked list, so we must fix
* up the bucket chain members behind and ahead of the overflow page being
* deleted. Concurrency issues are avoided by using lock chaining as
* described atop hashbucketcleanup.
*/
if (BlockNumberIsValid(prevblkno))
{
if (prevblkno == writeblkno)
prevbuf = wbuf;
else
prevbuf = _hash_getbuf_with_strategy(rel,
prevblkno,
HASH_WRITE,
LH_BUCKET_PAGE | LH_OVERFLOW_PAGE,
bstrategy);
}
if (BlockNumberIsValid(nextblkno))
nextbuf = _hash_getbuf_with_strategy(rel,
nextblkno,
HASH_WRITE,
LH_OVERFLOW_PAGE,
bstrategy);
/* Note: bstrategy is intentionally not used for metapage and bitmap */
/* Read the metapage so we can determine which bitmap page to use */
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
/* Identify which bit to set */
ovflbitno = _hash_ovflblkno_to_bitno(metap, ovflblkno);
bitmappage = ovflbitno >> BMPG_SHIFT(metap);
bitmapbit = ovflbitno & BMPG_MASK(metap);
if (bitmappage >= metap->hashm_nmaps)
elog(ERROR, "invalid overflow bit number %u", ovflbitno);
blkno = metap->hashm_mapp[bitmappage];
/* Release metapage lock while we access the bitmap page */
LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
/* read the bitmap page to clear the bitmap bit */
mapbuf = _hash_getbuf(rel, blkno, HASH_WRITE, LH_BITMAP_PAGE);
mappage = BufferGetPage(mapbuf);
freep = HashPageGetBitmap(mappage);
Assert(ISSET(freep, bitmapbit));
/* Get write-lock on metapage to update firstfree */
LockBuffer(metabuf, BUFFER_LOCK_EXCLUSIVE);
/* This operation needs to log multiple tuples, prepare WAL for that */
if (RelationNeedsWAL(rel))
XLogEnsureRecordSpace(HASH_XLOG_FREE_OVFL_BUFS, 4 + nitups);
START_CRIT_SECTION();
/*
* we have to insert tuples on the "write" page, being careful to preserve
* hashkey ordering. (If we insert many tuples into the same "write" page
* it would be worth qsort'ing them).
*/
if (nitups > 0)
{
_hash_pgaddmultitup(rel, wbuf, itups, itup_offsets, nitups);
MarkBufferDirty(wbuf);
}
/*
* Reinitialize the freed overflow page. Just zeroing the page won't
* work, because WAL replay routines expect pages to be initialized. See
* explanation of RBM_NORMAL mode atop XLogReadBufferExtended. We are
* careful to make the special space valid here so that tools like
* pageinspect won't get confused.
*/
_hash_pageinit(ovflpage, BufferGetPageSize(ovflbuf));
ovflopaque = (HashPageOpaque) PageGetSpecialPointer(ovflpage);
ovflopaque->hasho_prevblkno = InvalidBlockNumber;
ovflopaque->hasho_nextblkno = InvalidBlockNumber;
ovflopaque->hasho_bucket = -1;
ovflopaque->hasho_flag = LH_UNUSED_PAGE;
ovflopaque->hasho_page_id = HASHO_PAGE_ID;
MarkBufferDirty(ovflbuf);
if (BufferIsValid(prevbuf))
{
Page prevpage = BufferGetPage(prevbuf);
HashPageOpaque prevopaque = (HashPageOpaque) PageGetSpecialPointer(prevpage);
Assert(prevopaque->hasho_bucket == bucket);
prevopaque->hasho_nextblkno = nextblkno;
MarkBufferDirty(prevbuf);
}
if (BufferIsValid(nextbuf))
{
Page nextpage = BufferGetPage(nextbuf);
HashPageOpaque nextopaque = (HashPageOpaque) PageGetSpecialPointer(nextpage);
Assert(nextopaque->hasho_bucket == bucket);
nextopaque->hasho_prevblkno = prevblkno;
MarkBufferDirty(nextbuf);
}
/* Clear the bitmap bit to indicate that this overflow page is free */
CLRBIT(freep, bitmapbit);
MarkBufferDirty(mapbuf);
/* if this is now the first free page, update hashm_firstfree */
if (ovflbitno < metap->hashm_firstfree)
{
metap->hashm_firstfree = ovflbitno;
update_metap = true;
MarkBufferDirty(metabuf);
}
/* XLOG stuff */
if (RelationNeedsWAL(rel))
{
xl_hash_squeeze_page xlrec;
XLogRecPtr recptr;
int i;
xlrec.prevblkno = prevblkno;
xlrec.nextblkno = nextblkno;
xlrec.ntups = nitups;
xlrec.is_prim_bucket_same_wrt = (wbuf == bucketbuf);
xlrec.is_prev_bucket_same_wrt = (wbuf == prevbuf);
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHashSqueezePage);
/*
* bucket buffer needs to be registered to ensure that we can acquire
* a cleanup lock on it during replay.
*/
if (!xlrec.is_prim_bucket_same_wrt)
XLogRegisterBuffer(0, bucketbuf, REGBUF_STANDARD | REGBUF_NO_IMAGE);
XLogRegisterBuffer(1, wbuf, REGBUF_STANDARD);
if (xlrec.ntups > 0)
{
XLogRegisterBufData(1, (char *) itup_offsets,
nitups * sizeof(OffsetNumber));
for (i = 0; i < nitups; i++)
XLogRegisterBufData(1, (char *) itups[i], tups_size[i]);
}
XLogRegisterBuffer(2, ovflbuf, REGBUF_STANDARD);
/*
* If prevpage and the writepage (block in which we are moving tuples
* from overflow) are same, then no need to separately register
* prevpage. During replay, we can directly update the nextblock in
* writepage.
*/
if (BufferIsValid(prevbuf) && !xlrec.is_prev_bucket_same_wrt)
XLogRegisterBuffer(3, prevbuf, REGBUF_STANDARD);
if (BufferIsValid(nextbuf))
XLogRegisterBuffer(4, nextbuf, REGBUF_STANDARD);
XLogRegisterBuffer(5, mapbuf, REGBUF_STANDARD);
XLogRegisterBufData(5, (char *) &bitmapbit, sizeof(uint32));
if (update_metap)
{
XLogRegisterBuffer(6, metabuf, REGBUF_STANDARD);
XLogRegisterBufData(6, (char *) &metap->hashm_firstfree, sizeof(uint32));
}
recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_SQUEEZE_PAGE);
PageSetLSN(BufferGetPage(wbuf), recptr);
PageSetLSN(BufferGetPage(ovflbuf), recptr);
if (BufferIsValid(prevbuf) && !xlrec.is_prev_bucket_same_wrt)
PageSetLSN(BufferGetPage(prevbuf), recptr);
if (BufferIsValid(nextbuf))
PageSetLSN(BufferGetPage(nextbuf), recptr);
PageSetLSN(BufferGetPage(mapbuf), recptr);
if (update_metap)
PageSetLSN(BufferGetPage(metabuf), recptr);
}
END_CRIT_SECTION();
/* release previous bucket if it is not same as write bucket */
if (BufferIsValid(prevbuf) && prevblkno != writeblkno)
_hash_relbuf(rel, prevbuf);
if (BufferIsValid(ovflbuf))
_hash_relbuf(rel, ovflbuf);
if (BufferIsValid(nextbuf))
_hash_relbuf(rel, nextbuf);
_hash_relbuf(rel, mapbuf);
_hash_relbuf(rel, metabuf);
return nextblkno;
}
/*
* _hash_addovflpage
*
* Add an overflow page to the bucket whose last page is pointed to by 'buf'.
*
* On entry, the caller must hold a pin but no lock on 'buf'. The pin is
* dropped before exiting (we assume the caller is not interested in 'buf'
* anymore) if not asked to retain. The pin will be retained only for the
* primary bucket. The returned overflow page will be pinned and
* write-locked; it is guaranteed to be empty.
