mariadb/trx/trx0rec.c

/******************************************************
Transaction undo log record

(c) 1996 Innobase Oy

Created 3/26/1996 Heikki Tuuri
*******************************************************/

#include "trx0rec.h"

#ifdef UNIV_NONINL
#include "trx0rec.ic"
#endif

#include "fsp0fsp.h"
#include "mach0data.h"
#include "trx0rseg.h"
#include "trx0trx.h"
#include "trx0undo.h"
#include "dict0dict.h"
#include "ut0mem.h"
#include "row0ext.h"
#include "row0upd.h"
#include "que0que.h"
#include "trx0purge.h"
#include "row0row.h"

/*=========== UNDO LOG RECORD CREATION AND DECODING ====================*/

/**************************************************************************
Writes the mtr log entry of the inserted undo log record on the undo log
page. */
UNIV_INLINE
void
trx_undof_page_add_undo_rec_log(
/*============================*/
	page_t* undo_page,	/* in: undo log page */
	ulint	old_free,	/* in: start offset of the inserted entry */
	ulint	new_free,	/* in: end offset of the entry */
	mtr_t*	mtr)		/* in: mtr */
{
	byte*		log_ptr;
	const byte*	log_end;
	ulint		len;

	log_ptr = mlog_open(mtr, 11 + 13 + MLOG_BUF_MARGIN);

	if (log_ptr == NULL) {

		return;
	}

	log_end = &log_ptr[11 + 13 + MLOG_BUF_MARGIN];
	log_ptr = mlog_write_initial_log_record_fast(
		undo_page, MLOG_UNDO_INSERT, log_ptr, mtr);
	len = new_free - old_free - 4;

	mach_write_to_2(log_ptr, len);
	log_ptr += 2;

	if (log_ptr + len <= log_end) {
		memcpy(log_ptr, undo_page + old_free + 2, len);
		mlog_close(mtr, log_ptr + len);
	} else {
		mlog_close(mtr, log_ptr);
		mlog_catenate_string(mtr, undo_page + old_free + 2, len);
	}
}

/***************************************************************
Parses a redo log record of adding an undo log record. */

byte*
trx_undo_parse_add_undo_rec(
/*========================*/
			/* out: end of log record or NULL */
	byte*	ptr,	/* in: buffer */
	byte*	end_ptr,/* in: buffer end */
	page_t*	page)	/* in: page or NULL */
{
	ulint	len;
	byte*	rec;
	ulint	first_free;

	if (end_ptr < ptr + 2) {

		return(NULL);
	}

	len = mach_read_from_2(ptr);
	ptr += 2;

	if (end_ptr < ptr + len) {

		return(NULL);
	}

	if (page == NULL) {

		return(ptr + len);
	}

	first_free = mach_read_from_2(page + TRX_UNDO_PAGE_HDR
				      + TRX_UNDO_PAGE_FREE);
	rec = page + first_free;

	mach_write_to_2(rec, first_free + 4 + len);
	mach_write_to_2(rec + 2 + len, first_free);

	mach_write_to_2(page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE,
			first_free + 4 + len);
	ut_memcpy(rec + 2, ptr, len);

	return(ptr + len);
}

/**************************************************************************
Calculates the free space left for extending an undo log record. */
UNIV_INLINE
ulint
trx_undo_left(
/*==========*/
				/* out: bytes left */
	const page_t*	page,	/* in: undo log page */
	const byte*	ptr)	/* in: pointer to page */
{
	/* The '- 10' is a safety margin, in case we have some small
	calculation error below */

	return(UNIV_PAGE_SIZE - (ptr - page) - 10 - FIL_PAGE_DATA_END);
}

/**************************************************************************
Get the pointer to where the data for the undo log record will be written.*/
static
byte*
trx_undo_page_get_ptr(
/*==================*/
					/* out: ptr to where the undo log
					record data will be written,
					0 if not enough space.*/
	page_t*		undo_page,	/* in: undo log page */
	ulint		need)		/* in: need these man bytes */
{
	byte*		ptr;		/* pointer within undo_page */
	ulint		first_free;	/* offset within undo page */

	ut_ad(undo_page);

	ut_ad(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
			       + TRX_UNDO_PAGE_TYPE) == TRX_UNDO_INSERT);

	first_free = mach_read_from_2(
		undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE);

