innobase/include/page0zip.ic

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

Copyright (c) 2005, 2018, Oracle and/or its affiliates. All Rights Reserved.
Copyright (c) 2012, Facebook Inc.

This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License, version 2.0,
as published by the Free Software Foundation.

This program is also distributed with certain software (including
but not limited to OpenSSL) that is licensed under separate terms,
as designated in a particular file or component or in included license
documentation.  The authors of MySQL hereby grant you an additional
permission to link the program and your derivative works with the
separately licensed software that they have included with MySQL.

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, version 2.0, 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 Street, Suite 500, Boston, MA 02110-1335 USA

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

/**************************************************//**
@file include/page0zip.ic
Compressed page interface

Created June 2005 by Marko Makela
*******************************************************/

#ifdef UNIV_MATERIALIZE
# undef UNIV_INLINE
# define UNIV_INLINE
#endif

#include "page0zip.h"
#include "mtr0log.h"
#include "page0page.h"

/* The format of compressed pages is as follows.

The header and trailer of the uncompressed pages, excluding the page
directory in the trailer, are copied as is to the header and trailer
of the compressed page.

At the end of the compressed page, there is a dense page directory
pointing to every user record contained on the page, including deleted
records on the free list.  The dense directory is indexed in the
collation order, i.e., in the order in which the record list is
linked on the uncompressed page.  The infimum and supremum records are
excluded.  The two most significant bits of the entries are allocated
for the delete-mark and an n_owned flag indicating the last record in
a chain of records pointed to from the sparse page directory on the
uncompressed page.

The data between PAGE_ZIP_START and the last page directory entry will
be written in compressed format, starting at offset PAGE_DATA.
Infimum and supremum records are not stored.  We exclude the
REC_N_NEW_EXTRA_BYTES in every record header.  These can be recovered
from the dense page directory stored at the end of the compressed
page.

The fields node_ptr (in non-leaf B-tree nodes; level>0), trx_id and
roll_ptr (in leaf B-tree nodes; level=0), and BLOB pointers of
externally stored columns are stored separately, in ascending order of
heap_no and column index, starting backwards from the dense page
directory.

The compressed data stream may be followed by a modification log
covering the compressed portion of the page, as follows.

MODIFICATION LOG ENTRY FORMAT
- write record:
  - (heap_no - 1) << 1 (1..2 bytes)
  - extra bytes backwards
  - data bytes
- clear record:
  - (heap_no - 1) << 1 | 1 (1..2 bytes)

The integer values are stored in a variable-length format:
- 0xxxxxxx: 0..127
- 1xxxxxxx xxxxxxxx: 0..32767

The end of the modification log is marked by a 0 byte.

In summary, the compressed page looks like this:

(1) Uncompressed page header (PAGE_DATA bytes)
(2) Compressed index information
(3) Compressed page data
(4) Page modification log (page_zip->m_start..page_zip->m_end)
(5) Empty zero-filled space
(6) BLOB pointers (on leaf pages)
  - BTR_EXTERN_FIELD_REF_SIZE for each externally stored column
  - in descending collation order
(7) Uncompressed columns of user records, n_dense * uncompressed_size bytes,
  - indexed by heap_no
  - DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN for leaf pages of clustered indexes
  - REC_NODE_PTR_SIZE for non-leaf pages
  - 0 otherwise
(8) dense page directory, stored backwards
  - n_dense = n_heap - 2
  - existing records in ascending collation order
  - deleted records (free list) in link order
*/

/** Start offset of the area that will be compressed */
#define PAGE_ZIP_START		PAGE_NEW_SUPREMUM_END
/** Size of an compressed page directory entry */
#define PAGE_ZIP_DIR_SLOT_SIZE	2
/** Mask of record offsets */
#define PAGE_ZIP_DIR_SLOT_MASK	0x3fff
/** 'owned' flag */
#define PAGE_ZIP_DIR_SLOT_OWNED	0x4000
/** 'deleted' flag */
#define PAGE_ZIP_DIR_SLOT_DEL	0x8000