*
* The caller must hold a pin, but no lock, on the metapage buffer.
* That buffer is returned in the same state.
*
* NB: since this could be executed concurrently by multiple processes,
* one should not assume that the returned overflow page will be the
* immediate successor of the originally passed 'buf'. Additional overflow
* pages might have been added to the bucket chain in between.
*/
Buffer
_hash_addovflpage(Relation rel, Buffer metabuf, Buffer buf, bool retain_pin)
{
Buffer ovflbuf;
Page page;
Page ovflpage;
HashPageOpaque pageopaque;
HashPageOpaque ovflopaque;
HashMetaPage metap;
Buffer mapbuf = InvalidBuffer;
Buffer newmapbuf = InvalidBuffer;
BlockNumber blkno;
uint32 orig_firstfree;
uint32 splitnum;
uint32 *freep = NULL;
uint32 max_ovflpg;
uint32 bit;
uint32 bitmap_page_bit;
uint32 first_page;
uint32 last_bit;
uint32 last_page;
uint32 i,
j;
bool page_found = false;
/*
* Write-lock the tail page. Here, we need to maintain locking order such
* that, first acquire the lock on tail page of bucket, then on meta page
* to find and lock the bitmap page and if it is found, then lock on meta
* page is released, then finally acquire the lock on new overflow buffer.
* We need this locking order to avoid deadlock with backends that are
* doing inserts.
*
* Note: We could have avoided locking many buffers here if we made two
* WAL records for acquiring an overflow page (one to allocate an overflow
* page and another to add it to overflow bucket chain). However, doing
* so can leak an overflow page, if the system crashes after allocation.
* Needless to say, it is better to have a single record from a
* performance point of view as well.
*/
LockBuffer(buf, BUFFER_LOCK_EXCLUSIVE);
/* probably redundant... */
_hash_checkpage(rel, buf, LH_BUCKET_PAGE | LH_OVERFLOW_PAGE);
/* loop to find current tail page, in case someone else inserted too */
for (;;)
{
BlockNumber nextblkno;
page = BufferGetPage(buf);
pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
nextblkno = pageopaque->hasho_nextblkno;
if (!BlockNumberIsValid(nextblkno))
break;
/* we assume we do not need to write the unmodified page */
if (retain_pin)
{
/* pin will be retained only for the primary bucket page */
Assert((pageopaque->hasho_flag & LH_PAGE_TYPE) == LH_BUCKET_PAGE);
LockBuffer(buf, BUFFER_LOCK_UNLOCK);
}
else
_hash_relbuf(rel, buf);
retain_pin = false;
buf = _hash_getbuf(rel, nextblkno, HASH_WRITE, LH_OVERFLOW_PAGE);
}
/* Get exclusive lock on the meta page */
LockBuffer(metabuf, BUFFER_LOCK_EXCLUSIVE);
_hash_checkpage(rel, metabuf, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
/* start search at hashm_firstfree */
orig_firstfree = metap->hashm_firstfree;
first_page = orig_firstfree >> BMPG_SHIFT(metap);
bit = orig_firstfree & BMPG_MASK(metap);
i = first_page;
j = bit / BITS_PER_MAP;
bit &= ~(BITS_PER_MAP - 1);
/* outer loop iterates once per bitmap page */
for (;;)
{
BlockNumber mapblkno;
Page mappage;
uint32 last_inpage;
/* want to end search with the last existing overflow page */
splitnum = metap->hashm_ovflpoint;
max_ovflpg = metap->hashm_spares[splitnum] - 1;
last_page = max_ovflpg >> BMPG_SHIFT(metap);
last_bit = max_ovflpg & BMPG_MASK(metap);
if (i > last_page)
break;
Assert(i < metap->hashm_nmaps);
mapblkno = metap->hashm_mapp[i];
if (i == last_page)
last_inpage = last_bit;
else
last_inpage = BMPGSZ_BIT(metap) - 1;
/* Release exclusive lock on metapage while reading bitmap page */
LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
mapbuf = _hash_getbuf(rel, mapblkno, HASH_WRITE, LH_BITMAP_PAGE);
mappage = BufferGetPage(mapbuf);
freep = HashPageGetBitmap(mappage);
for (; bit <= last_inpage; j++, bit += BITS_PER_MAP)
{
if (freep[j] != ALL_SET)
{
page_found = true;
/* Reacquire exclusive lock on the meta page */
LockBuffer(metabuf, BUFFER_LOCK_EXCLUSIVE);
/* convert bit to bit number within page */
bit += _hash_firstfreebit(freep[j]);
bitmap_page_bit = bit;
/* convert bit to absolute bit number */
bit += (i << BMPG_SHIFT(metap));
/* Calculate address of the recycled overflow page */
blkno = bitno_to_blkno(metap, bit);
/* Fetch and init the recycled page */
ovflbuf = _hash_getinitbuf(rel, blkno);
goto found;
}
}
/* No free space here, try to advance to next map page */
_hash_relbuf(rel, mapbuf);
mapbuf = InvalidBuffer;
i++;
j = 0; /* scan from start of next map page */
bit = 0;
/* Reacquire exclusive lock on the meta page */
LockBuffer(metabuf, BUFFER_LOCK_EXCLUSIVE);
}
/*
* No free pages --- have to extend the relation to add an overflow page.
* First, check to see if we have to add a new bitmap page too.
*/
if (last_bit == (uint32) (BMPGSZ_BIT(metap) - 1))
{
/*
* We create the new bitmap page with all pages marked "in use".
* Actually two pages in the new bitmap's range will exist
* immediately: the bitmap page itself, and the following page which
* is the one we return to the caller. Both of these are correctly
* marked "in use". Subsequent pages do not exist yet, but it is
* convenient to pre-mark them as "in use" too.
*/
bit = metap->hashm_spares[splitnum];
/* metapage already has a write lock */
if (metap->hashm_nmaps >= HASH_MAX_BITMAPS)
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("out of overflow pages in hash index \"%s\"",
RelationGetRelationName(rel))));
newmapbuf = _hash_getnewbuf(rel, bitno_to_blkno(metap, bit), MAIN_FORKNUM);
}
else
{
/*
* Nothing to do here; since the page will be past the last used page,
* we know its bitmap bit was preinitialized to "in use".
*/
}
/* Calculate address of the new overflow page */
bit = BufferIsValid(newmapbuf) ?
metap->hashm_spares[splitnum] + 1 : metap->hashm_spares[splitnum];
blkno = bitno_to_blkno(metap, bit);
/*
* Fetch the page with _hash_getnewbuf to ensure smgr's idea of the
* relation length stays in sync with ours. XXX It's annoying to do this
* with metapage write lock held; would be better to use a lock that
* doesn't block incoming searches.
*
* It is okay to hold two buffer locks here (one on tail page of bucket
* and other on new overflow page) since there cannot be anyone else
* contending for access to ovflbuf.
*/
ovflbuf = _hash_getnewbuf(rel, blkno, MAIN_FORKNUM);
found:
/*
* Do the update. No ereport(ERROR) until changes are logged. We want to
* log the changes for bitmap page and overflow page together to avoid
* loss of pages in case the new page is added.
*/
START_CRIT_SECTION();
if (page_found)
{
Assert(BufferIsValid(mapbuf));
/* mark page "in use" in the bitmap */
SETBIT(freep, bitmap_page_bit);
MarkBufferDirty(mapbuf);
}
else
{
/* update the count to indicate new overflow page is added */
metap->hashm_spares[splitnum]++;
if (BufferIsValid(newmapbuf))
{
_hash_initbitmapbuffer(newmapbuf, metap->hashm_bmsize, false);
MarkBufferDirty(newmapbuf);
/* add the new bitmap page to the metapage's list of bitmaps */
metap->hashm_mapp[metap->hashm_nmaps] = BufferGetBlockNumber(newmapbuf);
metap->hashm_nmaps++;
metap->hashm_spares[splitnum]++;
}
MarkBufferDirty(metabuf);
/*
* for new overflow page, we don't need to explicitly set the bit in
* bitmap page, as by default that will be set to "in use".
