~drizzle-trunk/drizzle/development

/*****************************************************************************

Copyright (c) 1997, 2009, Innobase Oy. All Rights Reserved.

This program is free software; you can redistribute it and/or modify it under

the terms of the GNU General Public License as published by the Free Software

Foundation; version 2 of the License.

This program is distributed in the hope that it will be useful, but WITHOUT

ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS

FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with

this program; if not, write to the Free Software Foundation, Inc., 51 Franklin

St, Fifth Floor, Boston, MA 02110-1301 USA

*****************************************************************************/

/**************************************************//**

@file ibuf/ibuf0ibuf.c

Insert buffer

Created 7/19/1997 Heikki Tuuri

*******************************************************/

#include "ibuf0ibuf.h"

/** Number of bits describing a single page */

#define IBUF_BITS_PER_PAGE	4

#if IBUF_BITS_PER_PAGE % 2

# error "IBUF_BITS_PER_PAGE must be an even number!"

#endif

/** The start address for an insert buffer bitmap page bitmap */

#define IBUF_BITMAP		PAGE_DATA

#ifdef UNIV_NONINL

#include "ibuf0ibuf.ic"

#endif

#ifndef UNIV_HOTBACKUP

#include "buf0buf.h"

#include "buf0rea.h"

#include "fsp0fsp.h"

#include "trx0sys.h"

#include "fil0fil.h"

#include "thr0loc.h"

#include "rem0rec.h"

#include "btr0cur.h"

#include "btr0pcur.h"

#include "btr0btr.h"

#include "sync0sync.h"

#include "dict0boot.h"

#include "fut0lst.h"

#include "lock0lock.h"

#include "log0recv.h"

#include "que0que.h"

/*	STRUCTURE OF AN INSERT BUFFER RECORD

In versions < 4.1.x:

1. The first field is the page number.

2. The second field is an array which stores type info for each subsequent

   field. We store the information which affects the ordering of records, and

   also the physical storage size of an SQL NULL value. E.g., for CHAR(10) it

   is 10 bytes.

3. Next we have the fields of the actual index record.

In versions >= 4.1.x:

Note that contary to what we planned in the 1990's, there will only be one

insert buffer tree, and that is in the system tablespace of InnoDB.

1. The first field is the space id.

2. The second field is a one-byte marker (0) which differentiates records from

   the < 4.1.x storage format.

3. The third field is the page number.

4. The fourth field contains the type info, where we have also added 2 bytes to

   store the charset. In the compressed table format of 5.0.x we must add more

   information here so that we can build a dummy 'index' struct which 5.0.x

   can use in the binary search on the index page in the ibuf merge phase.

5. The rest of the fields contain the fields of the actual index record.

In versions >= 5.0.3:

The first byte of the fourth field is an additional marker (0) if the record

is in the compact format.  The presence of this marker can be detected by

looking at the length of the field modulo DATA_NEW_ORDER_NULL_TYPE_BUF_SIZE.

The high-order bit of the character set field in the type info is the

"nullable" flag for the field. */

/*	PREVENTING DEADLOCKS IN THE INSERT BUFFER SYSTEM

If an OS thread performs any operation that brings in disk pages from

non-system tablespaces into the buffer pool, or creates such a page there,

then the operation may have as a side effect an insert buffer index tree

compression. Thus, the tree latch of the insert buffer tree may be acquired

in the x-mode, and also the file space latch of the system tablespace may

be acquired in the x-mode.

Also, an insert to an index in a non-system tablespace can have the same

effect. How do we know this cannot lead to a deadlock of OS threads? There

is a problem with the i\o-handler threads: they break the latching order

because they own x-latches to pages which are on a lower level than the

insert buffer tree latch, its page latches, and the tablespace latch an

insert buffer operation can reserve.

The solution is the following: Let all the tree and page latches connected

with the insert buffer be later in the latching order than the fsp latch and

fsp page latches.

Insert buffer pages must be such that the insert buffer is never invoked

when these pages are accessed as this would result in a recursion violating

the latching order. We let a special i/o-handler thread take care of i/o to

the insert buffer pages and the ibuf bitmap pages, as well as the fsp bitmap

pages and the first inode page, which contains the inode of the ibuf tree: let

us call all these ibuf pages. To prevent deadlocks, we do not let a read-ahead

access both non-ibuf and ibuf pages.

