/*
* Write XLOG record describing a page update. The update can include any
* number of deletions and/or insertions of tuples on a single index page.
*
* If this update inserts a downlink for a split page, also record that
* the F_FOLLOW_RIGHT flag on the child page is cleared and NSN set.
*
* Note that both the todelete array and the tuples are marked as belonging
* to the target buffer; they need not be stored in XLOG if XLogInsert decides
* to log the whole buffer contents instead.
*/
XLogRecPtr
gistXLogUpdate(Buffer buffer,
OffsetNumber *todelete, int ntodelete,
IndexTuple *itup, int ituplen,
Buffer leftchildbuf)
{
gistxlogPageUpdate xlrec;
int i;
XLogRecPtr recptr;
xlrec.ntodelete = ntodelete;
xlrec.ntoinsert = ituplen;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, sizeof(gistxlogPageUpdate));
XLogRegisterBuffer(0, buffer, REGBUF_STANDARD);
XLogRegisterBufData(0, (char *) todelete, sizeof(OffsetNumber) * ntodelete);
/* new tuples */
for (i = 0; i < ituplen; i++)
XLogRegisterBufData(0, (char *) (itup[i]), IndexTupleSize(itup[i]));
/*
* Include a full page image of the child buf. (only necessary if a
* checkpoint happened since the child page was split)
*/
if (BufferIsValid(leftchildbuf))
XLogRegisterBuffer(1, leftchildbuf, REGBUF_STANDARD);
recptr = XLogInsert(RM_GIST_ID, XLOG_GIST_PAGE_UPDATE);
return recptr;
}
/*
* Write WAL record of a page split.
*/
XLogRecPtr
gistXLogSplit(bool page_is_leaf,
SplitedPageLayout *dist,
BlockNumber origrlink, GistNSN orignsn,
Buffer leftchildbuf, bool markfollowright)
{
gistxlogPageSplit xlrec;
SplitedPageLayout *ptr;
int npage = 0;
XLogRecPtr recptr;
int i;
for (ptr = dist; ptr; ptr = ptr->next)
npage++;
xlrec.origrlink = origrlink;
xlrec.orignsn = orignsn;
xlrec.origleaf = page_is_leaf;
xlrec.npage = (uint16) npage;
xlrec.markfollowright = markfollowright;
XLogBeginInsert();
/*
* Include a full page image of the child buf. (only necessary if a
* checkpoint happened since the child page was split)
*/
if (BufferIsValid(leftchildbuf))
XLogRegisterBuffer(0, leftchildbuf, REGBUF_STANDARD);
/*
* NOTE: We register a lot of data. The caller must've called
* XLogEnsureRecordSpace() to prepare for that. We cannot do it here,
* because we're already in a critical section. If you change the number
* of buffer or data registrations here, make sure you modify the
* XLogEnsureRecordSpace() calls accordingly!
*/
XLogRegisterData((char *) &xlrec, sizeof(gistxlogPageSplit));
i = 1;
for (ptr = dist; ptr; ptr = ptr->next)
{
XLogRegisterBuffer(i, ptr->buffer, REGBUF_WILL_INIT);
XLogRegisterBufData(i, (char *) &(ptr->block.num), sizeof(int));
XLogRegisterBufData(i, (char *) ptr->list, ptr->lenlist);
i++;
}
recptr = XLogInsert(RM_GIST_ID, XLOG_GIST_PAGE_SPLIT);
return recptr;
}
/*
* Place tuple on page and fills WAL record
*
* If the tuple doesn't fit, returns false without modifying the page.
*
* On insertion to an internal node, in addition to inserting the given item,
* the downlink of the existing item at 'off' is updated to point to
* 'updateblkno'.
*
* On INSERTED, registers the buffer as buffer ID 0, with data.
* On SPLIT, returns rdata that represents the split pages in *prdata.
*/
static GinPlaceToPageRC
entryPlaceToPage(GinBtree btree, Buffer buf, GinBtreeStack *stack,
void *insertPayload, BlockNumber updateblkno,
Page *newlpage, Page *newrpage)
{
GinBtreeEntryInsertData *insertData = insertPayload;
Page page = BufferGetPage(buf, NULL, NULL, BGP_NO_SNAPSHOT_TEST);
OffsetNumber off = stack->off;
OffsetNumber placed;
/* this must be static so it can be returned to caller. */
static ginxlogInsertEntry data;
/* quick exit if it doesn't fit */
if (!entryIsEnoughSpace(btree, buf, off, insertData))
{
entrySplitPage(btree, buf, stack, insertPayload, updateblkno,
newlpage, newrpage);
return SPLIT;
}
START_CRIT_SECTION();
entryPreparePage(btree, page, off, insertData, updateblkno);
placed = PageAddItem(page,
(Item) insertData->entry,
IndexTupleSize(insertData->entry),
off, false, false);
if (placed != off)
elog(ERROR, "failed to add item to index page in \"%s\"",
RelationGetRelationName(btree->index));
if (RelationNeedsWAL(btree->index))
{
data.isDelete = insertData->isDelete;
data.offset = off;
XLogBeginInsert();
XLogRegisterBuffer(0, buf, REGBUF_STANDARD);
XLogRegisterBufData(0, (char *) &data,
offsetof(ginxlogInsertEntry, tuple));
XLogRegisterBufData(0, (char *) insertData->entry,
IndexTupleSize(insertData->entry));
}
return INSERTED;
}
/*
* Perform data insertion after beginPlaceToPage has decided it will fit.