	/* Start writing the undo information from the first free
	bytes in the undo page */
	ptr = undo_page + first_free;

	ut_ad(first_free <= UNIV_PAGE_SIZE);

	/* NOTE: the value need must be big enough such that the
	general fields written below fit on the undo log page */

	if (UNIV_UNLIKELY(trx_undo_left(undo_page, ptr) < need)) {

		/* Error, not enough space */
		ptr = 0;

	} else {
		/* Reserve 2 bytes for the pointer to the next undo log
		record */
		ptr += 2;
	}

	return(ptr);
}

/**************************************************************************
Set the next and previous pointers in the undo page for the undo record
that was written to ptr. Update the first free value by the number of bytes
written for this undo record.*/
static
ulint
trx_undo_page_set_next_prev_and_add(
/*================================*/
					/* out: offset of the inserted entry
					on the page if succeeded, 0 if fail */
	page_t*		undo_page,	/* in/out: undo log page */
	byte*		ptr,		/* in: ptr up to where data has been
					written on this undo page. */
	mtr_t*		mtr)		/* in: mtr */
{
	ulint		first_free;	/* offset within undo_page */
	ulint		end_of_rec;	/* offset within undo_page */
	byte*		ptr_to_first_free;
					/* pointer within undo_page
					that points to the next free
					offset value within undo_page.*/

	ut_ad(ptr > undo_page);
	ut_ad(ptr < undo_page + UNIV_PAGE_SIZE);

	if (UNIV_UNLIKELY(trx_undo_left(undo_page, ptr) < 2)) {

		return(0);
	}

	ptr_to_first_free = undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE;

	first_free = mach_read_from_2(ptr_to_first_free);

	/* Write offset of the previous undo log record */
	mach_write_to_2(ptr, first_free);
	ptr += 2;

	end_of_rec = ptr - undo_page;

	/* Write offset of the next undo log record */
	mach_write_to_2(undo_page + first_free, end_of_rec);

	/* Update the offset to first free undo record */
	mach_write_to_2(ptr_to_first_free, end_of_rec);

	/* Write this log entry to the UNDO log */
	trx_undof_page_add_undo_rec_log(undo_page, first_free,
					end_of_rec, mtr);

	return(first_free);
}

/**************************************************************************
Reports in the undo log of an insert of a clustered index record. */
static
ulint
trx_undo_page_report_insert(
/*========================*/
					/* out: offset of the inserted entry
					on the page if succeed, 0 if fail */
	page_t*		undo_page,	/* in: undo log page */
	trx_t*		trx,		/* in: transaction */
	dict_index_t*	index,		/* in: clustered index */
	const dtuple_t*	clust_entry,	/* in: index entry which will be
					inserted to the clustered index */
	mtr_t*		mtr)		/* in: mtr */
{
	ulint		first_free;
	byte*		ptr;
	ulint		i;

	ut_ad(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
			       + TRX_UNDO_PAGE_TYPE) == TRX_UNDO_INSERT);

	first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
				      + TRX_UNDO_PAGE_FREE);
	ptr = undo_page + first_free;

	ut_ad(first_free <= UNIV_PAGE_SIZE);

	if (trx_undo_left(undo_page, ptr) < 2 + 1 + 11 + 11) {

		/* Not enough space for writing the general parameters */

		return(0);
	}

	/* Reserve 2 bytes for the pointer to the next undo log record */
	ptr += 2;

	/* Store first some general parameters to the undo log */
	*ptr++ = TRX_UNDO_INSERT_REC;
	ptr += mach_dulint_write_much_compressed(ptr, trx->undo_no);
	ptr += mach_dulint_write_much_compressed(ptr, index->table->id);
	/*----------------------------------------*/
	/* Store then the fields required to uniquely determine the record
	to be inserted in the clustered index */

	for (i = 0; i < dict_index_get_n_unique(index); i++) {

		const dfield_t*	field	= dtuple_get_nth_field(clust_entry, i);
		ulint		flen	= dfield_get_len(field);

		if (trx_undo_left(undo_page, ptr) < 5) {

			return(0);
		}

		ptr += mach_write_compressed(ptr, flen);