Then an i/o-handler for the insert buffer never needs to access recursively the

insert buffer tree and thus obeys the latching order. On the other hand, other

i/o-handlers for other tablespaces may require access to the insert buffer,

but because all kinds of latches they need to access there are later in the

latching order, no violation of the latching order occurs in this case,

either.

A problem is how to grow and contract an insert buffer tree. As it is later

in the latching order than the fsp management, we have to reserve the fsp

latch first, before adding or removing pages from the insert buffer tree.

We let the insert buffer tree have its own file space management: a free

list of pages linked to the tree root. To prevent recursive using of the

insert buffer when adding pages to the tree, we must first load these pages

to memory, obtaining a latch on them, and only after that add them to the

free list of the insert buffer tree. More difficult is removing of pages

from the free list. If there is an excess of pages in the free list of the

ibuf tree, they might be needed if some thread reserves the fsp latch,

intending to allocate more file space. So we do the following: if a thread

reserves the fsp latch, we check the writer count field of the latch. If

this field has value 1, it means that the thread did not own the latch

before entering the fsp system, and the mtr of the thread contains no

modifications to the fsp pages. Now we are free to reserve the ibuf latch,

and check if there is an excess of pages in the free list. We can then, in a

separate mini-transaction, take them out of the free list and free them to

the fsp system.

To avoid deadlocks in the ibuf system, we divide file pages into three levels:

(1) non-ibuf pages,

(2) ibuf tree pages and the pages in the ibuf tree free list, and

(3) ibuf bitmap pages.

No OS thread is allowed to access higher level pages if it has latches to

lower level pages; even if the thread owns a B-tree latch it must not access

the B-tree non-leaf pages if it has latches on lower level pages. Read-ahead

is only allowed for level 1 and 2 pages. Dedicated i/o-handler threads handle

exclusively level 1 i/o. A dedicated i/o handler thread handles exclusively

level 2 i/o. However, if an OS thread does the i/o handling for itself, i.e.,

it uses synchronous aio, it can access any pages, as long as it obeys the

access order rules. */

/** Buffer pool size per the maximum insert buffer size */

#define IBUF_POOL_SIZE_PER_MAX_SIZE	2

/** Table name for the insert buffer. */

#define IBUF_TABLE_NAME		"SYS_IBUF_TABLE"

/** Operations that can currently be buffered. */

UNIV_INTERN ibuf_use_t	ibuf_use		= IBUF_USE_INSERT;

/** The insert buffer control structure */

UNIV_INTERN ibuf_t*	ibuf			= NULL;

/** Counter for ibuf_should_try() */

UNIV_INTERN ulint	ibuf_flush_count	= 0;

#ifdef UNIV_IBUF_COUNT_DEBUG

/** Number of tablespaces in the ibuf_counts array */

#define IBUF_COUNT_N_SPACES	4

/** Number of pages within each tablespace in the ibuf_counts array */

#define IBUF_COUNT_N_PAGES	130000

/** Buffered entry counts for file pages, used in debugging */

static ulint	ibuf_counts[IBUF_COUNT_N_SPACES][IBUF_COUNT_N_PAGES];

/******************************************************************//**

Checks that the indexes to ibuf_counts[][] are within limits. */

UNIV_INLINE

void

ibuf_count_check(

/*=============*/

	ulint	space_id,	/*!< in: space identifier */

	ulint	page_no)	/*!< in: page number */

{

	if (space_id < IBUF_COUNT_N_SPACES && page_no < IBUF_COUNT_N_PAGES) {

		return;

	}

	fprintf(stderr,

		"InnoDB: UNIV_IBUF_COUNT_DEBUG limits space_id and page_no\n"

		"InnoDB: and breaks crash recovery.\n"

		"InnoDB: space_id=%lu, should be 0<=space_id<%lu\n"

		"InnoDB: page_no=%lu, should be 0<=page_no<%lu\n",

		(ulint) space_id, (ulint) IBUF_COUNT_N_SPACES,

		(ulint) page_no, (ulint) IBUF_COUNT_N_PAGES);

	ut_error;