/**********************************************************************//**
Determine the size of a compressed page in bytes.
@return	size in bytes */
UNIV_INLINE
ulint
page_zip_get_size(
/*==============*/
	const page_zip_des_t*	page_zip)	/*!< in: compressed page */
{
	ulint	size;

	if (!page_zip->ssize) {
		return(0);
	}

	size = (UNIV_ZIP_SIZE_MIN >> 1) << page_zip->ssize;

	ut_ad(size >= UNIV_ZIP_SIZE_MIN);
	ut_ad(size <= UNIV_PAGE_SIZE);

	return(size);
}
/**********************************************************************//**
Set the size of a compressed page in bytes. */
UNIV_INLINE
void
page_zip_set_size(
/*==============*/
	page_zip_des_t*	page_zip,	/*!< in/out: compressed page */
	ulint		size)		/*!< in: size in bytes */
{
	if (size) {
		int	ssize;

		ut_ad(ut_is_2pow(size));

		for (ssize = 1; size > (ulint) (512 << ssize); ssize++) {
		}

		page_zip->ssize = ssize;
	} else {
		page_zip->ssize = 0;
	}

	ut_ad(page_zip_get_size(page_zip) == size);
}

#ifndef UNIV_HOTBACKUP
/**********************************************************************//**
Determine if a record is so big that it needs to be stored externally.
@return	FALSE if the entire record can be stored locally on the page */
UNIV_INLINE
ibool
page_zip_rec_needs_ext(
/*===================*/
	ulint	rec_size,	/*!< in: length of the record in bytes */
	ulint	comp,		/*!< in: nonzero=compact format */
	ulint	n_fields,	/*!< in: number of fields in the record;
				ignored if zip_size == 0 */
	ulint	zip_size)	/*!< in: compressed page size in bytes, or 0 */
{
	ut_ad(rec_size > (comp ? REC_N_NEW_EXTRA_BYTES : REC_N_OLD_EXTRA_BYTES));
	ut_ad(ut_is_2pow(zip_size));
	ut_ad(comp || !zip_size);

#if UNIV_PAGE_SIZE_MAX > REC_MAX_DATA_SIZE
	if (rec_size >= REC_MAX_DATA_SIZE) {
		return(TRUE);
	}
#endif

	if (zip_size) {
		ut_ad(comp);
		/* On a compressed page, there is a two-byte entry in
		the dense page directory for every record.  But there
		is no record header.  There should be enough room for
		one record on an empty leaf page.  Subtract 1 byte for
		the encoded heap number.  Check also the available space
		on the uncompressed page. */
		return(rec_size - (REC_N_NEW_EXTRA_BYTES - 2 - 1)
		       >= page_zip_empty_size(n_fields, zip_size)
		       || rec_size >= page_get_free_space_of_empty(TRUE) / 2);
	}

	return(rec_size >= page_get_free_space_of_empty(comp) / 2);
}
#endif /* !UNIV_HOTBACKUP */

#ifdef UNIV_DEBUG
/**********************************************************************//**
Validate a compressed page descriptor.
@return	TRUE if ok */
UNIV_INLINE
ibool
page_zip_simple_validate(
/*=====================*/
	const page_zip_des_t*	page_zip)/*!< in: compressed page descriptor */
{
	ut_ad(page_zip);
	ut_ad(page_zip->data);
	ut_ad(page_zip->ssize <= PAGE_ZIP_SSIZE_MAX);
	ut_ad(page_zip_get_size(page_zip)
	      > PAGE_DATA + PAGE_ZIP_DIR_SLOT_SIZE);
	ut_ad(page_zip->m_start <= page_zip->m_end);
	ut_ad(page_zip->m_end < page_zip_get_size(page_zip));
	ut_ad(page_zip->n_blobs
	      < page_zip_get_size(page_zip) / BTR_EXTERN_FIELD_REF_SIZE);
	return(TRUE);
}
#endif /* UNIV_DEBUG */