*
* This is invoked within a critical section, and XLOG record creation (if
* needed) is already started. The target buffer is registered in slot 0.
*/
static void
entryExecPlaceToPage(GinBtree btree, Buffer buf, GinBtreeStack *stack,
void *insertPayload, BlockNumber updateblkno,
void *ptp_workspace)
{
GinBtreeEntryInsertData *insertData = insertPayload;
Page page = BufferGetPage(buf);
OffsetNumber off = stack->off;
OffsetNumber placed;
entryPreparePage(btree, page, off, insertData, updateblkno);
placed = PageAddItem(page,
(Item) insertData->entry,
IndexTupleSize(insertData->entry),
off, false, false);
if (placed != off)
elog(ERROR, "failed to add item to index page in \"%s\"",
RelationGetRelationName(btree->index));
if (RelationNeedsWAL(btree->index))
{
/*
* This must be static, because it has to survive until XLogInsert,
* and we can't palloc here. Ugly, but the XLogInsert infrastructure
* isn't reentrant anyway.
*/
static ginxlogInsertEntry data;
data.isDelete = insertData->isDelete;
data.offset = off;
XLogRegisterBufData(0, (char *) &data,
offsetof(ginxlogInsertEntry, tuple));
XLogRegisterBufData(0, (char *) insertData->entry,
IndexTupleSize(insertData->entry));
}
}
/*
* 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);
}
/*
* Write the index tuples contained in *collector into the index's
* pending list.
*
* Function guarantees that all these tuples will be inserted consecutively,
* preserving order
*/
void
ginHeapTupleFastInsert(GinState *ginstate, GinTupleCollector *collector)
{
Relation index = ginstate->index;
Buffer metabuffer;
Page metapage;
GinMetaPageData *metadata = NULL;
Buffer buffer = InvalidBuffer;
Page page = NULL;
ginxlogUpdateMeta data;
bool separateList = false;
bool needCleanup = false;
int cleanupSize;
bool needWal;
if (collector->ntuples == 0)
return;
needWal = RelationNeedsWAL(index);
data.node = index->rd_node;
data.ntuples = 0;
data.newRightlink = data.prevTail = InvalidBlockNumber;
metabuffer = ReadBuffer(index, GIN_METAPAGE_BLKNO);
metapage = BufferGetPage(metabuffer);
if (collector->sumsize + collector->ntuples * sizeof(ItemIdData) > GinListPageSize)
{
/*
* Total size is greater than one page => make sublist
*/
separateList = true;
}
else
{
LockBuffer(metabuffer, GIN_EXCLUSIVE);
metadata = GinPageGetMeta(metapage);
if (metadata->head == InvalidBlockNumber ||
collector->sumsize + collector->ntuples * sizeof(ItemIdData) > metadata->tailFreeSize)
{
/*
* Pending list is empty or total size is greater than freespace
* on tail page => make sublist
*
* We unlock metabuffer to keep high concurrency
*/
separateList = true;
LockBuffer(metabuffer, GIN_UNLOCK);
}
}
if (separateList)
{
/*
* We should make sublist separately and append it to the tail
*/
GinMetaPageData sublist;
memset(&sublist, 0, sizeof(GinMetaPageData));
makeSublist(index, collector->tuples, collector->ntuples, &sublist);
if (needWal)
XLogBeginInsert();
/*
* metapage was unlocked, see above
*/
LockBuffer(metabuffer, GIN_EXCLUSIVE);
metadata = GinPageGetMeta(metapage);
if (metadata->head == InvalidBlockNumber)
{
/*
* Main list is empty, so just insert sublist as main list
*/
START_CRIT_SECTION();
metadata->head = sublist.head;
metadata->tail = sublist.tail;
metadata->tailFreeSize = sublist.tailFreeSize;
metadata->nPendingPages = sublist.nPendingPages;
metadata->nPendingHeapTuples = sublist.nPendingHeapTuples;
}
else
{
/*
* Merge lists
*/
data.prevTail = metadata->tail;
data.newRightlink = sublist.head;
buffer = ReadBuffer(index, metadata->tail);
LockBuffer(buffer, GIN_EXCLUSIVE);
page = BufferGetPage(buffer);
Assert(GinPageGetOpaque(page)->rightlink == InvalidBlockNumber);
START_CRIT_SECTION();
GinPageGetOpaque(page)->rightlink = sublist.head;
MarkBufferDirty(buffer);
metadata->tail = sublist.tail;
metadata->tailFreeSize = sublist.tailFreeSize;
metadata->nPendingPages += sublist.nPendingPages;
metadata->nPendingHeapTuples += sublist.nPendingHeapTuples;
if (needWal)
XLogRegisterBuffer(1, buffer, REGBUF_STANDARD);
}
}
else
{
/*
* Insert into tail page. Metapage is already locked
*/
OffsetNumber l,
off;
int i,
tupsize;
char *ptr;
char *collectordata;
buffer = ReadBuffer(index, metadata->tail);
LockBuffer(buffer, GIN_EXCLUSIVE);
page = BufferGetPage(buffer);
off = (PageIsEmpty(page)) ? FirstOffsetNumber :