}

#endif

/** @name Offsets to the per-page bits in the insert buffer bitmap */

/* @{ */

#define	IBUF_BITMAP_FREE	0	/*!< Bits indicating the

					amount of free space */

#define IBUF_BITMAP_BUFFERED	2	/*!< TRUE if there are buffered

					changes for the page */

#define IBUF_BITMAP_IBUF	3	/*!< TRUE if page is a part of

					the ibuf tree, excluding the

					root page, or is in the free

					list of the ibuf */

/* @} */

/** The mutex used to block pessimistic inserts to ibuf trees */

static mutex_t	ibuf_pessimistic_insert_mutex;

/** The mutex protecting the insert buffer structs */

static mutex_t	ibuf_mutex;

/** The mutex protecting the insert buffer bitmaps */

static mutex_t	ibuf_bitmap_mutex;

/** The area in pages from which contract looks for page numbers for merge */

#define	IBUF_MERGE_AREA			8

/** Inside the merge area, pages which have at most 1 per this number less

buffered entries compared to maximum volume that can buffered for a single

page are merged along with the page whose buffer became full */

#define IBUF_MERGE_THRESHOLD		4

/** In ibuf_contract at most this number of pages is read to memory in one

batch, in order to merge the entries for them in the insert buffer */

#define	IBUF_MAX_N_PAGES_MERGED		IBUF_MERGE_AREA

/** If the combined size of the ibuf trees exceeds ibuf->max_size by this

many pages, we start to contract it in connection to inserts there, using

non-synchronous contract */

#define IBUF_CONTRACT_ON_INSERT_NON_SYNC	0

/** If the combined size of the ibuf trees exceeds ibuf->max_size by this

many pages, we start to contract it in connection to inserts there, using

synchronous contract */

#define IBUF_CONTRACT_ON_INSERT_SYNC		5

/** If the combined size of the ibuf trees exceeds ibuf->max_size by

this many pages, we start to contract it synchronous contract, but do

not insert */

#define IBUF_CONTRACT_DO_NOT_INSERT		10

/* TODO: how to cope with drop table if there are records in the insert

buffer for the indexes of the table? Is there actually any problem,

because ibuf merge is done to a page when it is read in, and it is

still physically like the index page even if the index would have been

dropped! So, there seems to be no problem. */

/******************************************************************//**

Sets the flag in the current OS thread local storage denoting that it is

inside an insert buffer routine. */

UNIV_INLINE

void

ibuf_enter(void)

/*============*/

{

	ibool*	ptr;

	ptr = thr_local_get_in_ibuf_field();

	ut_ad(*ptr == FALSE);

	*ptr = TRUE;

}

/******************************************************************//**

Sets the flag in the current OS thread local storage denoting that it is

exiting an insert buffer routine. */

UNIV_INLINE

void

ibuf_exit(void)

/*===========*/

{

	ibool*	ptr;

	ptr = thr_local_get_in_ibuf_field();

	ut_ad(*ptr == TRUE);

	*ptr = FALSE;

}

/******************************************************************//**

Returns TRUE if the current OS thread is performing an insert buffer

routine.

For instance, a read-ahead of non-ibuf pages is forbidden by threads

that are executing an insert buffer routine.

@return TRUE if inside an insert buffer routine */

UNIV_INTERN

ibool

ibuf_inside(void)

/*=============*/

{

	return(*thr_local_get_in_ibuf_field());

}

/******************************************************************//**

Gets the ibuf header page and x-latches it.

@return	insert buffer header page */

static

page_t*

ibuf_header_page_get(

/*=================*/

	mtr_t*	mtr)	/*!< in: mtr */

{

	buf_block_t*	block;

	ut_ad(!ibuf_inside());

	block = buf_page_get(

		IBUF_SPACE_ID, 0, FSP_IBUF_HEADER_PAGE_NO, RW_X_LATCH, mtr);

	buf_block_dbg_add_level(block, SYNC_IBUF_HEADER);

	return(buf_block_get_frame(block));

}

/******************************************************************//**

Gets the root page and x-latches it.

@return	insert buffer tree root page */

static

page_t*

ibuf_tree_root_get(

/*===============*/

	mtr_t*		mtr)	/*!< in: mtr */

{

	buf_block_t*	block;

	ut_ad(ibuf_inside());

	mtr_x_lock(dict_index_get_lock(ibuf->index), mtr);

	block = buf_page_get(

		IBUF_SPACE_ID, 0, FSP_IBUF_TREE_ROOT_PAGE_NO, RW_X_LATCH, mtr);

	buf_block_dbg_add_level(block, SYNC_TREE_NODE);

	return(buf_block_get_frame(block));

}

#ifdef UNIV_IBUF_COUNT_DEBUG

/******************************************************************//**

Gets the ibuf count for a given page.