		if (flen != UNIV_SQL_NULL) {
			if (trx_undo_left(undo_page, ptr) < flen) {

				return(0);
			}

			ut_memcpy(ptr, dfield_get_data(field), flen);
			ptr += flen;
		}
	}

	return(trx_undo_page_set_next_prev_and_add(undo_page, ptr, mtr));
}

/**************************************************************************
Reports in the undo log of an index create */
static
ulint
trx_undo_page_report_index_create(
/*==============================*/
					/* out: offset of the inserted entry
					on the page if succeed, 0 if fail */
	page_t*		undo_page,	/* in: undo log page */
	trx_t*		trx,		/* in: transaction */
	dict_index_t*	index,		/* in: index */
	mtr_t*		mtr)		/* in: mtr */
{
	byte*		ptr;

	ut_ad(undo_page && trx && index && mtr);

	/* Get the pointer to where we will write our undo data. */

	ptr = trx_undo_page_get_ptr(undo_page, 1 + 11 + 1 + 11 + 11);

	if (UNIV_UNLIKELY(!ptr)) {
		return(0);
	}

	/* This is our internal dictionary undo log record. */
	*ptr++ = TRX_UNDO_DICTIONARY_REC;

	ptr += mach_dulint_write_much_compressed(ptr, trx->undo_no);

	/* The sub type (discriminator) of this undo dictionary record */
	*ptr++ = TRX_UNDO_INDEX_CREATE_REC;

	/* For index create, we need both the table id and the index id
	to be stored in the undo log record.*/

	ptr += mach_dulint_write_much_compressed(ptr, index->table->id);
	ptr += mach_dulint_write_much_compressed(ptr, index->id);

	return(trx_undo_page_set_next_prev_and_add(undo_page, ptr, mtr));
}

/**************************************************************************
Reports in the undo log of a table create */
static
ulint
trx_undo_page_report_table_create(
/*==============================*/
					/* out: offset of the inserted entry
					on the page if succeed, 0 if fail */
	page_t*		undo_page,	/* in: undo log page */
	trx_t*		trx,		/* in: transaction */
	const char*	table_name,	/* in: table name */
	mtr_t*		mtr)		/* in: mtr */
{
	byte*		ptr;
	ulint		name_len;

	ut_ad(undo_page && trx && table_name && mtr);

	name_len = strlen(table_name) + 1;

	/* Get the pointer to where we will write our undo data */

	ptr = trx_undo_page_get_ptr(undo_page, 1 + 11 + 1 + name_len);

	if (UNIV_UNLIKELY(!ptr)) {
		return(0);
	}

	/* The type (discriminator) of this undo log */
	*ptr++ = TRX_UNDO_DICTIONARY_REC;

	ptr += mach_dulint_write_much_compressed(ptr, trx->undo_no);

	/* The sub type (discriminator) of this dictionary undo log */
	*ptr++ = TRX_UNDO_TABLE_CREATE_REC;

	/* For table create we need to store table name */
	memcpy(ptr, table_name, name_len);
	ptr += name_len;

	return(trx_undo_page_set_next_prev_and_add(undo_page, ptr, mtr));
}

/**************************************************************************
Reports in the undo log of a table drop */
static
ulint
trx_undo_page_report_table_drop(
/*============================*/
					/* out: offset of the inserted entry
					on the page if succeed, 0 if fail */
	page_t*		undo_page,	/* in: undo log page */
	trx_t*		trx,		/* in: transaction */
	const char*	table_name,	/* in: table name */
	mtr_t*		mtr)		/* in: mtr */
{
	byte*		ptr;
	ulint		name_len;

	ut_ad(undo_page && trx && table_name && mtr);

	name_len = strlen(table_name) + 1;

	/* Get the pointer to where we will write our undo data */

	ptr = trx_undo_page_get_ptr(undo_page, 1 + 11 + 1 + name_len);

	if (UNIV_UNLIKELY(!ptr)) {
		return(0);
	}

	*ptr++ = TRX_UNDO_DICTIONARY_REC;

	ptr += mach_dulint_write_much_compressed(ptr, trx->undo_no);

	/* The sub type (discriminator) of this dictionary undo log */
	*ptr++ = TRX_UNDO_TABLE_DROP_REC;