OffsetNumberNext(PageGetMaxOffsetNumber(page));
collectordata = ptr = (char *) palloc(collector->sumsize);
data.ntuples = collector->ntuples;
if (needWal)
XLogBeginInsert();
START_CRIT_SECTION();
/*
* Increase counter of heap tuples
*/
Assert(GinPageGetOpaque(page)->maxoff <= metadata->nPendingHeapTuples);
GinPageGetOpaque(page)->maxoff++;
metadata->nPendingHeapTuples++;
for (i = 0; i < collector->ntuples; i++)
{
tupsize = IndexTupleSize(collector->tuples[i]);
l = PageAddItem(page, (Item) collector->tuples[i], tupsize, off, false, false);
if (l == InvalidOffsetNumber)
elog(ERROR, "failed to add item to index page in \"%s\"",
RelationGetRelationName(index));
memcpy(ptr, collector->tuples[i], tupsize);
ptr += tupsize;
off++;
}
Assert((ptr - collectordata) <= collector->sumsize);
if (needWal)
{
XLogRegisterBuffer(1, buffer, REGBUF_STANDARD);
XLogRegisterBufData(1, collectordata, collector->sumsize);
}
metadata->tailFreeSize = PageGetExactFreeSpace(page);
MarkBufferDirty(buffer);
}
/*
* Write metabuffer, make xlog entry
*/
MarkBufferDirty(metabuffer);
if (needWal)
{
XLogRecPtr recptr;
memcpy(&data.metadata, metadata, sizeof(GinMetaPageData));
XLogRegisterBuffer(0, metabuffer, REGBUF_WILL_INIT);
XLogRegisterData((char *) &data, sizeof(ginxlogUpdateMeta));
recptr = XLogInsert(RM_GIN_ID, XLOG_GIN_UPDATE_META_PAGE);
PageSetLSN(metapage, recptr);
if (buffer != InvalidBuffer)
{
PageSetLSN(page, recptr);
}
}
if (buffer != InvalidBuffer)
UnlockReleaseBuffer(buffer);
/*
* Force pending list cleanup when it becomes too long. And,
* ginInsertCleanup could take significant amount of time, so we prefer to
* call it when it can do all the work in a single collection cycle. In
* non-vacuum mode, it shouldn't require maintenance_work_mem, so fire it
* while pending list is still small enough to fit into
* gin_pending_list_limit.
*
* ginInsertCleanup() should not be called inside our CRIT_SECTION.
*/
cleanupSize = GinGetPendingListCleanupSize(index);
if (metadata->nPendingPages * GIN_PAGE_FREESIZE > cleanupSize * 1024L)
needCleanup = true;
UnlockReleaseBuffer(metabuffer);
END_CRIT_SECTION();
if (needCleanup)
ginInsertCleanup(ginstate, false, true, NULL);
}
/*
* Build a pending-list page from the given array of tuples, and write it out.
*
* Returns amount of free space left on the page.
*/
static int32
writeListPage(Relation index, Buffer buffer,
IndexTuple *tuples, int32 ntuples, BlockNumber rightlink)
{
Page page = BufferGetPage(buffer);
int32 i,
freesize,
size = 0;
OffsetNumber l,
off;
char *workspace;
char *ptr;
/* workspace could be a local array; we use palloc for alignment */
workspace = palloc(BLCKSZ);
START_CRIT_SECTION();
GinInitBuffer(buffer, GIN_LIST);
off = FirstOffsetNumber;
ptr = workspace;
for (i = 0; i < ntuples; i++)
{
int this_size = IndexTupleSize(tuples[i]);
memcpy(ptr, tuples[i], this_size);
ptr += this_size;
size += this_size;
l = PageAddItem(page, (Item) tuples[i], this_size, off, false, false);
if (l == InvalidOffsetNumber)
elog(ERROR, "failed to add item to index page in \"%s\"",
RelationGetRelationName(index));
off++;
}
Assert(size <= BLCKSZ); /* else we overran workspace */
GinPageGetOpaque(page)->rightlink = rightlink;
/*
* tail page may contain only whole row(s) or final part of row placed on
* previous pages (a "row" here meaning all the index tuples generated for
* one heap tuple)
*/
if (rightlink == InvalidBlockNumber)
{
GinPageSetFullRow(page);
GinPageGetOpaque(page)->maxoff = 1;
}
else
{
GinPageGetOpaque(page)->maxoff = 0;
}
MarkBufferDirty(buffer);
if (RelationNeedsWAL(index))
{
ginxlogInsertListPage data;
XLogRecPtr recptr;
data.rightlink = rightlink;
data.ntuples = ntuples;
XLogBeginInsert();
XLogRegisterData((char *) &data, sizeof(ginxlogInsertListPage));
XLogRegisterBuffer(0, buffer, REGBUF_WILL_INIT);
XLogRegisterBufData(0, workspace, size);
recptr = XLogInsert(RM_GIN_ID, XLOG_GIN_INSERT_LISTPAGE);
PageSetLSN(page, recptr);
}
/* get free space before releasing buffer */
freesize = PageGetExactFreeSpace(page);
UnlockReleaseBuffer(buffer);
END_CRIT_SECTION();
pfree(workspace);
return freesize;
}
/*
* Insert an index tuple into the index relation. The revmap is updated to
* mark the range containing the given page as pointing to the inserted entry.
* A WAL record is written.