*/
}
/*
* Adjust hashm_firstfree to avoid redundant searches. But don't risk
* changing it if someone moved it while we were searching bitmap pages.
*/
if (metap->hashm_firstfree == orig_firstfree)
{
metap->hashm_firstfree = bit + 1;
MarkBufferDirty(metabuf);
}
/* initialize new overflow page */
ovflpage = BufferGetPage(ovflbuf);
ovflopaque = (HashPageOpaque) PageGetSpecialPointer(ovflpage);
ovflopaque->hasho_prevblkno = BufferGetBlockNumber(buf);
ovflopaque->hasho_nextblkno = InvalidBlockNumber;
ovflopaque->hasho_bucket = pageopaque->hasho_bucket;
ovflopaque->hasho_flag = LH_OVERFLOW_PAGE;
ovflopaque->hasho_page_id = HASHO_PAGE_ID;
MarkBufferDirty(ovflbuf);
/* logically chain overflow page to previous page */
pageopaque->hasho_nextblkno = BufferGetBlockNumber(ovflbuf);
MarkBufferDirty(buf);
/* XLOG stuff */
if (RelationNeedsWAL(rel))
{
XLogRecPtr recptr;
xl_hash_add_ovfl_page xlrec;
xlrec.bmpage_found = page_found;
xlrec.bmsize = metap->hashm_bmsize;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfHashAddOvflPage);
XLogRegisterBuffer(0, ovflbuf, REGBUF_WILL_INIT);
XLogRegisterBufData(0, (char *) &pageopaque->hasho_bucket, sizeof(Bucket));
XLogRegisterBuffer(1, buf, REGBUF_STANDARD);
if (BufferIsValid(mapbuf))
{
XLogRegisterBuffer(2, mapbuf, REGBUF_STANDARD);
XLogRegisterBufData(2, (char *) &bitmap_page_bit, sizeof(uint32));
}
if (BufferIsValid(newmapbuf))
XLogRegisterBuffer(3, newmapbuf, REGBUF_WILL_INIT);
XLogRegisterBuffer(4, metabuf, REGBUF_STANDARD);
XLogRegisterBufData(4, (char *) &metap->hashm_firstfree, sizeof(uint32));
recptr = XLogInsert(RM_HASH_ID, XLOG_HASH_ADD_OVFL_PAGE);
PageSetLSN(BufferGetPage(ovflbuf), recptr);
PageSetLSN(BufferGetPage(buf), recptr);
if (BufferIsValid(mapbuf))
PageSetLSN(BufferGetPage(mapbuf), recptr);
if (BufferIsValid(newmapbuf))
PageSetLSN(BufferGetPage(newmapbuf), recptr);
PageSetLSN(BufferGetPage(metabuf), recptr);
}
END_CRIT_SECTION();
if (retain_pin)
LockBuffer(buf, BUFFER_LOCK_UNLOCK);
else
_hash_relbuf(rel, buf);
if (BufferIsValid(mapbuf))
_hash_relbuf(rel, mapbuf);
LockBuffer(metabuf, BUFFER_LOCK_UNLOCK);
if (BufferIsValid(newmapbuf))
_hash_relbuf(rel, newmapbuf);
return ovflbuf;
}
/*
* _hash_doinsert() -- Handle insertion of a single index tuple.
*
* This routine is called by the public interface routines, hashbuild
* and hashinsert. By here, itup is completely filled in.
*/
void
_hash_doinsert(Relation rel, IndexTuple itup)
{
Buffer buf;
Buffer metabuf;
HashMetaPage metap;
BlockNumber blkno;
Page page;
HashPageOpaque pageopaque;
Size itemsz;
bool do_expand;
uint32 hashkey;
Bucket bucket;
/*
* Get the hash key for the item (it's stored in the index tuple itself).
*/
hashkey = _hash_get_indextuple_hashkey(itup);
/* compute item size too */
itemsz = IndexTupleDSize(*itup);
itemsz = MAXALIGN(itemsz); /* be safe, PageAddItem will do this but we
* need to be consistent */
/*
* Acquire shared split lock so we can compute the target bucket safely
* (see README).
*/
_hash_getlock(rel, 0, HASH_SHARE);
/* Read the metapage */
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
/*
* Check whether the item can fit on a hash page at all. (Eventually, we
* ought to try to apply TOAST methods if not.) Note that at this point,
* itemsz doesn't include the ItemId.
*
* XXX this is useless code if we are only storing hash keys.
*/
if (itemsz > HashMaxItemSize((Page) metap))
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("index row size %lu exceeds hash maximum %lu",
(unsigned long) itemsz,
(unsigned long) HashMaxItemSize((Page) metap)),
errhint("Values larger than a buffer page cannot be indexed.")));
/*
* Compute the target bucket number, and convert to block number.
*/
bucket = _hash_hashkey2bucket(hashkey,
metap->hashm_maxbucket,
metap->hashm_highmask,
metap->hashm_lowmask);
blkno = BUCKET_TO_BLKNO(metap, bucket);
/* release lock on metapage, but keep pin since we'll need it again */
_hash_chgbufaccess(rel, metabuf, HASH_READ, HASH_NOLOCK);
/*
* Acquire share lock on target bucket; then we can release split lock.
*/
_hash_getlock(rel, blkno, HASH_SHARE);
_hash_droplock(rel, 0, HASH_SHARE);
/* Fetch the primary bucket page for the bucket */
buf = _hash_getbuf(rel, blkno, HASH_WRITE, LH_BUCKET_PAGE);
page = BufferGetPage(buf);
pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(pageopaque->hasho_bucket == bucket);
/* Do the insertion */
while (PageGetFreeSpace(page) < itemsz)
{
/*
* no space on this page; check for an overflow page
*/
BlockNumber nextblkno = pageopaque->hasho_nextblkno;
if (BlockNumberIsValid(nextblkno))
{
/*
* ovfl page exists; go get it. if it doesn't have room, we'll
* find out next pass through the loop test above.
*/
_hash_relbuf(rel, buf);
buf = _hash_getbuf(rel, nextblkno, HASH_WRITE, LH_OVERFLOW_PAGE);
page = BufferGetPage(buf);
}
else
{
/*
* we're at the end of the bucket chain and we haven't found a
* page with enough room. allocate a new overflow page.
*/
/* release our write lock without modifying buffer */
_hash_chgbufaccess(rel, buf, HASH_READ, HASH_NOLOCK);
/* chain to a new overflow page */
buf = _hash_addovflpage(rel, metabuf, buf);
page = BufferGetPage(buf);
/* should fit now, given test above */
Assert(PageGetFreeSpace(page) >= itemsz);
}
pageopaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(pageopaque->hasho_flag == LH_OVERFLOW_PAGE);
Assert(pageopaque->hasho_bucket == bucket);
}
/* found page with enough space, so add the item here */
(void) _hash_pgaddtup(rel, buf, itemsz, itup);
/* write and release the modified page */
_hash_wrtbuf(rel, buf);
/* We can drop the bucket lock now */
_hash_droplock(rel, blkno, HASH_SHARE);
/*
* Write-lock the metapage so we can increment the tuple count. After
* incrementing it, check to see if it's time for a split.
*/
_hash_chgbufaccess(rel, metabuf, HASH_NOLOCK, HASH_WRITE);
metap->hashm_ntuples += 1;
/* Make sure this stays in sync with _hash_expandtable() */
do_expand = metap->hashm_ntuples >
(double) metap->hashm_ffactor * (metap->hashm_maxbucket + 1);
/* Write out the metapage and drop lock, but keep pin */
_hash_chgbufaccess(rel, metabuf, HASH_WRITE, HASH_NOLOCK);
/* Attempt to split if a split is needed */
if (do_expand)
_hash_expandtable(rel, metabuf);
/* Finally drop our pin on the metapage */
_hash_dropbuf(rel, metabuf);
}
/*
* _hash_squeezebucket(rel, bucket)
*
* Try to squeeze the tuples onto pages occurring earlier in the
* bucket chain in an attempt to free overflow pages. When we start
* the "squeezing", the page from which we start taking tuples (the
* "read" page) is the last bucket in the bucket chain and the page
* onto which we start squeezing tuples (the "write" page) is the
* first page in the bucket chain. The read page works backward and
* the write page works forward; the procedure terminates when the
* read page and write page are the same page.