	/* For table drop we need to store a table name */
	memcpy(ptr, table_name, name_len);
	ptr += name_len;

	return(trx_undo_page_set_next_prev_and_add(undo_page, ptr, mtr));
}

/**************************************************************************
Reports in the undo log of a table rename */
static
ulint
trx_undo_page_report_table_rename(
/*==============================*/
					/* out: offset of the inserted entry
					on the page if succeed, 0 if fail */
	page_t*		undo_page,	/* in: undo log page */
	trx_t*		trx,		/* in: transaction */
	const char*	new_table_name,	/* in: new table name */
	const char*	old_table_name,	/* in: old table name */
	const char*	tmp_table_name,	/* in: the temp name */
	mtr_t*		mtr)		/* in: mtr */
{
	byte*		ptr;
	ulint		new_name_len;
	ulint		old_name_len;
	ulint		tmp_name_len;

	ut_ad(undo_page && trx && new_table_name && old_table_name && mtr);

	new_name_len = strlen(new_table_name) + 1;
	old_name_len = strlen(old_table_name) + 1;
	tmp_name_len = strlen(tmp_table_name) + 1;

	/* Get the pointer to where we will write our undo data. */

	ptr = trx_undo_page_get_ptr(undo_page, 1 + 11 + 1
				    + new_name_len + old_name_len
				    + tmp_name_len);

	if (UNIV_UNLIKELY(!ptr)) {
		return(0);
	}

	/* The type (discriminator) of this undo log */
	*ptr++ = TRX_UNDO_DICTIONARY_REC;
	ptr += mach_dulint_write_much_compressed(ptr, trx->undo_no);

	/* The sub type (discriminator) of this dictionary undo log */
	*ptr++ = TRX_UNDO_TABLE_RENAME_REC;

	/* For table rename we need to store the new table name and
	the old table name */

	memcpy(ptr, new_table_name, new_name_len);
	ptr += new_name_len;

	memcpy(ptr, old_table_name, old_name_len);
	ptr += old_name_len;

	memcpy(ptr, tmp_table_name, tmp_name_len);
	ptr += tmp_name_len;

	return(trx_undo_page_set_next_prev_and_add(undo_page, ptr, mtr));
}

/**************************************************************************
Reads from an undo log record the general parameters. */

byte*
trx_undo_rec_get_pars(
/*==================*/
					/* out: remaining part of undo log
					record after reading these values */
	trx_undo_rec_t*	undo_rec,	/* in: undo log record */
	ulint*		type,		/* out: undo record type:
					TRX_UNDO_INSERT_REC, ... */
	ulint*		cmpl_info,	/* out: compiler info, relevant only
					for update type records */
	ibool*		updated_extern,	/* out: TRUE if we updated an
					externally stored fild */
	dulint*		undo_no,	/* out: undo log record number */
	dulint*		table_id)	/* out: table id */
{
	byte*		ptr;
	ulint		type_cmpl;

	ptr = undo_rec + 2;

	type_cmpl = mach_read_from_1(ptr);
	ptr++;

	if (type_cmpl & TRX_UNDO_UPD_EXTERN) {
		*updated_extern = TRUE;
		type_cmpl -= TRX_UNDO_UPD_EXTERN;
	} else {
		*updated_extern = FALSE;
	}

	*type = type_cmpl & (TRX_UNDO_CMPL_INFO_MULT - 1);
	*cmpl_info = type_cmpl / TRX_UNDO_CMPL_INFO_MULT;

	*undo_no = mach_dulint_read_much_compressed(ptr);
	ptr += mach_dulint_get_much_compressed_size(*undo_no);

	*table_id = mach_dulint_read_much_compressed(ptr);
	ptr += mach_dulint_get_much_compressed_size(*table_id);

	return(ptr);
}

/**************************************************************************
Reads from an undo log record a stored column value. */
static
byte*
trx_undo_rec_get_col_val(
/*=====================*/
			/* out: remaining part of undo log record after
			reading these values */
	byte*	ptr,	/* in: pointer to remaining part of undo log record */
	byte**	field,	/* out: pointer to stored field */
	ulint*	len)	/* out: length of the field, or UNIV_SQL_NULL */
{
	*len = mach_read_compressed(ptr);
	ptr += mach_get_compressed_size(*len);

