nbtree.c

PostgreSQL 8.1.4的源码 适用于Linux下的开源数据库系统

		if (BufferIsValid(so->btso_mrkbuf))
			ReleaseBuffer(so->btso_mrkbuf);
		so->btso_mrkbuf = InvalidBuffer;
		ItemPointerSetInvalid(iptr);
	}

	if (so->keyData != NULL)
		pfree(so->keyData);
	pfree(so);

	PG_RETURN_VOID();
}

/*
 *	btmarkpos() -- save current scan position
 */
Datum
btmarkpos(PG_FUNCTION_ARGS)
{
	IndexScanDesc scan = (IndexScanDesc) PG_GETARG_POINTER(0);
	ItemPointer iptr;
	BTScanOpaque so;

	so = (BTScanOpaque) scan->opaque;

	/* we aren't holding any read locks, but gotta drop the pin */
	if (ItemPointerIsValid(iptr = &(scan->currentMarkData)))
	{
		ReleaseBuffer(so->btso_mrkbuf);
		so->btso_mrkbuf = InvalidBuffer;
		ItemPointerSetInvalid(iptr);
	}

	/* bump pin on current buffer for assignment to mark buffer */
	if (ItemPointerIsValid(&(scan->currentItemData)))
	{
		IncrBufferRefCount(so->btso_curbuf);
		so->btso_mrkbuf = so->btso_curbuf;
		scan->currentMarkData = scan->currentItemData;
		so->mrkHeapIptr = so->curHeapIptr;
	}

	PG_RETURN_VOID();
}

/*
 *	btrestrpos() -- restore scan to last saved position
 */
Datum
btrestrpos(PG_FUNCTION_ARGS)
{
	IndexScanDesc scan = (IndexScanDesc) PG_GETARG_POINTER(0);
	ItemPointer iptr;
	BTScanOpaque so;

	so = (BTScanOpaque) scan->opaque;

	/* we aren't holding any read locks, but gotta drop the pin */
	if (ItemPointerIsValid(iptr = &(scan->currentItemData)))
	{
		ReleaseBuffer(so->btso_curbuf);
		so->btso_curbuf = InvalidBuffer;
		ItemPointerSetInvalid(iptr);
	}

	/* bump pin on marked buffer */
	if (ItemPointerIsValid(&(scan->currentMarkData)))
	{
		IncrBufferRefCount(so->btso_mrkbuf);
		so->btso_curbuf = so->btso_mrkbuf;
		scan->currentItemData = scan->currentMarkData;
		so->curHeapIptr = so->mrkHeapIptr;
	}

	PG_RETURN_VOID();
}

/*
 * 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.
 */
Datum
btbulkdelete(PG_FUNCTION_ARGS)
{
	Relation	rel = (Relation) PG_GETARG_POINTER(0);
	IndexBulkDeleteCallback callback = (IndexBulkDeleteCallback) PG_GETARG_POINTER(1);
	void	   *callback_state = (void *) PG_GETARG_POINTER(2);
	IndexBulkDeleteResult *result;
	double		tuples_removed;
	double		num_index_tuples;
	OffsetNumber deletable[MaxOffsetNumber];
	int			ndeletable;
	Buffer		buf;
	BlockNumber num_pages;

	tuples_removed = 0;
	num_index_tuples = 0;

	/*
	 * The outer loop iterates over index leaf pages, the inner over items on
	 * a leaf page.  We issue just one _bt_delitems() call per page, so as to
	 * minimize WAL traffic.
	 *
	 * Note that we exclusive-lock every leaf page containing data items, in
	 * sequence left to right.	It sounds attractive to only exclusive-lock
	 * those containing items we need to delete, but unfortunately that is not
	 * safe: we could then pass a stopped indexscan, which could in rare cases
	 * lead to deleting the item it needs to find when it resumes.	(See
	 * _bt_restscan --- this could only happen if an indexscan stops on a
	 * deletable item and then a page split moves that item into a page
	 * further to its right, which the indexscan will have no pin on.)	We can
	 * skip obtaining exclusive lock on empty pages though, since no indexscan
	 * could be stopped on those.
	 */
	buf = _bt_get_endpoint(rel, 0, false);
	if (BufferIsValid(buf))		/* check for empty index */
	{
		for (;;)
		{
			Page		page;
			BTPageOpaque opaque;
			OffsetNumber offnum,
						minoff,
						maxoff;
			BlockNumber nextpage;