@return number of entries in the insert buffer currently buffered for

this page */

UNIV_INTERN

ulint

ibuf_count_get(

/*===========*/

	ulint	space,	/*!< in: space id */

	ulint	page_no)/*!< in: page number */

{

	ibuf_count_check(space, page_no);

	return(ibuf_counts[space][page_no]);

}

/******************************************************************//**

Sets the ibuf count for a given page. */

static

void

ibuf_count_set(

/*===========*/

	ulint	space,	/*!< in: space id */

	ulint	page_no,/*!< in: page number */

	ulint	val)	/*!< in: value to set */

{

	ibuf_count_check(space, page_no);

	ut_a(val < UNIV_PAGE_SIZE);

	ibuf_counts[space][page_no] = val;

}

#endif

/******************************************************************//**

Closes insert buffer and frees the data structures. */

UNIV_INTERN

void

ibuf_close(void)

/*============*/

{

	mutex_free(&ibuf_pessimistic_insert_mutex);

	memset(&ibuf_pessimistic_insert_mutex,

	       0x0, sizeof(ibuf_pessimistic_insert_mutex));

	mutex_free(&ibuf_mutex);

	memset(&ibuf_mutex, 0x0, sizeof(ibuf_mutex));

	mutex_free(&ibuf_bitmap_mutex);

	memset(&ibuf_bitmap_mutex, 0x0, sizeof(ibuf_mutex));

	mem_free(ibuf);

	ibuf = NULL;

}

/******************************************************************//**

Updates the size information of the ibuf, assuming the segment size has not

changed. */

static

void

ibuf_size_update(

/*=============*/

	const page_t*	root,	/*!< in: ibuf tree root */

	mtr_t*		mtr)	/*!< in: mtr */

{

	ut_ad(mutex_own(&ibuf_mutex));

	ibuf->free_list_len = flst_get_len(root + PAGE_HEADER

					   + PAGE_BTR_IBUF_FREE_LIST, mtr);

	ibuf->height = 1 + btr_page_get_level(root, mtr);

	/* the '1 +' is the ibuf header page */

	ibuf->size = ibuf->seg_size - (1 + ibuf->free_list_len);

	ibuf->empty = page_get_n_recs(root) == 0;

}

/******************************************************************//**

Creates the insert buffer data structure at a database startup and initializes

the data structures for the insert buffer. */

UNIV_INTERN

void

ibuf_init_at_db_start(void)

/*=======================*/

{

	page_t*		root;

	mtr_t		mtr;

	dict_table_t*	table;

	mem_heap_t*	heap;

	dict_index_t*	index;

	ulint		n_used;

	page_t*		header_page;

	ulint		error;

	ibuf = mem_alloc(sizeof(ibuf_t));

	memset(ibuf, 0, sizeof(*ibuf));

	/* Note that also a pessimistic delete can sometimes make a B-tree

	grow in size, as the references on the upper levels of the tree can

	change */

	ibuf->max_size = buf_pool_get_curr_size() / UNIV_PAGE_SIZE

		/ IBUF_POOL_SIZE_PER_MAX_SIZE;

	mutex_create(&ibuf_pessimistic_insert_mutex,

		     SYNC_IBUF_PESS_INSERT_MUTEX);

	mutex_create(&ibuf_mutex, SYNC_IBUF_MUTEX);

	mutex_create(&ibuf_bitmap_mutex, SYNC_IBUF_BITMAP_MUTEX);

	mtr_start(&mtr);

	mutex_enter(&ibuf_mutex);

	mtr_x_lock(fil_space_get_latch(IBUF_SPACE_ID, NULL), &mtr);

	header_page = ibuf_header_page_get(&mtr);

	fseg_n_reserved_pages(header_page + IBUF_HEADER + IBUF_TREE_SEG_HEADER,

			      &n_used, &mtr);

	ibuf_enter();

	ut_ad(n_used >= 2);

	ibuf->seg_size = n_used;

	{

		buf_block_t*	block;

		block = buf_page_get(

			IBUF_SPACE_ID, 0, FSP_IBUF_TREE_ROOT_PAGE_NO,

			RW_X_LATCH, &mtr);

		buf_block_dbg_add_level(block, SYNC_TREE_NODE);

		root = buf_block_get_frame(block);