	*field = ptr;

	if (*len != UNIV_SQL_NULL) {
		if (*len >= UNIV_EXTERN_STORAGE_FIELD) {
			ptr += (*len - UNIV_EXTERN_STORAGE_FIELD);
		} else {
			ptr += *len;
		}
	}

	return(ptr);
}

/***********************************************************************
Builds a row reference from an undo log record. */

byte*
trx_undo_rec_get_row_ref(
/*=====================*/
				/* out: pointer to remaining part of undo
				record */
	byte*		ptr,	/* in: remaining part of a copy of an undo log
				record, at the start of the row reference;
				NOTE that this copy of the undo log record must
				be preserved as long as the row reference is
				used, as we do NOT copy the data in the
				record! */
	dict_index_t*	index,	/* in: clustered index */
	dtuple_t**	ref,	/* out, own: row reference */
	mem_heap_t*	heap)	/* in: memory heap from which the memory
				needed is allocated */
{
	dfield_t*	dfield;
	byte*		field;
	ulint		len;
	ulint		ref_len;
	ulint		i;

	ut_ad(index && ptr && ref && heap);
	ut_a(dict_index_is_clust(index));

	ref_len = dict_index_get_n_unique(index);

	*ref = dtuple_create(heap, ref_len);

	dict_index_copy_types(*ref, index, ref_len);

	for (i = 0; i < ref_len; i++) {
		dfield = dtuple_get_nth_field(*ref, i);

		ptr = trx_undo_rec_get_col_val(ptr, &field, &len);

		dfield_set_data(dfield, field, len);
	}

	return(ptr);
}

/***********************************************************************
Skips a row reference from an undo log record. */

byte*
trx_undo_rec_skip_row_ref(
/*======================*/
				/* out: pointer to remaining part of undo
				record */
	byte*		ptr,	/* in: remaining part in update undo log
				record, at the start of the row reference */
	dict_index_t*	index)	/* in: clustered index */
{
	byte*	field;
	ulint	len;
	ulint	ref_len;
	ulint	i;

	ut_ad(index && ptr);
	ut_a(dict_index_is_clust(index));

	ref_len = dict_index_get_n_unique(index);

	for (i = 0; i < ref_len; i++) {
		ptr = trx_undo_rec_get_col_val(ptr, &field, &len);
	}

	return(ptr);
}

/**************************************************************************
Reports in the undo log of an update or delete marking of a clustered index
record. */
static
ulint
trx_undo_page_report_modify(
/*========================*/
					/* out: byte offset of the inserted
					undo log entry on the page if succeed,
					0 if fail */
	page_t*		undo_page,	/* in: undo log page */
	trx_t*		trx,		/* in: transaction */
	dict_index_t*	index,		/* in: clustered index where update or
					delete marking is done */
	const rec_t*	rec,		/* in: clustered index record which
					has NOT yet been modified */
	const ulint*	offsets,	/* in: rec_get_offsets(rec, index) */
	const upd_t*	update,		/* in: update vector which tells the
					columns to be updated; in the case of
					a delete, this should be set to NULL */
	ulint		cmpl_info,	/* in: compiler info on secondary
					index updates */
	mtr_t*		mtr)		/* in: mtr */
{
	dict_table_t*	table;
	ulint		first_free;
	byte*		ptr;
	const byte*	field;
	ulint		flen;
	ulint		col_no;
	ulint		type_cmpl;
	byte*		type_cmpl_ptr;
	ulint		i;

	ut_a(dict_index_is_clust(index));
	ut_ad(rec_offs_validate(rec, index, offsets));
	ut_ad(mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
			       + TRX_UNDO_PAGE_TYPE) == TRX_UNDO_UPDATE);
	table = index->table;

	first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
				      + TRX_UNDO_PAGE_FREE);
	ptr = undo_page + first_free;

	ut_ad(first_free <= UNIV_PAGE_SIZE);

	if (trx_undo_left(undo_page, ptr) < 50) {

		/* NOTE: the value 50 must be big enough so that the general
		fields written below fit on the undo log page */

		return(0);
	}

	/* Reserve 2 bytes for the pointer to the next undo log record */
	ptr += 2;