*
* The buffer, if valid, is first checked for free space to insert the new
* entry; if there isn't enough, a new buffer is obtained and pinned. No
* buffer lock must be held on entry, no buffer lock is held on exit.
*
* Return value is the offset number where the tuple was inserted.
*/
OffsetNumber
brin_doinsert(Relation idxrel, BlockNumber pagesPerRange,
BrinRevmap *revmap, Buffer *buffer, BlockNumber heapBlk,
BrinTuple *tup, Size itemsz)
{
Page page;
BlockNumber blk;
OffsetNumber off;
Buffer revmapbuf;
ItemPointerData tid;
bool extended;
Assert(itemsz == MAXALIGN(itemsz));
/* If the item is oversized, don't even bother. */
if (itemsz > BrinMaxItemSize)
{
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("index row size %lu exceeds maximum %lu for index \"%s\"",
(unsigned long) itemsz,
(unsigned long) BrinMaxItemSize,
RelationGetRelationName(idxrel))));
return InvalidOffsetNumber; /* keep compiler quiet */
}
/* Make sure the revmap is long enough to contain the entry we need */
brinRevmapExtend(revmap, heapBlk);
/*
* Acquire lock on buffer supplied by caller, if any. If it doesn't have
* enough space, unpin it to obtain a new one below.
*/
if (BufferIsValid(*buffer))
{
/*
* It's possible that another backend (or ourselves!) extended the
* revmap over the page we held a pin on, so we cannot assume that
* it's still a regular page.
*/
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
if (br_page_get_freespace(BufferGetPage(*buffer)) < itemsz)
{
UnlockReleaseBuffer(*buffer);
*buffer = InvalidBuffer;
}
}
/*
* If we still don't have a usable buffer, have brin_getinsertbuffer
* obtain one for us.
*/
if (!BufferIsValid(*buffer))
{
do
*buffer = brin_getinsertbuffer(idxrel, InvalidBuffer, itemsz, &extended);
while (!BufferIsValid(*buffer));
}
else
extended = false;
/* Now obtain lock on revmap buffer */
revmapbuf = brinLockRevmapPageForUpdate(revmap, heapBlk);
page = BufferGetPage(*buffer);
blk = BufferGetBlockNumber(*buffer);
/* Execute the actual insertion */
START_CRIT_SECTION();
if (extended)
brin_page_init(BufferGetPage(*buffer), BRIN_PAGETYPE_REGULAR);
off = PageAddItem(page, (Item) tup, itemsz, InvalidOffsetNumber,
false, false);
if (off == InvalidOffsetNumber)
elog(ERROR, "could not insert new index tuple to page");
MarkBufferDirty(*buffer);
BRIN_elog((DEBUG2, "inserted tuple (%u,%u) for range starting at %u",
blk, off, heapBlk));
ItemPointerSet(&tid, blk, off);
brinSetHeapBlockItemptr(revmapbuf, pagesPerRange, heapBlk, tid);
MarkBufferDirty(revmapbuf);
/* XLOG stuff */
if (RelationNeedsWAL(idxrel))
{
xl_brin_insert xlrec;
XLogRecPtr recptr;
uint8 info;
info = XLOG_BRIN_INSERT | (extended ? XLOG_BRIN_INIT_PAGE : 0);
xlrec.heapBlk = heapBlk;
xlrec.pagesPerRange = pagesPerRange;
xlrec.offnum = off;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfBrinInsert);
XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD | (extended ? REGBUF_WILL_INIT : 0));
XLogRegisterBufData(0, (char *) tup, itemsz);
XLogRegisterBuffer(1, revmapbuf, 0);
recptr = XLogInsert(RM_BRIN_ID, info);
PageSetLSN(page, recptr);
PageSetLSN(BufferGetPage(revmapbuf), recptr);
}
END_CRIT_SECTION();
/* Tuple is firmly on buffer; we can release our locks */
LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
LockBuffer(revmapbuf, BUFFER_LOCK_UNLOCK);
if (extended)
FreeSpaceMapVacuum(idxrel);
return off;
}
/*
* Update tuple origtup (size origsz), located in offset oldoff of buffer
* oldbuf, to newtup (size newsz) as summary tuple for the page range starting
* at heapBlk. oldbuf must not be locked on entry, and is not locked at exit.
*
* If samepage is true, attempt to put the new tuple in the same page, but if
* there's no room, use some other one.
*
* If the update is successful, return true; the revmap is updated to point to
* the new tuple. If the update is not done for whatever reason, return false.
* Caller may retry the update if this happens.
*/
bool
brin_doupdate(Relation idxrel, BlockNumber pagesPerRange,
BrinRevmap *revmap, BlockNumber heapBlk,
Buffer oldbuf, OffsetNumber oldoff,
const BrinTuple *origtup, Size origsz,
const BrinTuple *newtup, Size newsz,
bool samepage)
{
Page oldpage;
ItemId oldlp;
BrinTuple *oldtup;
Size oldsz;
Buffer newbuf;
bool extended;
Assert(newsz == MAXALIGN(newsz));
/* If the item is oversized, don't bother. */
if (newsz > BrinMaxItemSize)
{
ereport(ERROR,
(errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
errmsg("index row size %lu exceeds maximum %lu for index \"%s\"",
(unsigned long) newsz,
(unsigned long) BrinMaxItemSize,
RelationGetRelationName(idxrel))));
return false; /* keep compiler quiet */
}
/* make sure the revmap is long enough to contain the entry we need */
brinRevmapExtend(revmap, heapBlk);
if (!samepage)
{
/* need a page on which to put the item */
newbuf = brin_getinsertbuffer(idxrel, oldbuf, newsz, &extended);
if (!BufferIsValid(newbuf))
{
Assert(!extended);
return false;
}
/*
* Note: it's possible (though unlikely) that the returned newbuf is
* the same as oldbuf, if brin_getinsertbuffer determined that the old
* buffer does in fact have enough space.