	}

	ibuf_size_update(root, &mtr);

	mutex_exit(&ibuf_mutex);

	mtr_commit(&mtr);

	ibuf_exit();

	heap = mem_heap_create(450);

	/* Use old-style record format for the insert buffer. */

	table = dict_mem_table_create(IBUF_TABLE_NAME, IBUF_SPACE_ID, 1, 0);

	dict_mem_table_add_col(table, heap, "DUMMY_COLUMN", DATA_BINARY, 0, 0);

	table->id = ut_dulint_add(DICT_IBUF_ID_MIN, IBUF_SPACE_ID);

	dict_table_add_to_cache(table, heap);

	mem_heap_free(heap);

	index = dict_mem_index_create(

		IBUF_TABLE_NAME, "CLUST_IND",

		IBUF_SPACE_ID, DICT_CLUSTERED | DICT_UNIVERSAL | DICT_IBUF, 1);

	dict_mem_index_add_field(index, "DUMMY_COLUMN", 0);

	index->id = ut_dulint_add(DICT_IBUF_ID_MIN, IBUF_SPACE_ID);

	error = dict_index_add_to_cache(table, index,

					FSP_IBUF_TREE_ROOT_PAGE_NO, FALSE);

	ut_a(error == DB_SUCCESS);

	ibuf->index = dict_table_get_first_index(table);

}

#endif /* !UNIV_HOTBACKUP */

/*********************************************************************//**

Initializes an ibuf bitmap page. */

UNIV_INTERN

void

ibuf_bitmap_page_init(

/*==================*/

	buf_block_t*	block,	/*!< in: bitmap page */

	mtr_t*		mtr)	/*!< in: mtr */

{

	page_t*	page;

	ulint	byte_offset;

	ulint	zip_size = buf_block_get_zip_size(block);

	ut_a(ut_is_2pow(zip_size));

	page = buf_block_get_frame(block);

	fil_page_set_type(page, FIL_PAGE_IBUF_BITMAP);

	/* Write all zeros to the bitmap */

	if (!zip_size) {

		byte_offset = UT_BITS_IN_BYTES(UNIV_PAGE_SIZE

					       * IBUF_BITS_PER_PAGE);

	} else {

		byte_offset = UT_BITS_IN_BYTES(zip_size * IBUF_BITS_PER_PAGE);

	}

	memset(page + IBUF_BITMAP, 0, byte_offset);

	/* The remaining area (up to the page trailer) is uninitialized. */

#ifndef UNIV_HOTBACKUP

	mlog_write_initial_log_record(page, MLOG_IBUF_BITMAP_INIT, mtr);

#endif /* !UNIV_HOTBACKUP */

}

/*********************************************************************//**

Parses a redo log record of an ibuf bitmap page init.

@return	end of log record or NULL */

UNIV_INTERN

byte*

ibuf_parse_bitmap_init(

/*===================*/

	byte*		ptr,	/*!< in: buffer */

	byte*		end_ptr __attribute__((unused)), /*!< in: buffer end */

	buf_block_t*	block,	/*!< in: block or NULL */

	mtr_t*		mtr)	/*!< in: mtr or NULL */

{

	ut_ad(ptr && end_ptr);

	if (block) {

		ibuf_bitmap_page_init(block, mtr);

	}

	return(ptr);

}

#ifndef UNIV_HOTBACKUP

/********************************************************************//**

Gets the desired bits for a given page from a bitmap page.

@return	value of bits */

UNIV_INLINE

ulint

ibuf_bitmap_page_get_bits(

/*======================*/

	const page_t*	page,	/*!< in: bitmap page */

	ulint		page_no,/*!< in: page whose bits to get */

	ulint		zip_size,/*!< in: compressed page size in bytes;

				0 for uncompressed pages */

	ulint		bit,	/*!< in: IBUF_BITMAP_FREE,

				IBUF_BITMAP_BUFFERED, ... */

	mtr_t*		mtr __attribute__((unused)))

				/*!< in: mtr containing an

				x-latch to the bitmap page */

{

	ulint	byte_offset;

	ulint	bit_offset;

	ulint	map_byte;

	ulint	value;

	ut_ad(bit < IBUF_BITS_PER_PAGE);