	/* Store first some general parameters to the undo log */

	if (!update) {
		type_cmpl = TRX_UNDO_DEL_MARK_REC;
	} else if (rec_get_deleted_flag(rec, dict_table_is_comp(table))) {
		type_cmpl = TRX_UNDO_UPD_DEL_REC;
	} else {
		type_cmpl = TRX_UNDO_UPD_EXIST_REC;
	}

	type_cmpl |= cmpl_info * TRX_UNDO_CMPL_INFO_MULT;
	type_cmpl_ptr = ptr;

	*ptr++ = type_cmpl;
	ptr += mach_dulint_write_much_compressed(ptr, trx->undo_no);

	ptr += mach_dulint_write_much_compressed(ptr, table->id);

	/*----------------------------------------*/
	/* Store the state of the info bits */

	*ptr++ = rec_get_info_bits(rec, dict_table_is_comp(table));

	/* Store the values of the system columns */
	field = rec_get_nth_field(rec, offsets,
				  dict_index_get_sys_col_pos(
					  index, DATA_TRX_ID), &flen);
	ut_ad(flen == DATA_TRX_ID_LEN);

	ptr += mach_dulint_write_compressed(ptr, trx_read_trx_id(field));

	field = rec_get_nth_field(rec, offsets,
				  dict_index_get_sys_col_pos(
					  index, DATA_ROLL_PTR), &flen);
	ut_ad(flen == DATA_ROLL_PTR_LEN);

	ptr += mach_dulint_write_compressed(ptr, trx_read_roll_ptr(field));

	/*----------------------------------------*/
	/* Store then the fields required to uniquely determine the
	record which will be modified in the clustered index */

	for (i = 0; i < dict_index_get_n_unique(index); i++) {

		field = rec_get_nth_field(rec, offsets, i, &flen);

		if (trx_undo_left(undo_page, ptr) < 4) {

			return(0);
		}

		ptr += mach_write_compressed(ptr, flen);

		if (flen != UNIV_SQL_NULL) {
			if (trx_undo_left(undo_page, ptr) < flen) {

				return(0);
			}

			ut_memcpy(ptr, field, flen);
			ptr += flen;
		}
	}

	/*----------------------------------------*/
	/* Save to the undo log the old values of the columns to be updated. */

	if (update) {
		if (trx_undo_left(undo_page, ptr) < 5) {

			return(0);
		}

		ptr += mach_write_compressed(ptr, upd_get_n_fields(update));

		for (i = 0; i < upd_get_n_fields(update); i++) {

			ulint	pos = upd_get_nth_field(update, i)->field_no;

			/* Write field number to undo log */
			if (trx_undo_left(undo_page, ptr) < 5) {

				return(0);
			}

			ptr += mach_write_compressed(ptr, pos);

			/* Save the old value of field */
			field = rec_get_nth_field(rec, offsets, pos, &flen);

			if (trx_undo_left(undo_page, ptr) < 5) {

				return(0);
			}

			if (rec_offs_nth_extern(offsets, pos)) {
				/* If a field has external storage, we add
				to flen the flag */

				ptr += mach_write_compressed(
					ptr,
					UNIV_EXTERN_STORAGE_FIELD + flen);

				/* Notify purge that it eventually has to
				free the old externally stored field */

				trx->update_undo->del_marks = TRUE;

				*type_cmpl_ptr |= TRX_UNDO_UPD_EXTERN;
			} else {
				ptr += mach_write_compressed(ptr, flen);
			}

			if (flen != UNIV_SQL_NULL) {
				if (trx_undo_left(undo_page, ptr) < flen) {

					return(0);
				}

				ut_memcpy(ptr, field, flen);
				ptr += flen;
			}
		}
	}

	/*----------------------------------------*/
	/* In the case of a delete marking, and also in the case of an update
	where any ordering field of any index changes, store the values of all
	columns which occur as ordering fields in any index. This info is used
	in the purge of old versions where we use it to build and search the
	delete marked index records, to look if we can remove them from the
	index tree. Note that starting from 4.0.14 also externally stored
	fields can be ordering in some index. Starting from 5.2, we no longer
	store REC_MAX_INDEX_COL_LEN first bytes to the undo log record,
	but we can construct the column prefix fields in the index by
	fetching the first page of the BLOB that is pointed to by the
	clustered index. This works also in crash recovery, because all pages
	(including BLOBs) are recovered before anything is rolled back. */

	if (!update || !(cmpl_info & UPD_NODE_NO_ORD_CHANGE)) {
		byte*	old_ptr = ptr;

		trx->update_undo->del_marks = TRUE;

		if (trx_undo_left(undo_page, ptr) < 5) {

			return(0);
		}

		/* Reserve 2 bytes to write the number of bytes the stored
		fields take in this undo record */

		ptr += 2;

		for (col_no = 0; col_no < dict_table_get_n_cols(table);
		     col_no++) {

			const dict_col_t*	col
				= dict_table_get_nth_col(table, col_no);

			if (col->ord_part) {
				ulint	pos;