*/
if (newbuf == oldbuf)
{
Assert(!extended);
newbuf = InvalidBuffer;
}
}
else
{
LockBuffer(oldbuf, BUFFER_LOCK_EXCLUSIVE);
newbuf = InvalidBuffer;
extended = false;
}
oldpage = BufferGetPage(oldbuf);
oldlp = PageGetItemId(oldpage, oldoff);
/*
* Check that the old tuple wasn't updated concurrently: it might have
* moved someplace else entirely ...
*/
if (!ItemIdIsNormal(oldlp))
{
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
/*
* If this happens, and the new buffer was obtained by extending the
* relation, then we need to ensure we don't leave it uninitialized or
* forget about it.
*/
if (BufferIsValid(newbuf))
{
if (extended)
brin_initialize_empty_new_buffer(idxrel, newbuf);
UnlockReleaseBuffer(newbuf);
if (extended)
FreeSpaceMapVacuum(idxrel);
}
return false;
}
oldsz = ItemIdGetLength(oldlp);
oldtup = (BrinTuple *) PageGetItem(oldpage, oldlp);
/*
* ... or it might have been updated in place to different contents.
*/
if (!brin_tuples_equal(oldtup, oldsz, origtup, origsz))
{
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
if (BufferIsValid(newbuf))
{
if (extended)
brin_initialize_empty_new_buffer(idxrel, newbuf);
UnlockReleaseBuffer(newbuf);
if (extended)
FreeSpaceMapVacuum(idxrel);
}
return false;
}
/*
* Great, the old tuple is intact. We can proceed with the update.
*
* If there's enough room in the old page for the new tuple, replace it.
*
* Note that there might now be enough space on the page even though the
* caller told us there isn't, if a concurrent update moved another tuple
* elsewhere or replaced a tuple with a smaller one.
*/
if (((BrinPageFlags(oldpage) & BRIN_EVACUATE_PAGE) == 0) &&
brin_can_do_samepage_update(oldbuf, origsz, newsz))
{
if (BufferIsValid(newbuf))
{
/* as above */
if (extended)
brin_initialize_empty_new_buffer(idxrel, newbuf);
UnlockReleaseBuffer(newbuf);
}
START_CRIT_SECTION();
if (!PageIndexTupleOverwrite(oldpage, oldoff, (Item) newtup, newsz))
elog(ERROR, "failed to replace BRIN tuple");
MarkBufferDirty(oldbuf);
/* XLOG stuff */
if (RelationNeedsWAL(idxrel))
{
xl_brin_samepage_update xlrec;
XLogRecPtr recptr;
uint8 info = XLOG_BRIN_SAMEPAGE_UPDATE;
xlrec.offnum = oldoff;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfBrinSamepageUpdate);
XLogRegisterBuffer(0, oldbuf, REGBUF_STANDARD);
XLogRegisterBufData(0, (char *) newtup, newsz);
recptr = XLogInsert(RM_BRIN_ID, info);
PageSetLSN(oldpage, recptr);
}
END_CRIT_SECTION();
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
if (extended)
FreeSpaceMapVacuum(idxrel);
return true;
}
else if (newbuf == InvalidBuffer)
{
/*
* Not enough space, but caller said that there was. Tell them to
* start over.
*/
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
return false;
}
else
{
/*
* Not enough free space on the oldpage. Put the new tuple on the new
* page, and update the revmap.
*/
Page newpage = BufferGetPage(newbuf);
Buffer revmapbuf;
ItemPointerData newtid;
OffsetNumber newoff;
BlockNumber newblk = InvalidBlockNumber;
Size freespace = 0;
revmapbuf = brinLockRevmapPageForUpdate(revmap, heapBlk);
START_CRIT_SECTION();
/*
* We need to initialize the page if it's newly obtained. Note we
* will WAL-log the initialization as part of the update, so we don't
* need to do that here.