/**********************************************************************//**
Determine if the length of the page trailer.
@return length of the page trailer, in bytes, not including the
terminating zero byte of the modification log */
UNIV_INLINE
ibool
page_zip_get_trailer_len(
/*=====================*/
	const page_zip_des_t*	page_zip,/*!< in: compressed page */
	ibool			is_clust)/*!< in: TRUE if clustered index */
{
	ulint	uncompressed_size;

	ut_ad(page_zip_simple_validate(page_zip));
	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));

	if (!page_is_leaf(page_zip->data)) {
		uncompressed_size = PAGE_ZIP_DIR_SLOT_SIZE
			+ REC_NODE_PTR_SIZE;
		ut_ad(!page_zip->n_blobs);
	} else if (is_clust) {
		uncompressed_size = PAGE_ZIP_DIR_SLOT_SIZE
			+ DATA_TRX_ID_LEN + DATA_ROLL_PTR_LEN;
	} else {
		uncompressed_size = PAGE_ZIP_DIR_SLOT_SIZE;
		ut_ad(!page_zip->n_blobs);
	}

	return((page_dir_get_n_heap(page_zip->data) - 2)
	       * uncompressed_size
	       + page_zip->n_blobs * BTR_EXTERN_FIELD_REF_SIZE);
}

/**********************************************************************//**
Determine how big record can be inserted without recompressing the page.
@return a positive number indicating the maximum size of a record
whose insertion is guaranteed to succeed, or zero or negative */
UNIV_INLINE
lint
page_zip_max_ins_size(
/*==================*/
	const page_zip_des_t*	page_zip,/*!< in: compressed page */
	ibool			is_clust)/*!< in: TRUE if clustered index */
{
	ulint	trailer_len;

	trailer_len = page_zip_get_trailer_len(page_zip, is_clust);

	/* When a record is created, a pointer may be added to
	the dense directory.
	Likewise, space for the columns that will not be
	compressed will be allocated from the page trailer.
	Also the BLOB pointers will be allocated from there, but
	we may as well count them in the length of the record. */

	trailer_len += PAGE_ZIP_DIR_SLOT_SIZE;

	return((lint) page_zip_get_size(page_zip)
	       - trailer_len - page_zip->m_end
	       - (REC_N_NEW_EXTRA_BYTES - 2));
}

/**********************************************************************//**
Determine if enough space is available in the modification log.
@return	TRUE if enough space is available */
UNIV_INLINE
ibool
page_zip_available(
/*===============*/
	const page_zip_des_t*	page_zip,/*!< in: compressed page */
	ibool			is_clust,/*!< in: TRUE if clustered index */
	ulint			length,	/*!< in: combined size of the record */
	ulint			create)	/*!< in: nonzero=add the record to
					the heap */
{
	ulint	trailer_len;

	ut_ad(length > REC_N_NEW_EXTRA_BYTES);

	trailer_len = page_zip_get_trailer_len(page_zip, is_clust);

	/* Subtract the fixed extra bytes and add the maximum
	space needed for identifying the record (encoded heap_no). */
	length -= REC_N_NEW_EXTRA_BYTES - 2;

	if (create > 0) {
		/* When a record is created, a pointer may be added to
		the dense directory.
		Likewise, space for the columns that will not be
		compressed will be allocated from the page trailer.
		Also the BLOB pointers will be allocated from there, but
		we may as well count them in the length of the record. */

		trailer_len += PAGE_ZIP_DIR_SLOT_SIZE;
	}

	return(length + trailer_len + page_zip->m_end
	       < page_zip_get_size(page_zip));
}

/**********************************************************************//**
Initialize a compressed page descriptor. */
UNIV_INLINE
void
page_zip_des_init(
/*==============*/
	page_zip_des_t*	page_zip)	/*!< in/out: compressed page
					descriptor */
{
	memset(page_zip, 0, sizeof *page_zip);
}