#if IBUF_BITS_PER_PAGE % 2

# error "IBUF_BITS_PER_PAGE % 2 != 0"

#endif

	ut_ad(ut_is_2pow(zip_size));

	ut_ad(mtr_memo_contains_page(mtr, page, MTR_MEMO_PAGE_X_FIX));

	if (!zip_size) {

		bit_offset = (page_no % UNIV_PAGE_SIZE) * IBUF_BITS_PER_PAGE

			+ bit;

	} else {

		bit_offset = (page_no & (zip_size - 1)) * IBUF_BITS_PER_PAGE

			+ bit;

	}

	byte_offset = bit_offset / 8;

	bit_offset = bit_offset % 8;

	ut_ad(byte_offset + IBUF_BITMAP < UNIV_PAGE_SIZE);

	map_byte = mach_read_from_1(page + IBUF_BITMAP + byte_offset);

	value = ut_bit_get_nth(map_byte, bit_offset);

	if (bit == IBUF_BITMAP_FREE) {

		ut_ad(bit_offset + 1 < 8);

		value = value * 2 + ut_bit_get_nth(map_byte, bit_offset + 1);

	}

	return(value);

}

/********************************************************************//**

Sets the desired bit for a given page in a bitmap page. */

static

void

ibuf_bitmap_page_set_bits(

/*======================*/

	page_t*	page,	/*!< in: bitmap page */

	ulint	page_no,/*!< in: page whose bits to set */

	ulint	zip_size,/*!< in: compressed page size in bytes;

			0 for uncompressed pages */

	ulint	bit,	/*!< in: IBUF_BITMAP_FREE, IBUF_BITMAP_BUFFERED, ... */

	ulint	val,	/*!< in: value to set */

	mtr_t*	mtr)	/*!< in: mtr containing an x-latch to the bitmap page */

{

	ulint	byte_offset;

	ulint	bit_offset;

	ulint	map_byte;

	ut_ad(bit < IBUF_BITS_PER_PAGE);

#if IBUF_BITS_PER_PAGE % 2

# error "IBUF_BITS_PER_PAGE % 2 != 0"

#endif

	ut_ad(ut_is_2pow(zip_size));

	ut_ad(mtr_memo_contains_page(mtr, page, MTR_MEMO_PAGE_X_FIX));

#ifdef UNIV_IBUF_COUNT_DEBUG

	ut_a((bit != IBUF_BITMAP_BUFFERED) || (val != FALSE)

	     || (0 == ibuf_count_get(page_get_space_id(page),

				     page_no)));

#endif

	if (!zip_size) {

		bit_offset = (page_no % UNIV_PAGE_SIZE) * IBUF_BITS_PER_PAGE

			+ bit;

	} else {

		bit_offset = (page_no & (zip_size - 1)) * IBUF_BITS_PER_PAGE

			+ bit;

	}

	byte_offset = bit_offset / 8;

	bit_offset = bit_offset % 8;

	ut_ad(byte_offset + IBUF_BITMAP < UNIV_PAGE_SIZE);

	map_byte = mach_read_from_1(page + IBUF_BITMAP + byte_offset);

	if (bit == IBUF_BITMAP_FREE) {

		ut_ad(bit_offset + 1 < 8);

		ut_ad(val <= 3);

		map_byte = ut_bit_set_nth(map_byte, bit_offset, val / 2);

		map_byte = ut_bit_set_nth(map_byte, bit_offset + 1, val % 2);

	} else {

		ut_ad(val <= 1);

		map_byte = ut_bit_set_nth(map_byte, bit_offset, val);

	}

	mlog_write_ulint(page + IBUF_BITMAP + byte_offset, map_byte,

			 MLOG_1BYTE, mtr);

}

/********************************************************************//**

Calculates the bitmap page number for a given page number.

@return	the bitmap page number where the file page is mapped */

UNIV_INLINE

ulint

ibuf_bitmap_page_no_calc(

/*=====================*/

	ulint	zip_size,	/*!< in: compressed page size in bytes;

				0 for uncompressed pages */

	ulint	page_no)	/*!< in: tablespace page number */

{

	ut_ad(ut_is_2pow(zip_size));

	if (!zip_size) {

		return(FSP_IBUF_BITMAP_OFFSET

		       + (page_no & ~(UNIV_PAGE_SIZE - 1)));

	} else {

		return(FSP_IBUF_BITMAP_OFFSET

		       + (page_no & ~(zip_size - 1)));

	}

}

/********************************************************************//**

Gets the ibuf bitmap page where the bits describing a given file page are

stored.