*/
if (extended)
brin_page_init(BufferGetPage(newbuf), BRIN_PAGETYPE_REGULAR);
PageIndexTupleDeleteNoCompact(oldpage, oldoff);
newoff = PageAddItem(newpage, (Item) newtup, newsz,
InvalidOffsetNumber, false, false);
if (newoff == InvalidOffsetNumber)
elog(ERROR, "failed to add BRIN tuple to new page");
MarkBufferDirty(oldbuf);
MarkBufferDirty(newbuf);
/* needed to update FSM below */
if (extended)
{
newblk = BufferGetBlockNumber(newbuf);
freespace = br_page_get_freespace(newpage);
}
ItemPointerSet(&newtid, BufferGetBlockNumber(newbuf), newoff);
brinSetHeapBlockItemptr(revmapbuf, pagesPerRange, heapBlk, newtid);
MarkBufferDirty(revmapbuf);
/* XLOG stuff */
if (RelationNeedsWAL(idxrel))
{
xl_brin_update xlrec;
XLogRecPtr recptr;
uint8 info;
info = XLOG_BRIN_UPDATE | (extended ? XLOG_BRIN_INIT_PAGE : 0);
xlrec.insert.offnum = newoff;
xlrec.insert.heapBlk = heapBlk;
xlrec.insert.pagesPerRange = pagesPerRange;
xlrec.oldOffnum = oldoff;
XLogBeginInsert();
/* new page */
XLogRegisterData((char *) &xlrec, SizeOfBrinUpdate);
XLogRegisterBuffer(0, newbuf, REGBUF_STANDARD | (extended ? REGBUF_WILL_INIT : 0));
XLogRegisterBufData(0, (char *) newtup, newsz);
/* revmap page */
XLogRegisterBuffer(1, revmapbuf, 0);
/* old page */
XLogRegisterBuffer(2, oldbuf, REGBUF_STANDARD);
recptr = XLogInsert(RM_BRIN_ID, info);
PageSetLSN(oldpage, recptr);
PageSetLSN(newpage, recptr);
PageSetLSN(BufferGetPage(revmapbuf), recptr);
}
END_CRIT_SECTION();
LockBuffer(revmapbuf, BUFFER_LOCK_UNLOCK);
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
UnlockReleaseBuffer(newbuf);
if (extended)
{
Assert(BlockNumberIsValid(newblk));
RecordPageWithFreeSpace(idxrel, newblk, freespace);
FreeSpaceMapVacuum(idxrel);
}
return true;
}
}
/*
* _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 */
}
/*
* Apply changes represented by GenericXLogState to the actual buffers,
* and emit a generic xlog record.
*/
XLogRecPtr
GenericXLogFinish(GenericXLogState *state)
{
XLogRecPtr lsn;
int i;
if (state->isLogged)
{
/* Logged relation: make xlog record in critical section. */
XLogBeginInsert();
START_CRIT_SECTION();
for (i = 0; i < MAX_GENERIC_XLOG_PAGES; i++)
{
PageData *pageData = &state->pages[i];
Page page;
PageHeader pageHeader;
if (BufferIsInvalid(pageData->buffer))
continue;
page = BufferGetPage(pageData->buffer);
pageHeader = (PageHeader) pageData->image;
if (pageData->flags & GENERIC_XLOG_FULL_IMAGE)
{
/*
* A full-page image does not require us to supply any xlog
* data. Just apply the image, being careful to zero the
* "hole" between pd_lower and pd_upper in order to avoid
* divergence between actual page state and what replay would
* produce.
*/
memcpy(page, pageData->image, pageHeader->pd_lower);
memset(page + pageHeader->pd_lower, 0,
pageHeader->pd_upper - pageHeader->pd_lower);
memcpy(page + pageHeader->pd_upper,
pageData->image + pageHeader->pd_upper,
BLCKSZ - pageHeader->pd_upper);
XLogRegisterBuffer(i, pageData->buffer,
REGBUF_FORCE_IMAGE | REGBUF_STANDARD);
}
else
{
/*
* In normal mode, calculate delta and write it as xlog data
* associated with this page.
*/
computeDelta(pageData, page, (Page) pageData->image);
/* Apply the image, with zeroed "hole" as above */
memcpy(page, pageData->image, pageHeader->pd_lower);
memset(page + pageHeader->pd_lower, 0,
pageHeader->pd_upper - pageHeader->pd_lower);
memcpy(page + pageHeader->pd_upper,
pageData->image + pageHeader->pd_upper,
BLCKSZ - pageHeader->pd_upper);
XLogRegisterBuffer(i, pageData->buffer, REGBUF_STANDARD);
XLogRegisterBufData(i, pageData->delta, pageData->deltaLen);
}
}
/* Insert xlog record */
lsn = XLogInsert(RM_GENERIC_ID, 0);
/* Set LSN and mark buffers dirty */
for (i = 0; i < MAX_GENERIC_XLOG_PAGES; i++)
{
PageData *pageData = &state->pages[i];
if (BufferIsInvalid(pageData->buffer))
continue;
PageSetLSN(BufferGetPage(pageData->buffer), lsn);
MarkBufferDirty(pageData->buffer);
}
END_CRIT_SECTION();
}
else
{
/* Unlogged relation: skip xlog-related stuff */
START_CRIT_SECTION();
for (i = 0; i < MAX_GENERIC_XLOG_PAGES; i++)
{
PageData *pageData = &state->pages[i];
if (BufferIsInvalid(pageData->buffer))
continue;
memcpy(BufferGetPage(pageData->buffer),
pageData->image,
BLCKSZ);
/* We don't worry about zeroing the "hole" in this case */
MarkBufferDirty(pageData->buffer);
}
END_CRIT_SECTION();
/* We don't have a LSN to return, in this case */
lsn = InvalidXLogRecPtr;
}
for (i = 0; i < MAX_GENERIC_XLOG_PAGES; i++)
pfree(state->pages[i].image);
pfree(state);
return lsn;
}
/*
* _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;
}
/*
* 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);
}
/*
* Update tuple origtup (size origsz), located in offset oldoff of buffer
* oldbuf, to newtup (size newsz) as summary tuple for the page range starting
* at heapBlk. oldbuf must not be locked on entry, and is not locked at exit.