				/* Write field number to undo log */
				if (trx_undo_left(undo_page, ptr) < 5 + 5) {

					return(0);
				}

				pos = dict_index_get_nth_col_pos(index,
								 col_no);
				ptr += mach_write_compressed(ptr, pos);

				/* Save the old value of field */
				field = rec_get_nth_field(rec, offsets, pos,
							  &flen);

				ptr += mach_write_compressed(ptr, flen);

				if (flen != UNIV_SQL_NULL) {
					if (trx_undo_left(undo_page, ptr)
					    < flen) {

						return(0);
					}

					ut_memcpy(ptr, field, flen);
					ptr += flen;
				}
			}
		}

		mach_write_to_2(old_ptr, ptr - old_ptr);
	}

	/*----------------------------------------*/
	/* Write pointers to the previous and the next undo log records */
	if (trx_undo_left(undo_page, ptr) < 2) {

		return(0);
	}

	mach_write_to_2(ptr, first_free);
	ptr += 2;
	mach_write_to_2(undo_page + first_free, ptr - undo_page);

	mach_write_to_2(undo_page + TRX_UNDO_PAGE_HDR + TRX_UNDO_PAGE_FREE,
			ptr - undo_page);

	/* Write to the REDO log about this change in the UNDO log */

	trx_undof_page_add_undo_rec_log(undo_page, first_free,
					ptr - undo_page, mtr);
	return(first_free);
}

/**************************************************************************
Reads from an undo log update record the system field values of the old
version. */

byte*
trx_undo_update_rec_get_sys_cols(
/*=============================*/
				/* out: remaining part of undo log
				record after reading these values */
	byte*	ptr,		/* in: remaining part of undo log
				record after reading general
				parameters */
	dulint*	trx_id,		/* out: trx id */
	dulint*	roll_ptr,	/* out: roll ptr */
	ulint*	info_bits)	/* out: info bits state */
{
	/* Read the state of the info bits */
	*info_bits = mach_read_from_1(ptr);
	ptr += 1;

	/* Read the values of the system columns */

	*trx_id = mach_dulint_read_compressed(ptr);
	ptr += mach_dulint_get_compressed_size(*trx_id);

	*roll_ptr = mach_dulint_read_compressed(ptr);
	ptr += mach_dulint_get_compressed_size(*roll_ptr);

	return(ptr);
}

/**************************************************************************
Reads from an update undo log record the number of updated fields. */
UNIV_INLINE
byte*
trx_undo_update_rec_get_n_upd_fields(
/*=================================*/
			/* out: remaining part of undo log record after
			reading this value */
	byte*	ptr,	/* in: pointer to remaining part of undo log record */
	ulint*	n)	/* out: number of fields */
{
	*n = mach_read_compressed(ptr);
	ptr += mach_get_compressed_size(*n);

	return(ptr);
}

/**************************************************************************
Reads from an update undo log record a stored field number. */
UNIV_INLINE
byte*
trx_undo_update_rec_get_field_no(
/*=============================*/
			/* out: remaining part of undo log record after
			reading this value */
	byte*	ptr,	/* in: pointer to remaining part of undo log record */
	ulint*	field_no)/* out: field number */
{
	*field_no = mach_read_compressed(ptr);
	ptr += mach_get_compressed_size(*field_no);

	return(ptr);
}

/***********************************************************************
Builds an update vector based on a remaining part of an undo log record. */