@return bitmap page where the file page is mapped, that is, the bitmap

page containing the descriptor bits for the file page; the bitmap page

is x-latched */

static

page_t*

ibuf_bitmap_get_map_page(

/*=====================*/

	ulint	space,	/*!< in: space id of the file page */

	ulint	page_no,/*!< in: page number of the file page */

	ulint	zip_size,/*!< in: compressed page size in bytes;

			0 for uncompressed pages */

	mtr_t*	mtr)	/*!< in: mtr */

{

	buf_block_t*	block;

	block = buf_page_get(space, zip_size,

			     ibuf_bitmap_page_no_calc(zip_size, page_no),

			     RW_X_LATCH, mtr);

	buf_block_dbg_add_level(block, SYNC_IBUF_BITMAP);

	return(buf_block_get_frame(block));

}

/************************************************************************//**

Sets the free bits of the page in the ibuf bitmap. This is done in a separate

mini-transaction, hence this operation does not restrict further work to only

ibuf bitmap operations, which would result if the latch to the bitmap page

were kept. */

UNIV_INLINE

void

ibuf_set_free_bits_low(

/*===================*/

	ulint			zip_size,/*!< in: compressed page size in bytes;

					0 for uncompressed pages */

	const buf_block_t*	block,	/*!< in: index page; free bits are set if

					the index is non-clustered and page

					level is 0 */

	ulint			val,	/*!< in: value to set: < 4 */

	mtr_t*			mtr)	/*!< in/out: mtr */

{

	page_t*	bitmap_page;

	ulint	space;

	ulint	page_no;

	if (!page_is_leaf(buf_block_get_frame(block))) {

		return;

	}

	space = buf_block_get_space(block);

	page_no = buf_block_get_page_no(block);

	bitmap_page = ibuf_bitmap_get_map_page(space, page_no, zip_size, mtr);

#ifdef UNIV_IBUF_DEBUG

# if 0

	fprintf(stderr,

		"Setting space %lu page %lu free bits to %lu should be %lu\n",

		space, page_no, val,

		ibuf_index_page_calc_free(zip_size, block));

# endif

	ut_a(val <= ibuf_index_page_calc_free(zip_size, block));

#endif /* UNIV_IBUF_DEBUG */

	ibuf_bitmap_page_set_bits(bitmap_page, page_no, zip_size,

				  IBUF_BITMAP_FREE, val, mtr);

}

/************************************************************************//**

Sets the free bit of the page in the ibuf bitmap. This is done in a separate

mini-transaction, hence this operation does not restrict further work to only

ibuf bitmap operations, which would result if the latch to the bitmap page

were kept. */

UNIV_INTERN

void

ibuf_set_free_bits_func(

/*====================*/

	buf_block_t*	block,	/*!< in: index page of a non-clustered index;

				free bit is reset if page level is 0 */

#ifdef UNIV_IBUF_DEBUG

	ulint		max_val,/*!< in: ULINT_UNDEFINED or a maximum

				value which the bits must have before

				setting; this is for debugging */

#endif /* UNIV_IBUF_DEBUG */

	ulint		val)	/*!< in: value to set: < 4 */

{

	mtr_t	mtr;

	page_t*	page;

	page_t*	bitmap_page;

	ulint	space;

	ulint	page_no;

	ulint	zip_size;

	page = buf_block_get_frame(block);

	if (!page_is_leaf(page)) {

		return;

	}

	mtr_start(&mtr);

	space = buf_block_get_space(block);

	page_no = buf_block_get_page_no(block);

	zip_size = buf_block_get_zip_size(block);

	bitmap_page = ibuf_bitmap_get_map_page(space, page_no, zip_size, &mtr);

#ifdef UNIV_IBUF_DEBUG

	if (max_val != ULINT_UNDEFINED) {

		ulint	old_val;

		old_val = ibuf_bitmap_page_get_bits(

			bitmap_page, page_no, zip_size,

			IBUF_BITMAP_FREE, &mtr);

# if 0

		if (old_val != max_val) {

			fprintf(stderr,

				"Ibuf: page %lu old val %lu max val %lu\n",

				page_get_page_no(page),

				old_val, max_val);

		}

# endif