*
* If samepage is true, attempt to put the new tuple in the same page, but if
* there's no room, use some other one.
*
* If the update is successful, return true; the revmap is updated to point to
* the new tuple. If the update is not done for whatever reason, return false.
* Caller may retry the update if this happens.
*/
bool
brin_doupdate(Relation idxrel, BlockNumber pagesPerRange,
BrinRevmap *revmap, BlockNumber heapBlk,
Buffer oldbuf, OffsetNumber oldoff,
const BrinTuple *origtup, Size origsz,
const BrinTuple *newtup, Size newsz,
bool samepage)
{
Page oldpage;
ItemId oldlp;
BrinTuple *oldtup;
Size oldsz;
Buffer newbuf;
BrinSpecialSpace *special;
bool extended = false;
newsz = MAXALIGN(newsz);
/* make sure the revmap is long enough to contain the entry we need */
brinRevmapExtend(revmap, heapBlk);
if (!samepage)
{
/* need a page on which to put the item */
newbuf = brin_getinsertbuffer(idxrel, oldbuf, newsz, &extended);
/* XXX delay vacuuming FSM until locks are released? */
if (extended)
FreeSpaceMapVacuum(idxrel);
if (!BufferIsValid(newbuf))
return false;
/*
* Note: it's possible (though unlikely) that the returned newbuf is
* the same as oldbuf, if brin_getinsertbuffer determined that the old
* buffer does in fact have enough space.
*/
if (newbuf == oldbuf)
newbuf = InvalidBuffer;
}
else
{
LockBuffer(oldbuf, BUFFER_LOCK_EXCLUSIVE);
newbuf = InvalidBuffer;
}
oldpage = BufferGetPage(oldbuf);
oldlp = PageGetItemId(oldpage, oldoff);
/*
* Check that the old tuple wasn't updated concurrently: it might have
* moved someplace else entirely ...
*/
if (!ItemIdIsNormal(oldlp))
{
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
if (BufferIsValid(newbuf))
UnlockReleaseBuffer(newbuf);
return false;
}
oldsz = ItemIdGetLength(oldlp);
oldtup = (BrinTuple *) PageGetItem(oldpage, oldlp);
/*
* ... or it might have been updated in place to different contents.
*/
if (!brin_tuples_equal(oldtup, oldsz, origtup, origsz))
{
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
if (BufferIsValid(newbuf))
UnlockReleaseBuffer(newbuf);
return false;
}
special = (BrinSpecialSpace *) PageGetSpecialPointer(oldpage);
/*
* Great, the old tuple is intact. We can proceed with the update.
*
* If there's enough room in the old page for the new tuple, replace it.
*
* Note that there might now be enough space on the page even though the
* caller told us there isn't, if a concurrent update moved another tuple
* elsewhere or replaced a tuple with a smaller one.
*/
if (((special->flags & BRIN_EVACUATE_PAGE) == 0) &&
brin_can_do_samepage_update(oldbuf, origsz, newsz))
{
if (BufferIsValid(newbuf))
UnlockReleaseBuffer(newbuf);
START_CRIT_SECTION();
PageIndexDeleteNoCompact(oldpage, &oldoff, 1);
if (PageAddItem(oldpage, (Item) newtup, newsz, oldoff, true,
false) == InvalidOffsetNumber)
elog(ERROR, "failed to add BRIN tuple");
MarkBufferDirty(oldbuf);
/* XLOG stuff */
if (RelationNeedsWAL(idxrel))
{
xl_brin_samepage_update xlrec;
XLogRecPtr recptr;
uint8 info = XLOG_BRIN_SAMEPAGE_UPDATE;
xlrec.offnum = oldoff;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfBrinSamepageUpdate);
XLogRegisterBuffer(0, oldbuf, REGBUF_STANDARD);
XLogRegisterBufData(0, (char *) newtup, newsz);
recptr = XLogInsert(RM_BRIN_ID, info);
PageSetLSN(oldpage, recptr);
}
END_CRIT_SECTION();
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
return true;
}
else if (newbuf == InvalidBuffer)
{
/*
* Not enough space, but caller said that there was. Tell them to
* start over.
*/
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
return false;
}
else
{
/*
* Not enough free space on the oldpage. Put the new tuple on the new
* page, and update the revmap.