*
* At completion of this procedure, it is guaranteed that all pages in
* the bucket are nonempty, unless the bucket is totally empty (in
* which case all overflow pages will be freed). The original implementation
* required that to be true on entry as well, but it's a lot easier for
* callers to leave empty overflow pages and let this guy clean it up.
*
* Caller must hold exclusive lock on the target bucket. This allows
* us to safely lock multiple pages in the bucket.
*/
void
_hash_squeezebucket(Relation rel,
Bucket bucket,
BlockNumber bucket_blkno)
{
Buffer wbuf;
Buffer rbuf = 0;
BlockNumber wblkno;
BlockNumber rblkno;
Page wpage;
Page rpage;
HashPageOpaque wopaque;
HashPageOpaque ropaque;
OffsetNumber woffnum;
OffsetNumber roffnum;
IndexTuple itup;
Size itemsz;
/*
* start squeezing into the base bucket page.
*/
wblkno = bucket_blkno;
wbuf = _hash_getbuf(rel, wblkno, HASH_WRITE);
_hash_checkpage(rel, wbuf, LH_BUCKET_PAGE);
wpage = BufferGetPage(wbuf);
wopaque = (HashPageOpaque) PageGetSpecialPointer(wpage);
/*
* if there aren't any overflow pages, there's nothing to squeeze.
*/
if (!BlockNumberIsValid(wopaque->hasho_nextblkno))
{
_hash_relbuf(rel, wbuf);
return;
}
/*
* find the last page in the bucket chain by starting at the base bucket
* page and working forward.
*/
ropaque = wopaque;
do
{
rblkno = ropaque->hasho_nextblkno;
if (ropaque != wopaque)
_hash_relbuf(rel, rbuf);
rbuf = _hash_getbuf(rel, rblkno, HASH_WRITE);
_hash_checkpage(rel, rbuf, LH_OVERFLOW_PAGE);
rpage = BufferGetPage(rbuf);
ropaque = (HashPageOpaque) PageGetSpecialPointer(rpage);
Assert(ropaque->hasho_bucket == bucket);
} while (BlockNumberIsValid(ropaque->hasho_nextblkno));
/*
* squeeze the tuples.
*/
roffnum = FirstOffsetNumber;
for (;;)
{
/* this test is needed in case page is empty on entry */
if (roffnum <= PageGetMaxOffsetNumber(rpage))
{
itup = (IndexTuple) PageGetItem(rpage,
PageGetItemId(rpage, roffnum));
itemsz = IndexTupleDSize(*itup);
itemsz = MAXALIGN(itemsz);
/*
* Walk up the bucket chain, looking for a page big enough for
* this item. Exit if we reach the read page.
*/
while (PageGetFreeSpace(wpage) < itemsz)
{
Assert(!PageIsEmpty(wpage));
wblkno = wopaque->hasho_nextblkno;
Assert(BlockNumberIsValid(wblkno));
_hash_wrtbuf(rel, wbuf);
if (rblkno == wblkno)
{
/* wbuf is already released */
_hash_wrtbuf(rel, rbuf);
return;
}
wbuf = _hash_getbuf(rel, wblkno, HASH_WRITE);
_hash_checkpage(rel, wbuf, LH_OVERFLOW_PAGE);
wpage = BufferGetPage(wbuf);
wopaque = (HashPageOpaque) PageGetSpecialPointer(wpage);
Assert(wopaque->hasho_bucket == bucket);
}
/*
* we have found room so insert on the "write" page.
*/
woffnum = OffsetNumberNext(PageGetMaxOffsetNumber(wpage));
if (PageAddItem(wpage, (Item) itup, itemsz, woffnum, LP_USED)
== InvalidOffsetNumber)
elog(ERROR, "failed to add index item to \"%s\"",
RelationGetRelationName(rel));
/*
* delete the tuple from the "read" page. PageIndexTupleDelete
* repacks the ItemId array, so 'roffnum' will be "advanced" to
* the "next" ItemId.
*/
PageIndexTupleDelete(rpage, roffnum);
}
/*
* if the "read" page is now empty because of the deletion (or because
* it was empty when we got to it), free it.
*
* Tricky point here: if our read and write pages are adjacent in the
* bucket chain, our write lock on wbuf will conflict with
* _hash_freeovflpage's attempt to update the sibling links of the
* removed page. However, in that case we are done anyway, so we can
* simply drop the write lock before calling _hash_freeovflpage.
*/
if (PageIsEmpty(rpage))
{
rblkno = ropaque->hasho_prevblkno;
Assert(BlockNumberIsValid(rblkno));
/* are we freeing the page adjacent to wbuf? */
if (rblkno == wblkno)
{
/* yes, so release wbuf lock first */
_hash_wrtbuf(rel, wbuf);
/* free this overflow page (releases rbuf) */
_hash_freeovflpage(rel, rbuf);
/* done */
return;
}
/* free this overflow page, then get the previous one */
_hash_freeovflpage(rel, rbuf);
rbuf = _hash_getbuf(rel, rblkno, HASH_WRITE);
_hash_checkpage(rel, rbuf, LH_OVERFLOW_PAGE);
rpage = BufferGetPage(rbuf);
ropaque = (HashPageOpaque) PageGetSpecialPointer(rpage);
Assert(ropaque->hasho_bucket == bucket);
roffnum = FirstOffsetNumber;
}
}
/* NOTREACHED */
}
/*
* _hash_getovflpage()
*
* Find an available overflow page and return it. The returned buffer
* is pinned and write-locked, and has had _hash_pageinit() applied,
* but it is caller's responsibility to fill the special space.
*
* The caller must hold a pin, but no lock, on the metapage buffer.
* That buffer is left in the same state at exit.
*/
static Buffer
_hash_getovflpage(Relation rel, Buffer metabuf)
{
HashMetaPage metap;
Buffer mapbuf = 0;
Buffer newbuf;
BlockNumber blkno;
uint32 orig_firstfree;
uint32 splitnum;
uint32 *freep = NULL;
uint32 max_ovflpg;
uint32 bit;
uint32 first_page;
uint32 last_bit;
uint32 last_page;
uint32 i,
j;
/* Get exclusive lock on the meta page */
_hash_chgbufaccess(rel, metabuf, HASH_NOLOCK, HASH_WRITE);
_hash_checkpage(rel, metabuf, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
/* start search at hashm_firstfree */
orig_firstfree = metap->hashm_firstfree;
first_page = orig_firstfree >> BMPG_SHIFT(metap);
bit = orig_firstfree & BMPG_MASK(metap);
i = first_page;
j = bit / BITS_PER_MAP;
bit &= ~(BITS_PER_MAP - 1);
/* outer loop iterates once per bitmap page */
for (;;)
{
BlockNumber mapblkno;
Page mappage;
uint32 last_inpage;
/* want to end search with the last existing overflow page */
splitnum = metap->hashm_ovflpoint;
max_ovflpg = metap->hashm_spares[splitnum] - 1;
last_page = max_ovflpg >> BMPG_SHIFT(metap);
last_bit = max_ovflpg & BMPG_MASK(metap);
if (i > last_page)
break;
Assert(i < metap->hashm_nmaps);
mapblkno = metap->hashm_mapp[i];
if (i == last_page)
last_inpage = last_bit;
else
last_inpage = BMPGSZ_BIT(metap) - 1;
/* Release exclusive lock on metapage while reading bitmap page */
_hash_chgbufaccess(rel, metabuf, HASH_READ, HASH_NOLOCK);
mapbuf = _hash_getbuf(rel, mapblkno, HASH_WRITE, LH_BITMAP_PAGE);
mappage = BufferGetPage(mapbuf);
freep = HashPageGetBitmap(mappage);
for (; bit <= last_inpage; j++, bit += BITS_PER_MAP)
{
if (freep[j] != ALL_SET)
goto found;
}
/* No free space here, try to advance to next map page */
_hash_relbuf(rel, mapbuf);
i++;
j = 0; /* scan from start of next map page */
bit = 0;
/* Reacquire exclusive lock on the meta page */
_hash_chgbufaccess(rel, metabuf, HASH_NOLOCK, HASH_WRITE);
}
/*
* No free pages --- have to extend the relation to add an overflow page.