			ndeletable = 0;
			page = BufferGetPage(buf);
			opaque = (BTPageOpaque) PageGetSpecialPointer(page);
			minoff = P_FIRSTDATAKEY(opaque);
			maxoff = PageGetMaxOffsetNumber(page);
			/* We probably cannot see deleted pages, but skip 'em if so */
			if (minoff <= maxoff && !P_ISDELETED(opaque))
			{
				/*
				 * Trade in the initial read lock for a super-exclusive write
				 * lock on this page.
				 */
				LockBuffer(buf, BUFFER_LOCK_UNLOCK);
				LockBufferForCleanup(buf);

				/*
				 * Recompute minoff/maxoff, both of which could have changed
				 * while we weren't holding the lock.
				 */
				minoff = P_FIRSTDATAKEY(opaque);
				maxoff = PageGetMaxOffsetNumber(page);

				/*
				 * Scan over all items to see which ones need deleted
				 * according to the callback function.
				 */
				for (offnum = minoff;
					 offnum <= maxoff;
					 offnum = OffsetNumberNext(offnum))
				{
					BTItem		btitem;
					ItemPointer htup;

					btitem = (BTItem) PageGetItem(page,
												PageGetItemId(page, offnum));
					htup = &(btitem->bti_itup.t_tid);
					if (callback(htup, callback_state))
					{
						deletable[ndeletable++] = offnum;
						tuples_removed += 1;
					}
					else
						num_index_tuples += 1;
				}
			}

			/*
			 * If we need to delete anything, do it and write the buffer; else
			 * just release the buffer.
			 */
			nextpage = opaque->btpo_next;
			if (ndeletable > 0)
			{
				_bt_delitems(rel, buf, deletable, ndeletable);
				_bt_wrtbuf(rel, buf);
			}
			else
				_bt_relbuf(rel, buf);

			/* call vacuum_delay_point while not holding any buffer lock */
			vacuum_delay_point();

			/* And advance to next page, if any */
			if (nextpage == P_NONE)
				break;
			buf = _bt_getbuf(rel, nextpage, BT_READ);
		}
	}

	/* return statistics */
	num_pages = RelationGetNumberOfBlocks(rel);

	result = (IndexBulkDeleteResult *) palloc0(sizeof(IndexBulkDeleteResult));
	result->num_pages = num_pages;
	result->num_index_tuples = num_index_tuples;
	result->tuples_removed = tuples_removed;

	PG_RETURN_POINTER(result);
}

/*
 * Post-VACUUM cleanup.
 *
 * Here, we scan looking for pages we can delete or return to the freelist.
 *
 * Result: a palloc'd struct containing statistical info for VACUUM displays.
 */
Datum
btvacuumcleanup(PG_FUNCTION_ARGS)
{
	Relation	rel = (Relation) PG_GETARG_POINTER(0);
	IndexVacuumCleanupInfo *info = (IndexVacuumCleanupInfo *) PG_GETARG_POINTER(1);
	IndexBulkDeleteResult *stats = (IndexBulkDeleteResult *) PG_GETARG_POINTER(2);
	BlockNumber num_pages;
	BlockNumber blkno;
	BlockNumber *freePages;
	int			nFreePages,
				maxFreePages;
	BlockNumber pages_deleted = 0;
	MemoryContext mycontext;
	MemoryContext oldcontext;
	bool		needLock;

	Assert(stats != NULL);

	/*
	 * First find out the number of pages in the index.  We must acquire the
	 * relation-extension lock while doing this to avoid a race condition: if
	 * someone else is extending the relation, there is a window where
	 * bufmgr/smgr have created a new all-zero page but it hasn't yet been
	 * write-locked by _bt_getbuf().  If we manage to scan such a page here,
	 * we'll improperly assume it can be recycled. Taking the lock
	 * synchronizes things enough to prevent a problem: either num_pages won't
	 * include the new page, or _bt_getbuf already has write lock on the
	 * buffer and it will be fully initialized before we can examine it.  (See
	 * also vacuumlazy.c, which has the same issue.)
	 *
	 * We can skip locking for new or temp relations, however, since no one
	 * else could be accessing them.
	 */
	needLock = !RELATION_IS_LOCAL(rel);

	if (needLock)
		LockRelationForExtension(rel, ExclusiveLock);

	num_pages = RelationGetNumberOfBlocks(rel);

	if (needLock)
		UnlockRelationForExtension(rel, ExclusiveLock);