*/
Page newpage = BufferGetPage(newbuf);
Buffer revmapbuf;
ItemPointerData newtid;
OffsetNumber newoff;
revmapbuf = brinLockRevmapPageForUpdate(revmap, heapBlk);
START_CRIT_SECTION();
PageIndexDeleteNoCompact(oldpage, &oldoff, 1);
newoff = PageAddItem(newpage, (Item) newtup, newsz,
InvalidOffsetNumber, false, false);
if (newoff == InvalidOffsetNumber)
elog(ERROR, "failed to add BRIN tuple to new page");
MarkBufferDirty(oldbuf);
MarkBufferDirty(newbuf);
ItemPointerSet(&newtid, BufferGetBlockNumber(newbuf), newoff);
brinSetHeapBlockItemptr(revmapbuf, pagesPerRange, heapBlk, newtid);
MarkBufferDirty(revmapbuf);
/* XLOG stuff */
if (RelationNeedsWAL(idxrel))
{
xl_brin_update xlrec;
XLogRecPtr recptr;
uint8 info;
info = XLOG_BRIN_UPDATE | (extended ? XLOG_BRIN_INIT_PAGE : 0);
xlrec.insert.offnum = newoff;
xlrec.insert.heapBlk = heapBlk;
xlrec.insert.pagesPerRange = pagesPerRange;
xlrec.oldOffnum = oldoff;
XLogBeginInsert();
/* new page */
XLogRegisterData((char *) &xlrec, SizeOfBrinUpdate);
XLogRegisterBuffer(0, newbuf, REGBUF_STANDARD | (extended ? REGBUF_WILL_INIT : 0));
XLogRegisterBufData(0, (char *) newtup, newsz);
/* revmap page */
XLogRegisterBuffer(1, revmapbuf, REGBUF_STANDARD);
/* old page */
XLogRegisterBuffer(2, oldbuf, REGBUF_STANDARD);
recptr = XLogInsert(RM_BRIN_ID, info);
PageSetLSN(oldpage, recptr);
PageSetLSN(newpage, recptr);
PageSetLSN(BufferGetPage(revmapbuf), recptr);
}
END_CRIT_SECTION();
LockBuffer(revmapbuf, BUFFER_LOCK_UNLOCK);
LockBuffer(oldbuf, BUFFER_LOCK_UNLOCK);
UnlockReleaseBuffer(newbuf);
return true;
}
}
/*
* Insert an index tuple into the index relation. The revmap is updated to
* mark the range containing the given page as pointing to the inserted entry.
* A WAL record is written.
*
* The buffer, if valid, is first checked for free space to insert the new
* entry; if there isn't enough, a new buffer is obtained and pinned. No
* buffer lock must be held on entry, no buffer lock is held on exit.
*
* Return value is the offset number where the tuple was inserted.
*/
OffsetNumber
brin_doinsert(Relation idxrel, BlockNumber pagesPerRange,
BrinRevmap *revmap, Buffer *buffer, BlockNumber heapBlk,
BrinTuple *tup, Size itemsz)
{
Page page;
BlockNumber blk;
OffsetNumber off;
Buffer revmapbuf;
ItemPointerData tid;
bool extended = false;
itemsz = MAXALIGN(itemsz);
/* Make sure the revmap is long enough to contain the entry we need */
brinRevmapExtend(revmap, heapBlk);
/*
* Obtain a locked buffer to insert the new tuple. Note
* brin_getinsertbuffer ensures there's enough space in the returned
* buffer.
*/
if (BufferIsValid(*buffer))
{
/*
* It's possible that another backend (or ourselves!) extended the
* revmap over the page we held a pin on, so we cannot assume that
* it's still a regular page.
*/
LockBuffer(*buffer, BUFFER_LOCK_EXCLUSIVE);
if (br_page_get_freespace(BufferGetPage(*buffer)) < itemsz)
{
UnlockReleaseBuffer(*buffer);
*buffer = InvalidBuffer;
}
}
if (!BufferIsValid(*buffer))
{
*buffer = brin_getinsertbuffer(idxrel, InvalidBuffer, itemsz, &extended);
Assert(BufferIsValid(*buffer));
Assert(br_page_get_freespace(BufferGetPage(*buffer)) >= itemsz);
}
/* Now obtain lock on revmap buffer */
revmapbuf = brinLockRevmapPageForUpdate(revmap, heapBlk);
page = BufferGetPage(*buffer);
blk = BufferGetBlockNumber(*buffer);
START_CRIT_SECTION();
off = PageAddItem(page, (Item) tup, itemsz, InvalidOffsetNumber,
false, false);
if (off == InvalidOffsetNumber)
elog(ERROR, "could not insert new index tuple to page");
MarkBufferDirty(*buffer);
BRIN_elog((DEBUG2, "inserted tuple (%u,%u) for range starting at %u",
blk, off, heapBlk));
ItemPointerSet(&tid, blk, off);
brinSetHeapBlockItemptr(revmapbuf, pagesPerRange, heapBlk, tid);
MarkBufferDirty(revmapbuf);
/* XLOG stuff */
if (RelationNeedsWAL(idxrel))
{
xl_brin_insert xlrec;
XLogRecPtr recptr;
uint8 info;
info = XLOG_BRIN_INSERT | (extended ? XLOG_BRIN_INIT_PAGE : 0);
xlrec.heapBlk = heapBlk;
xlrec.pagesPerRange = pagesPerRange;
xlrec.offnum = off;
XLogBeginInsert();
XLogRegisterData((char *) &xlrec, SizeOfBrinInsert);
XLogRegisterBuffer(0, *buffer, REGBUF_STANDARD | (extended ? REGBUF_WILL_INIT : 0));
XLogRegisterBufData(0, (char *) tup, itemsz);
XLogRegisterBuffer(1, revmapbuf, 0);
recptr = XLogInsert(RM_BRIN_ID, info);
PageSetLSN(page, recptr);
PageSetLSN(BufferGetPage(revmapbuf), recptr);
}
END_CRIT_SECTION();
/* Tuple is firmly on buffer; we can release our locks */
LockBuffer(*buffer, BUFFER_LOCK_UNLOCK);
LockBuffer(revmapbuf, BUFFER_LOCK_UNLOCK);
if (extended)
FreeSpaceMapVacuum(idxrel);
return off;
}