* First, check to see if we have to add a new bitmap page too.
*/
if (last_bit == (uint32) (BMPGSZ_BIT(metap) - 1))
{
/*
* We create the new bitmap page with all pages marked "in use".
* Actually two pages in the new bitmap's range will exist
* immediately: the bitmap page itself, and the following page which
* is the one we return to the caller. Both of these are correctly
* marked "in use". Subsequent pages do not exist yet, but it is
* convenient to pre-mark them as "in use" too.
*/
bit = metap->hashm_spares[splitnum];
_hash_initbitmap(rel, metap, bitno_to_blkno(metap, bit), MAIN_FORKNUM);
metap->hashm_spares[splitnum]++;
}
else
{
/*
* Nothing to do here; since the page will be past the last used page,
* we know its bitmap bit was preinitialized to "in use".
*/
}
/* Calculate address of the new overflow page */
bit = metap->hashm_spares[splitnum];
blkno = bitno_to_blkno(metap, bit);
/*
* Fetch the page with _hash_getnewbuf to ensure smgr's idea of the
* relation length stays in sync with ours. XXX It's annoying to do this
* with metapage write lock held; would be better to use a lock that
* doesn't block incoming searches.
*/
newbuf = _hash_getnewbuf(rel, blkno, MAIN_FORKNUM);
metap->hashm_spares[splitnum]++;
/*
* Adjust hashm_firstfree to avoid redundant searches. But don't risk
* changing it if someone moved it while we were searching bitmap pages.
*/
if (metap->hashm_firstfree == orig_firstfree)
metap->hashm_firstfree = bit + 1;
/* Write updated metapage and release lock, but not pin */
_hash_chgbufaccess(rel, metabuf, HASH_WRITE, HASH_NOLOCK);
return newbuf;
found:
/* convert bit to bit number within page */
bit += _hash_firstfreebit(freep[j]);
/* mark page "in use" in the bitmap */
SETBIT(freep, bit);
_hash_wrtbuf(rel, mapbuf);
/* Reacquire exclusive lock on the meta page */
_hash_chgbufaccess(rel, metabuf, HASH_NOLOCK, HASH_WRITE);
/* convert bit to absolute bit number */
bit += (i << BMPG_SHIFT(metap));
/* Calculate address of the recycled overflow page */
blkno = bitno_to_blkno(metap, bit);
/*
* Adjust hashm_firstfree to avoid redundant searches. But don't risk
* changing it if someone moved it while we were searching bitmap pages.
*/
if (metap->hashm_firstfree == orig_firstfree)
{
metap->hashm_firstfree = bit + 1;
/* Write updated metapage and release lock, but not pin */
_hash_chgbufaccess(rel, metabuf, HASH_WRITE, HASH_NOLOCK);
}
else
{
/* We didn't change the metapage, so no need to write */
_hash_chgbufaccess(rel, metabuf, HASH_READ, HASH_NOLOCK);
}
/* Fetch, init, and return the recycled page */
return _hash_getinitbuf(rel, blkno);
}
/*
* _hash_splitbucket -- split 'obucket' into 'obucket' and 'nbucket'
*
* We are splitting a bucket that consists of a base bucket page and zero
* or more overflow (bucket chain) pages. We must relocate tuples that
* belong in the new bucket, and compress out any free space in the old
* bucket.
*
* The caller must hold exclusive locks on both buckets to ensure that
* no one else is trying to access them (see README).
*
* The caller must hold a pin, but no lock, on the metapage buffer.
* The buffer is returned in the same state. (The metapage is only
* touched if it becomes necessary to add or remove overflow pages.)
*/
static void
_hash_splitbucket(Relation rel,
Buffer metabuf,
Bucket obucket,
Bucket nbucket,
BlockNumber start_oblkno,
BlockNumber start_nblkno,
uint32 maxbucket,
uint32 highmask,
uint32 lowmask)
{
BlockNumber oblkno;
BlockNumber nblkno;
Buffer obuf;
Buffer nbuf;
Page opage;
Page npage;
HashPageOpaque oopaque;
HashPageOpaque nopaque;
/*
* It should be okay to simultaneously write-lock pages from each bucket,
* since no one else can be trying to acquire buffer lock on pages of
* either bucket.
*/
oblkno = start_oblkno;
obuf = _hash_getbuf(rel, oblkno, HASH_WRITE, LH_BUCKET_PAGE);
opage = BufferGetPage(obuf);
oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
nblkno = start_nblkno;
nbuf = _hash_getnewbuf(rel, nblkno, MAIN_FORKNUM);
npage = BufferGetPage(nbuf);
/* initialize the new bucket's primary page */
nopaque = (HashPageOpaque) PageGetSpecialPointer(npage);
nopaque->hasho_prevblkno = InvalidBlockNumber;
nopaque->hasho_nextblkno = InvalidBlockNumber;
nopaque->hasho_bucket = nbucket;
nopaque->hasho_flag = LH_BUCKET_PAGE;
nopaque->hasho_page_id = HASHO_PAGE_ID;
/*
* Partition the tuples in the old bucket between the old bucket and the
* new bucket, advancing along the old bucket's overflow bucket chain and
* adding overflow pages to the new bucket as needed. Outer loop iterates
* once per page in old bucket.
*/
for (;;)
{
OffsetNumber ooffnum;
OffsetNumber omaxoffnum;
OffsetNumber deletable[MaxOffsetNumber];
int ndeletable = 0;
/* Scan each tuple in old page */
omaxoffnum = PageGetMaxOffsetNumber(opage);
for (ooffnum = FirstOffsetNumber;
ooffnum <= omaxoffnum;
ooffnum = OffsetNumberNext(ooffnum))
{
IndexTuple itup;
Size itemsz;
Bucket bucket;
/*
* Fetch the item's hash key (conveniently stored in the item) and
* determine which bucket it now belongs in.
*/
itup = (IndexTuple) PageGetItem(opage,
PageGetItemId(opage, ooffnum));
bucket = _hash_hashkey2bucket(_hash_get_indextuple_hashkey(itup),
maxbucket, highmask, lowmask);
if (bucket == nbucket)
{
/*
* insert the tuple into the new bucket. if it doesn't fit on
* the current page in the new bucket, we must allocate a new
* overflow page and place the tuple on that page instead.
*/
itemsz = IndexTupleDSize(*itup);
itemsz = MAXALIGN(itemsz);
if (PageGetFreeSpace(npage) < itemsz)
{
/* write out nbuf and drop lock, but keep pin */
_hash_chgbufaccess(rel, nbuf, HASH_WRITE, HASH_NOLOCK);
/* chain to a new overflow page */
nbuf = _hash_addovflpage(rel, metabuf, nbuf);
npage = BufferGetPage(nbuf);
/* we don't need nblkno or nopaque within the loop */
}
/*
* Insert tuple on new page, using _hash_pgaddtup to ensure
* correct ordering by hashkey. This is a tad inefficient
* since we may have to shuffle itempointers repeatedly.
* Possible future improvement: accumulate all the items for
* the new page and qsort them before insertion.
*/
(void) _hash_pgaddtup(rel, nbuf, itemsz, itup);
/*
* Mark tuple for deletion from old page.
*/
deletable[ndeletable++] = ooffnum;
}
else
{
/*
* the tuple stays on this page, so nothing to do.
*/
Assert(bucket == obucket);
}
}
oblkno = oopaque->hasho_nextblkno;
/*
* Done scanning this old page. If we moved any tuples, delete them
* from the old page.