	/* No point in remembering more than MaxFSMPages pages */
	maxFreePages = MaxFSMPages;
	if ((BlockNumber) maxFreePages > num_pages)
		maxFreePages = (int) num_pages;
	freePages = (BlockNumber *) palloc(maxFreePages * sizeof(BlockNumber));
	nFreePages = 0;

	/* Create a temporary memory context to run _bt_pagedel in */
	mycontext = AllocSetContextCreate(CurrentMemoryContext,
									  "_bt_pagedel",
									  ALLOCSET_DEFAULT_MINSIZE,
									  ALLOCSET_DEFAULT_INITSIZE,
									  ALLOCSET_DEFAULT_MAXSIZE);

	/*
	 * Scan through all pages of index, except metapage.  (Any pages added
	 * after we start the scan will not be examined; this should be fine,
	 * since they can't possibly be empty.)
	 */
	for (blkno = BTREE_METAPAGE + 1; blkno < num_pages; blkno++)
	{
		Buffer		buf;
		Page		page;
		BTPageOpaque opaque;

		vacuum_delay_point();

		buf = _bt_getbuf(rel, blkno, BT_READ);
		page = BufferGetPage(buf);
		opaque = (BTPageOpaque) PageGetSpecialPointer(page);
		if (_bt_page_recyclable(page))
		{
			/* Okay to recycle this page */
			if (nFreePages < maxFreePages)
				freePages[nFreePages++] = blkno;
			pages_deleted++;
		}
		else if (P_ISDELETED(opaque))
		{
			/* Already deleted, but can't recycle yet */
			pages_deleted++;
		}
		else if ((opaque->btpo_flags & BTP_HALF_DEAD) ||
				 P_FIRSTDATAKEY(opaque) > PageGetMaxOffsetNumber(page))
		{
			/* Empty, try to delete */
			int			ndel;

			/* Run pagedel in a temp context to avoid memory leakage */
			MemoryContextReset(mycontext);
			oldcontext = MemoryContextSwitchTo(mycontext);

			ndel = _bt_pagedel(rel, buf, info->vacuum_full);

			/* count only this page, else may double-count parent */
			if (ndel)
				pages_deleted++;

			/*
			 * During VACUUM FULL it's okay to recycle deleted pages
			 * immediately, since there can be no other transactions scanning
			 * the index.  Note that we will only recycle the current page and
			 * not any parent pages that _bt_pagedel might have recursed to;
			 * this seems reasonable in the name of simplicity.  (Trying to do
			 * otherwise would mean we'd have to sort the list of recyclable
			 * pages we're building.)
			 */
			if (ndel && info->vacuum_full)
			{
				if (nFreePages < maxFreePages)
					freePages[nFreePages++] = blkno;
			}

			MemoryContextSwitchTo(oldcontext);
			continue;			/* pagedel released buffer */
		}
		_bt_relbuf(rel, buf);
	}

	/*
	 * During VACUUM FULL, we truncate off any recyclable pages at the end of
	 * the index.  In a normal vacuum it'd be unsafe to do this except by
	 * acquiring exclusive lock on the index and then rechecking all the
	 * pages; doesn't seem worth it.
	 */
	if (info->vacuum_full && nFreePages > 0)
	{
		BlockNumber new_pages = num_pages;

		while (nFreePages > 0 && freePages[nFreePages - 1] == new_pages - 1)
		{
			new_pages--;
			pages_deleted--;
			nFreePages--;
		}
		if (new_pages != num_pages)
		{
			/*
			 * Okay to truncate.
			 */
			RelationTruncate(rel, new_pages);

			/* update statistics */
			stats->pages_removed = num_pages - new_pages;

			num_pages = new_pages;
		}
	}

	/*
	 * Update the shared Free Space Map with the info we now have about free
	 * pages in the index, discarding any old info the map may have. We do not
	 * need to sort the page numbers; they're in order already.
	 */
	RecordIndexFreeSpace(&rel->rd_node, nFreePages, freePages);

	pfree(freePages);