/**********************************************************************//**
Write a log record of writing to the uncompressed header portion of a page. */
UNIV_INTERN
void
page_zip_write_header_log(
/*======================*/
	const byte*	data,/*!< in: data on the uncompressed page */
	ulint		length,	/*!< in: length of the data */
	mtr_t*		mtr);	/*!< in: mini-transaction */

/**********************************************************************//**
Write data to the uncompressed header portion of a page.  The data must
already have been written to the uncompressed page.
However, the data portion of the uncompressed page may differ from
the compressed page when a record is being inserted in
page_cur_insert_rec_zip(). */
UNIV_INLINE
void
page_zip_write_header(
/*==================*/
	page_zip_des_t*	page_zip,/*!< in/out: compressed page */
	const byte*	str,	/*!< in: address on the uncompressed page */
	ulint		length,	/*!< in: length of the data */
	mtr_t*		mtr)	/*!< in: mini-transaction, or NULL */
{
	ulint	pos;

	ut_ad(PAGE_ZIP_MATCH(str, page_zip));
	ut_ad(page_zip_simple_validate(page_zip));
	UNIV_MEM_ASSERT_RW(page_zip->data, page_zip_get_size(page_zip));

	pos = page_offset(str);

	ut_ad(pos < PAGE_DATA);

	memcpy(page_zip->data + pos, str, length);

	/* The following would fail in page_cur_insert_rec_zip(). */
	/* ut_ad(page_zip_validate(page_zip, str - pos)); */

	if (mtr) {
#ifndef UNIV_HOTBACKUP
		page_zip_write_header_log(str, length, mtr);
#endif /* !UNIV_HOTBACKUP */
	}
}

/**********************************************************************//**
Write a log record of compressing an index page without the data on the page. */
UNIV_INLINE
void
page_zip_compress_write_log_no_data(
/*================================*/
	ulint		level,	/*!< in: compression level */
	const page_t*	page,	/*!< in: page that is compressed */
	dict_index_t*	index,	/*!< in: index */
	mtr_t*		mtr)	/*!< in: mtr */
{
	byte* log_ptr = mlog_open_and_write_index(
		mtr, page, index, MLOG_ZIP_PAGE_COMPRESS_NO_DATA, 1);

	if (log_ptr) {
		mach_write_to_1(log_ptr, level);
		mlog_close(mtr, log_ptr + 1);
	}
}

/**********************************************************************//**
Parses a log record of compressing an index page without the data.
@return	end of log record or NULL */
UNIV_INLINE
byte*
page_zip_parse_compress_no_data(
/*============================*/
	byte*		ptr,		/*!< in: buffer */
	byte*		end_ptr,	/*!< in: buffer end */
	page_t*		page,		/*!< in: uncompressed page */
	page_zip_des_t*	page_zip,	/*!< out: compressed page */
	dict_index_t*	index)		/*!< in: index */
{
	ulint	level;
	if (end_ptr == ptr) {
		return(NULL);
	}

	level = mach_read_from_1(ptr);

	/* If page compression fails then there must be something wrong
	because a compress log record is logged only if the compression
	was successful. Crash in this case. */

	if (page
	    && !page_zip_compress(page_zip, page, index, level, NULL)) {
		ut_error;
	}

	return(ptr + 1);
}

/**********************************************************************//**
Reset the counters used for filling
INFORMATION_SCHEMA.innodb_cmp_per_index. */
UNIV_INLINE
void
page_zip_reset_stat_per_index()
/*===========================*/
{
	mutex_enter(&page_zip_stat_per_index_mutex);

	page_zip_stat_per_index.erase(
		page_zip_stat_per_index.begin(),
		page_zip_stat_per_index.end());

	mutex_exit(&page_zip_stat_per_index_mutex);
}

#ifdef UNIV_MATERIALIZE
# undef UNIV_INLINE
# define UNIV_INLINE	UNIV_INLINE_ORIGINAL
#endif