*/
if (ndeletable > 0)
{
PageIndexMultiDelete(opage, deletable, ndeletable);
_hash_wrtbuf(rel, obuf);
}
else
_hash_relbuf(rel, obuf);
/* Exit loop if no more overflow pages in old bucket */
if (!BlockNumberIsValid(oblkno))
break;
/* Else, advance to next old page */
obuf = _hash_getbuf(rel, oblkno, HASH_WRITE, LH_OVERFLOW_PAGE);
opage = BufferGetPage(obuf);
oopaque = (HashPageOpaque) PageGetSpecialPointer(opage);
}
/*
* We're at the end of the old bucket chain, so we're done partitioning
* the tuples. Before quitting, call _hash_squeezebucket to ensure the
* tuples remaining in the old bucket (including the overflow pages) are
* packed as tightly as possible. The new bucket is already tight.
*/
_hash_wrtbuf(rel, nbuf);
_hash_squeezebucket(rel, obucket, start_oblkno, NULL);
}
/*
* _hash_squeezebucket(rel, bucket)
*
* Try to squeeze the tuples onto pages occurring earlier in the
* bucket chain in an attempt to free overflow pages. When we start
* the "squeezing", the page from which we start taking tuples (the
* "read" page) is the last bucket in the bucket chain and the page
* onto which we start squeezing tuples (the "write" page) is the
* first page in the bucket chain. The read page works backward and
* the write page works forward; the procedure terminates when the
* read page and write page are the same page.
*
* At completion of this procedure, it is guaranteed that all pages in
* the bucket are nonempty, unless the bucket is totally empty (in
* which case all overflow pages will be freed). The original implementation
* required that to be true on entry as well, but it's a lot easier for
* callers to leave empty overflow pages and let this guy clean it up.
*
* Caller must hold exclusive lock on the target bucket. This allows
* us to safely lock multiple pages in the bucket.
*
* Since this function is invoked in VACUUM, we provide an access strategy
* parameter that controls fetches of the bucket pages.
*/
void
_hash_squeezebucket(Relation rel,
Bucket bucket,
BlockNumber bucket_blkno,
BufferAccessStrategy bstrategy)
{
BlockNumber wblkno;
BlockNumber rblkno;
Buffer wbuf;
Buffer rbuf;
Page wpage;
Page rpage;
HashPageOpaque wopaque;
HashPageOpaque ropaque;
bool wbuf_dirty;
/*
* start squeezing into the base bucket page.
*/
wblkno = bucket_blkno;
wbuf = _hash_getbuf_with_strategy(rel,
wblkno,
HASH_WRITE,
LH_BUCKET_PAGE,
bstrategy);
wpage = BufferGetPage(wbuf);
wopaque = (HashPageOpaque) PageGetSpecialPointer(wpage);
/*
* if there aren't any overflow pages, there's nothing to squeeze.
*/
if (!BlockNumberIsValid(wopaque->hasho_nextblkno))
{
_hash_relbuf(rel, wbuf);
return;
}
/*
* Find the last page in the bucket chain by starting at the base bucket
* page and working forward. Note: we assume that a hash bucket chain is
* usually smaller than the buffer ring being used by VACUUM, else using
* the access strategy here would be counterproductive.
*/
rbuf = InvalidBuffer;
ropaque = wopaque;
do
{
rblkno = ropaque->hasho_nextblkno;
if (rbuf != InvalidBuffer)
_hash_relbuf(rel, rbuf);
rbuf = _hash_getbuf_with_strategy(rel,
rblkno,
HASH_WRITE,
LH_OVERFLOW_PAGE,
bstrategy);
rpage = BufferGetPage(rbuf);
ropaque = (HashPageOpaque) PageGetSpecialPointer(rpage);
Assert(ropaque->hasho_bucket == bucket);
} while (BlockNumberIsValid(ropaque->hasho_nextblkno));
/*
* squeeze the tuples.
*/
wbuf_dirty = false;
for (;;)
{
OffsetNumber roffnum;
OffsetNumber maxroffnum;
OffsetNumber deletable[MaxOffsetNumber];
int ndeletable = 0;
/* Scan each tuple in "read" page */
maxroffnum = PageGetMaxOffsetNumber(rpage);
for (roffnum = FirstOffsetNumber;
roffnum <= maxroffnum;
roffnum = OffsetNumberNext(roffnum))
{
IndexTuple itup;
Size itemsz;
itup = (IndexTuple) PageGetItem(rpage,
PageGetItemId(rpage, roffnum));
itemsz = IndexTupleDSize(*itup);
itemsz = MAXALIGN(itemsz);
/*
* Walk up the bucket chain, looking for a page big enough for
* this item. Exit if we reach the read page.
*/
while (PageGetFreeSpace(wpage) < itemsz)
{
Assert(!PageIsEmpty(wpage));
wblkno = wopaque->hasho_nextblkno;
Assert(BlockNumberIsValid(wblkno));
if (wbuf_dirty)
_hash_wrtbuf(rel, wbuf);
else
_hash_relbuf(rel, wbuf);
/* nothing more to do if we reached the read page */
if (rblkno == wblkno)
{
if (ndeletable > 0)
{
/* Delete tuples we already moved off read page */
PageIndexMultiDelete(rpage, deletable, ndeletable);
_hash_wrtbuf(rel, rbuf);
}
else
_hash_relbuf(rel, rbuf);
return;
}
wbuf = _hash_getbuf_with_strategy(rel,
wblkno,
HASH_WRITE,
LH_OVERFLOW_PAGE,
bstrategy);
wpage = BufferGetPage(wbuf);
wopaque = (HashPageOpaque) PageGetSpecialPointer(wpage);
Assert(wopaque->hasho_bucket == bucket);
wbuf_dirty = false;
}
/*
* we have found room so insert on the "write" page, being careful
* to preserve hashkey ordering. (If we insert many tuples into
* the same "write" page it would be worth qsort'ing instead of
* doing repeated _hash_pgaddtup.)
*/
(void) _hash_pgaddtup(rel, wbuf, itemsz, itup);
wbuf_dirty = true;
/* remember tuple for deletion from "read" page */
deletable[ndeletable++] = roffnum;
}
/*
* If we reach here, there are no live tuples on the "read" page ---
* it was empty when we got to it, or we moved them all. So we can
* just free the page without bothering with deleting tuples
* individually. Then advance to the previous "read" page.
*
* Tricky point here: if our read and write pages are adjacent in the
* bucket chain, our write lock on wbuf will conflict with
* _hash_freeovflpage's attempt to update the sibling links of the
* removed page. However, in that case we are done anyway, so we can
* simply drop the write lock before calling _hash_freeovflpage.
*/
rblkno = ropaque->hasho_prevblkno;
Assert(BlockNumberIsValid(rblkno));
/* are we freeing the page adjacent to wbuf? */
if (rblkno == wblkno)
{
/* yes, so release wbuf lock first */
if (wbuf_dirty)
_hash_wrtbuf(rel, wbuf);
else
_hash_relbuf(rel, wbuf);
/* free this overflow page (releases rbuf) */
_hash_freeovflpage(rel, rbuf, bstrategy);
/* done */
return;
}
/* free this overflow page, then get the previous one */
_hash_freeovflpage(rel, rbuf, bstrategy);
rbuf = _hash_getbuf_with_strategy(rel,
rblkno,
HASH_WRITE,
LH_OVERFLOW_PAGE,
bstrategy);
rpage = BufferGetPage(rbuf);
ropaque = (HashPageOpaque) PageGetSpecialPointer(rpage);
Assert(ropaque->hasho_bucket == bucket);
}
/* NOTREACHED */
}
/*
* Bulk deletion of all index entries pointing to a set of heap tuples.
* The set of target tuples is specified via a callback routine that tells
* whether any given heap tuple (identified by ItemPointer) is being deleted.
*
* This function also deletes the tuples that are moved by split to other
* bucket.
*
* Result: a palloc'd struct containing statistical info for VACUUM displays.