	MemoryContextDelete(mycontext);

	/* update statistics */
	stats->num_pages = num_pages;
	stats->pages_deleted = pages_deleted;
	stats->pages_free = nFreePages;

	PG_RETURN_POINTER(stats);
}

/*
 * Restore scan position when btgettuple is called to continue a scan.
 *
 * This is nontrivial because concurrent insertions might have moved the
 * index tuple we stopped on.  We assume the tuple can only have moved to
 * the right from our stop point, because we kept a pin on the buffer,
 * and so no deletion can have occurred on that page.
 *
 * On entry, we have a pin but no read lock on the buffer that contained
 * the index tuple we stopped the scan on.	On exit, we have pin and read
 * lock on the buffer that now contains that index tuple, and the scandesc's
 * current position is updated to point at it.
 */
static void
_bt_restscan(IndexScanDesc scan)
{
	Relation	rel = scan->indexRelation;
	BTScanOpaque so = (BTScanOpaque) scan->opaque;
	Buffer		buf = so->btso_curbuf;
	Page		page;
	ItemPointer current = &(scan->currentItemData);
	OffsetNumber offnum = ItemPointerGetOffsetNumber(current),
				maxoff;
	BTPageOpaque opaque;
	Buffer		nextbuf;
	ItemPointer target = &(so->curHeapIptr);
	BTItem		item;
	BlockNumber blkno;

	/*
	 * Reacquire read lock on the buffer.  (We should still have a
	 * reference-count pin on it, so need not get that.)
	 */
	LockBuffer(buf, BT_READ);

	page = BufferGetPage(buf);
	maxoff = PageGetMaxOffsetNumber(page);
	opaque = (BTPageOpaque) PageGetSpecialPointer(page);

	/*
	 * We use this as flag when first index tuple on page is deleted but we do
	 * not move left (this would slowdown vacuum) - so we set
	 * current->ip_posid before first index tuple on the current page
	 * (_bt_step will move it right)...  XXX still needed?
	 */
	if (!ItemPointerIsValid(target))
	{
		ItemPointerSetOffsetNumber(current,
								   OffsetNumberPrev(P_FIRSTDATAKEY(opaque)));
		return;
	}

	/*
	 * The item we were on may have moved right due to insertions. Find it
	 * again.  We use the heap TID to identify the item uniquely.
	 */
	for (;;)
	{
		/* Check for item on this page */
		for (;
			 offnum <= maxoff;
			 offnum = OffsetNumberNext(offnum))
		{
			item = (BTItem) PageGetItem(page, PageGetItemId(page, offnum));
			if (BTTidSame(item->bti_itup.t_tid, *target))
			{
				/* Found it */
				current->ip_posid = offnum;
				return;
			}
		}

		/*
		 * The item we're looking for moved right at least one page, so move
		 * right.  We are careful here to pin and read-lock the next non-dead
		 * page before releasing the current one.  This ensures that a
		 * concurrent btbulkdelete scan cannot pass our position --- if it
		 * did, it might be able to reach and delete our target item before we
		 * can find it again.
		 */
		if (P_RIGHTMOST(opaque))
			elog(ERROR, "failed to re-find previous key in \"%s\"",
				 RelationGetRelationName(rel));
		/* Advance to next non-dead page --- there must be one */
		nextbuf = InvalidBuffer;
		for (;;)
		{
			blkno = opaque->btpo_next;
			nextbuf = _bt_relandgetbuf(rel, nextbuf, blkno, BT_READ);
			page = BufferGetPage(nextbuf);
			opaque = (BTPageOpaque) PageGetSpecialPointer(page);
			if (!P_IGNORE(opaque))
				break;
			if (P_RIGHTMOST(opaque))
				elog(ERROR, "fell off the end of \"%s\"",
					 RelationGetRelationName(rel));
		}
		_bt_relbuf(rel, buf);
		so->btso_curbuf = buf = nextbuf;
		maxoff = PageGetMaxOffsetNumber(page);
		offnum = P_FIRSTDATAKEY(opaque);
		ItemPointerSet(current, blkno, offnum);
	}
}