*/
IndexBulkDeleteResult *
hashbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
IndexBulkDeleteCallback callback, void *callback_state)
{
Relation rel = info->index;
double tuples_removed;
double num_index_tuples;
double orig_ntuples;
Bucket orig_maxbucket;
Bucket cur_maxbucket;
Bucket cur_bucket;
Buffer metabuf;
HashMetaPage metap;
HashMetaPageData local_metapage;
tuples_removed = 0;
num_index_tuples = 0;
/*
* Read the metapage to fetch original bucket and tuple counts. Also, we
* keep a copy of the last-seen metapage so that we can use its
* hashm_spares[] values to compute bucket page addresses. This is a bit
* hokey but perfectly safe, since the interesting entries in the spares
* array cannot change under us; and it beats rereading the metapage for
* each bucket.
*/
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
orig_maxbucket = metap->hashm_maxbucket;
orig_ntuples = metap->hashm_ntuples;
memcpy(&local_metapage, metap, sizeof(local_metapage));
_hash_relbuf(rel, metabuf);
/* Scan the buckets that we know exist */
cur_bucket = 0;
cur_maxbucket = orig_maxbucket;
loop_top:
while (cur_bucket <= cur_maxbucket)
{
BlockNumber bucket_blkno;
BlockNumber blkno;
Buffer bucket_buf;
Buffer buf;
HashPageOpaque bucket_opaque;
Page page;
bool split_cleanup = false;
/* Get address of bucket's start page */
bucket_blkno = BUCKET_TO_BLKNO(&local_metapage, cur_bucket);
blkno = bucket_blkno;
/*
* We need to acquire a cleanup lock on the primary bucket page to out
* wait concurrent scans before deleting the dead tuples.
*/
buf = ReadBufferExtended(rel, MAIN_FORKNUM, blkno, RBM_NORMAL, info->strategy);
LockBufferForCleanup(buf);
_hash_checkpage(rel, buf, LH_BUCKET_PAGE);
page = BufferGetPage(buf);
bucket_opaque = (HashPageOpaque) PageGetSpecialPointer(page);
/*
* If the bucket contains tuples that are moved by split, then we need
* to delete such tuples. We can't delete such tuples if the split
* operation on bucket is not finished as those are needed by scans.
*/
if (!H_BUCKET_BEING_SPLIT(bucket_opaque) &&
H_NEEDS_SPLIT_CLEANUP(bucket_opaque))
split_cleanup = true;
bucket_buf = buf;
hashbucketcleanup(rel, cur_bucket, bucket_buf, blkno, info->strategy,
local_metapage.hashm_maxbucket,
local_metapage.hashm_highmask,
local_metapage.hashm_lowmask, &tuples_removed,
&num_index_tuples, split_cleanup,
callback, callback_state);
_hash_dropbuf(rel, bucket_buf);
/* Advance to next bucket */
cur_bucket++;
}
/* Write-lock metapage and check for split since we started */
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_WRITE, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
if (cur_maxbucket != metap->hashm_maxbucket)
{
/* There's been a split, so process the additional bucket(s) */
cur_maxbucket = metap->hashm_maxbucket;
memcpy(&local_metapage, metap, sizeof(local_metapage));
_hash_relbuf(rel, metabuf);
goto loop_top;
}
/* Okay, we're really done. Update tuple count in metapage. */
if (orig_maxbucket == metap->hashm_maxbucket &&
orig_ntuples == metap->hashm_ntuples)
{
/*
* No one has split or inserted anything since start of scan, so
* believe our count as gospel.
*/
metap->hashm_ntuples = num_index_tuples;
}
else
{
/*
* Otherwise, our count is untrustworthy since we may have
* double-scanned tuples in split buckets. Proceed by dead-reckoning.
* (Note: we still return estimated_count = false, because using this
* count is better than not updating reltuples at all.)
*/
if (metap->hashm_ntuples > tuples_removed)
metap->hashm_ntuples -= tuples_removed;
else
metap->hashm_ntuples = 0;
num_index_tuples = metap->hashm_ntuples;
}
_hash_wrtbuf(rel, metabuf);
/* return statistics */
if (stats == NULL)
stats = (IndexBulkDeleteResult *) palloc0(sizeof(IndexBulkDeleteResult));
stats->estimated_count = false;
stats->num_index_tuples = num_index_tuples;
stats->tuples_removed += tuples_removed;
/* hashvacuumcleanup will fill in num_pages */
return stats;
}
/*
* Bulk deletion of all index entries pointing to a set of heap tuples.
* The set of target tuples is specified via a callback routine that tells
* whether any given heap tuple (identified by ItemPointer) is being deleted.
*
* Result: a palloc'd struct containing statistical info for VACUUM displays.
*/
IndexBulkDeleteResult *
hashbulkdelete(IndexVacuumInfo *info, IndexBulkDeleteResult *stats,
IndexBulkDeleteCallback callback, void *callback_state)
{
Relation rel = info->index;
double tuples_removed;
double num_index_tuples;
double orig_ntuples;
Bucket orig_maxbucket;
Bucket cur_maxbucket;
Bucket cur_bucket;
Buffer metabuf;
HashMetaPage metap;
HashMetaPageData local_metapage;
tuples_removed = 0;
num_index_tuples = 0;
/*
* Read the metapage to fetch original bucket and tuple counts. Also, we
* keep a copy of the last-seen metapage so that we can use its
* hashm_spares[] values to compute bucket page addresses. This is a bit
* hokey but perfectly safe, since the interesting entries in the spares
* array cannot change under us; and it beats rereading the metapage for
* each bucket.
*/
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_READ, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
orig_maxbucket = metap->hashm_maxbucket;
orig_ntuples = metap->hashm_ntuples;
memcpy(&local_metapage, metap, sizeof(local_metapage));
_hash_relbuf(rel, metabuf);
/* Scan the buckets that we know exist */
cur_bucket = 0;
cur_maxbucket = orig_maxbucket;
loop_top:
while (cur_bucket <= cur_maxbucket)
{
BlockNumber bucket_blkno;
BlockNumber blkno;
bool bucket_dirty = false;
/* Get address of bucket's start page */
bucket_blkno = BUCKET_TO_BLKNO(&local_metapage, cur_bucket);
/* Exclusive-lock the bucket so we can shrink it */
_hash_getlock(rel, bucket_blkno, HASH_EXCLUSIVE);
/* Shouldn't have any active scans locally, either */
if (_hash_has_active_scan(rel, cur_bucket))
elog(ERROR, "hash index has active scan during VACUUM");
/* Scan each page in bucket */
blkno = bucket_blkno;
while (BlockNumberIsValid(blkno))
{
Buffer buf;
Page page;
HashPageOpaque opaque;
OffsetNumber offno;
OffsetNumber maxoffno;
OffsetNumber deletable[MaxOffsetNumber];
int ndeletable = 0;
vacuum_delay_point();
buf = _hash_getbuf_with_strategy(rel, blkno, HASH_WRITE,
LH_BUCKET_PAGE | LH_OVERFLOW_PAGE,
info->strategy);
page = BufferGetPage(buf);
opaque = (HashPageOpaque) PageGetSpecialPointer(page);
Assert(opaque->hasho_bucket == cur_bucket);
/* Scan each tuple in page */
maxoffno = PageGetMaxOffsetNumber(page);
for (offno = FirstOffsetNumber;
offno <= maxoffno;
offno = OffsetNumberNext(offno))
{
IndexTuple itup;
ItemPointer htup;
itup = (IndexTuple) PageGetItem(page,
PageGetItemId(page, offno));
htup = &(itup->t_tid);
if (callback(htup, callback_state))
{
/* mark the item for deletion */
deletable[ndeletable++] = offno;
tuples_removed += 1;
}
else
num_index_tuples += 1;
}
/*
* Apply deletions and write page if needed, advance to next page.
*/
blkno = opaque->hasho_nextblkno;
if (ndeletable > 0)
{
PageIndexMultiDelete(page, deletable, ndeletable);
_hash_wrtbuf(rel, buf);
bucket_dirty = true;
}
else
_hash_relbuf(rel, buf);
}
/* If we deleted anything, try to compact free space */
if (bucket_dirty)
_hash_squeezebucket(rel, cur_bucket, bucket_blkno,
info->strategy);
/* Release bucket lock */
_hash_droplock(rel, bucket_blkno, HASH_EXCLUSIVE);
/* Advance to next bucket */
cur_bucket++;
}
/* Write-lock metapage and check for split since we started */
metabuf = _hash_getbuf(rel, HASH_METAPAGE, HASH_WRITE, LH_META_PAGE);
metap = HashPageGetMeta(BufferGetPage(metabuf));
if (cur_maxbucket != metap->hashm_maxbucket)
{
/* There's been a split, so process the additional bucket(s) */
cur_maxbucket = metap->hashm_maxbucket;
memcpy(&local_metapage, metap, sizeof(local_metapage));
_hash_relbuf(rel, metabuf);
goto loop_top;
}
/* Okay, we're really done. Update tuple count in metapage. */
if (orig_maxbucket == metap->hashm_maxbucket &&
orig_ntuples == metap->hashm_ntuples)
{
/*
* No one has split or inserted anything since start of scan, so
* believe our count as gospel.
*/
metap->hashm_ntuples = num_index_tuples;
}
else
{
/*
* Otherwise, our count is untrustworthy since we may have
* double-scanned tuples in split buckets. Proceed by dead-reckoning.
* (Note: we still return estimated_count = false, because using this
* count is better than not updating reltuples at all.)
*/
if (metap->hashm_ntuples > tuples_removed)
metap->hashm_ntuples -= tuples_removed;
else
metap->hashm_ntuples = 0;
num_index_tuples = metap->hashm_ntuples;
}
_hash_wrtbuf(rel, metabuf);
/* return statistics */
if (stats == NULL)
stats = (IndexBulkDeleteResult *) palloc0(sizeof(IndexBulkDeleteResult));
stats->estimated_count = false;
stats->num_index_tuples = num_index_tuples;
stats->tuples_removed += tuples_removed;
/* hashvacuumcleanup will fill in num_pages */
return stats;
}
/*
* Helper function to perform deletion of index entries from a bucket.
*
* This function expects that the caller has acquired a cleanup lock on the
* primary bucket page, and will return with a write lock again held on the
* primary bucket page. The lock won't necessarily be held continuously,
* though, because we'll release it when visiting overflow pages.
*
* It would be very bad if this function cleaned a page while some other
* backend was in the midst of scanning it, because hashgettuple assumes
* that the next valid TID will be greater than or equal to the current
* valid TID. There can't be any concurrent scans in progress when we first
* enter this function because of the cleanup lock we hold on the primary
* bucket page, but as soon as we release that lock, there might be. We
* handle that by conspiring to prevent those scans from passing our cleanup
* scan. To do that, we lock the next page in the bucket chain before
* releasing the lock on the previous page. (This type of lock chaining is
* not ideal, so we might want to look for a better solution at some point.)
*
* We need to retain a pin on the primary bucket to ensure that no concurrent
* split can start.
*/
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)
{
BlockNumber blkno;
Buffer buf;
Bucket new_bucket PG_USED_FOR_ASSERTS_ONLY = InvalidBucket;
bool bucket_dirty = false;
blkno = bucket_blkno;
buf = bucket_buf;
if (split_cleanup)
new_bucket = _hash_get_newbucket_from_oldbucket(rel, cur_bucket,
lowmask, maxbucket);
/* Scan each page in bucket */
for (;;)
{
HashPageOpaque opaque;
OffsetNumber offno;
OffsetNumber maxoffno;
Buffer next_buf;
Page page;
OffsetNumber deletable[MaxOffsetNumber];
int ndeletable = 0;
bool retain_pin = false;
bool curr_page_dirty = false;
vacuum_delay_point();
page = BufferGetPage(buf);
opaque = (HashPageOpaque) PageGetSpecialPointer(page);
/* Scan each tuple in page */
maxoffno = PageGetMaxOffsetNumber(page);
for (offno = FirstOffsetNumber;
offno <= maxoffno;
offno = OffsetNumberNext(offno))
{
ItemPointer htup;
IndexTuple itup;
Bucket bucket;
bool kill_tuple = false;
itup = (IndexTuple) PageGetItem(page,
PageGetItemId(page, offno));
htup = &(itup->t_tid);
/*
* To remove the dead tuples, we strictly want to rely on results
* of callback function. refer btvacuumpage for detailed reason.
*/
if (callback && callback(htup, callback_state))
{
kill_tuple = true;
if (tuples_removed)
*tuples_removed += 1;
}
else if (split_cleanup)
{
/* delete the tuples that are moved by split. */
bucket = _hash_hashkey2bucket(_hash_get_indextuple_hashkey(itup),
maxbucket,
highmask,
lowmask);
/* mark the item for deletion */
if (bucket != cur_bucket)
{
/*
* We expect tuples to either belong to curent bucket or
* new_bucket. This is ensured because we don't allow
* further splits from bucket that contains garbage. See
* comments in _hash_expandtable.
*/
Assert(bucket == new_bucket);
kill_tuple = true;
}
}
if (kill_tuple)
{
/* mark the item for deletion */
deletable[ndeletable++] = offno;
}
else
{
/* we're keeping it, so count it */
if (num_index_tuples)
*num_index_tuples += 1;
}
}
/* retain the pin on primary bucket page till end of bucket scan */
if (blkno == bucket_blkno)
retain_pin = true;
else
retain_pin = false;
blkno = opaque->hasho_nextblkno;
/*
* Apply deletions, advance to next page and write page if needed.
*/
if (ndeletable > 0)
{
PageIndexMultiDelete(page, deletable, ndeletable);
bucket_dirty = true;
curr_page_dirty = true;
}
/* bail out if there are no more pages to scan. */
if (!BlockNumberIsValid(blkno))
break;
next_buf = _hash_getbuf_with_strategy(rel, blkno, HASH_WRITE,
LH_OVERFLOW_PAGE,
bstrategy);
/*
* release the lock on previous page after acquiring the lock on next
* page
*/
if (curr_page_dirty)
{
if (retain_pin)
_hash_chgbufaccess(rel, buf, HASH_WRITE, HASH_NOLOCK);
else
_hash_wrtbuf(rel, buf);
curr_page_dirty = false;
}
else if (retain_pin)
_hash_chgbufaccess(rel, buf, HASH_READ, HASH_NOLOCK);
else
_hash_relbuf(rel, buf);
buf = next_buf;
}
/*
* lock the bucket page to clear the garbage flag and squeeze the bucket.
* if the current buffer is same as bucket buffer, then we already have
* lock on bucket page.
*/
if (buf != bucket_buf)
{
_hash_relbuf(rel, buf);
_hash_chgbufaccess(rel, bucket_buf, HASH_NOLOCK, HASH_WRITE);
}
/*
* Clear the garbage flag from bucket after deleting the tuples that are
* moved by split. We purposefully clear the flag before squeeze bucket,
* so that after restart, vacuum shouldn't again try to delete the moved
* by split tuples.
*/
if (split_cleanup)
{
HashPageOpaque bucket_opaque;
Page page;
page = BufferGetPage(bucket_buf);
bucket_opaque = (HashPageOpaque) PageGetSpecialPointer(page);
bucket_opaque->hasho_flag &= ~LH_BUCKET_NEEDS_SPLIT_CLEANUP;
}
/*
* If we have deleted anything, try to compact free space. For squeezing
* the bucket, we must have a cleanup lock, else it can impact the
* ordering of tuples for a scan that has started before it.
*/
if (bucket_dirty && IsBufferCleanupOK(bucket_buf))
_hash_squeezebucket(rel, cur_bucket, bucket_blkno, bucket_buf,
bstrategy);
else
_hash_chgbufaccess(rel, bucket_buf, HASH_WRITE, HASH_NOLOCK);
}