byte*
trx_undo_update_rec_get_update(
/*===========================*/
				/* out: remaining part of the record,
				NULL if an error detected, which means that
				the record is corrupted */
	byte*		ptr,	/* in: remaining part in update undo log
				record, after reading the row reference
				NOTE that this copy of the undo log record must
				be preserved as long as the update vector is
				used, as we do NOT copy the data in the
				record! */
	dict_index_t*	index,	/* in: clustered index */
	ulint		type,	/* in: TRX_UNDO_UPD_EXIST_REC,
				TRX_UNDO_UPD_DEL_REC, or
				TRX_UNDO_DEL_MARK_REC; in the last case,
				only trx id and roll ptr fields are added to
				the update vector */
	dulint		trx_id,	/* in: transaction id from this undo record */
	dulint		roll_ptr,/* in: roll pointer from this undo record */
	ulint		info_bits,/* in: info bits from this undo record */
	trx_t*		trx,	/* in: transaction */
	mem_heap_t*	heap,	/* in: memory heap from which the memory
				needed is allocated */
	upd_t**		upd)	/* out, own: update vector */
{
	upd_field_t*	upd_field;
	upd_t*		update;
	ulint		n_fields;
	byte*		buf;
	byte*		field;
	ulint		len;
	ulint		field_no;
	ulint		i;

	ut_a(dict_index_is_clust(index));

	if (type != TRX_UNDO_DEL_MARK_REC) {
		ptr = trx_undo_update_rec_get_n_upd_fields(ptr, &n_fields);
	} else {
		n_fields = 0;
	}

	update = upd_create(n_fields + 2, heap);

	update->info_bits = info_bits;

	/* Store first trx id and roll ptr to update vector */

	upd_field = upd_get_nth_field(update, n_fields);
	buf = mem_heap_alloc(heap, DATA_TRX_ID_LEN);
	trx_write_trx_id(buf, trx_id);

	upd_field_set_field_no(upd_field,
			       dict_index_get_sys_col_pos(index, DATA_TRX_ID),
			       index, trx);
	dfield_set_data(&(upd_field->new_val), buf, DATA_TRX_ID_LEN);

	upd_field = upd_get_nth_field(update, n_fields + 1);
	buf = mem_heap_alloc(heap, DATA_ROLL_PTR_LEN);
	trx_write_roll_ptr(buf, roll_ptr);

	upd_field_set_field_no(
		upd_field, dict_index_get_sys_col_pos(index, DATA_ROLL_PTR),
		index, trx);
	dfield_set_data(&(upd_field->new_val), buf, DATA_ROLL_PTR_LEN);

	/* Store then the updated ordinary columns to the update vector */

	for (i = 0; i < n_fields; i++) {

		ptr = trx_undo_update_rec_get_field_no(ptr, &field_no);

		if (field_no >= dict_index_get_n_fields(index)) {
			fprintf(stderr,
				"InnoDB: Error: trying to access"
				" update undo rec field %lu in ",
				(ulong) field_no);
			dict_index_name_print(stderr, trx, index);
			fprintf(stderr, "\n"
				"InnoDB: but index has only %lu fields\n"
				"InnoDB: Submit a detailed bug report"
				" to http://bugs.mysql.com\n"
				"InnoDB: Run also CHECK TABLE ",
				(ulong) dict_index_get_n_fields(index));
			ut_print_name(stderr, trx, TRUE, index->table_name);
			fprintf(stderr, "\n"
				"InnoDB: n_fields = %lu, i = %lu, ptr %p\n",
				(ulong) n_fields, (ulong) i, ptr);
			return(NULL);
		}

		ptr = trx_undo_rec_get_col_val(ptr, &field, &len);

		upd_field = upd_get_nth_field(update, i);

		upd_field_set_field_no(upd_field, field_no, index, trx);

		if (len != UNIV_SQL_NULL && len >= UNIV_EXTERN_STORAGE_FIELD) {

			upd_field->extern_storage = TRUE;

			len -= UNIV_EXTERN_STORAGE_FIELD;
		}

		dfield_set_data(&(upd_field->new_val), field, len);
	}

	*upd = update;

	return(ptr);
}

/***********************************************************************
Builds a partial row from an update undo log record. It contains the
columns which occur as ordering in any index of the table. */

byte*
trx_undo_rec_get_partial_row(
/*=========================*/
				/* out: pointer to remaining part of undo
				record */
	byte*		ptr,	/* in: remaining part in update undo log
				record of a suitable type, at the start of
				the stored index columns;
				NOTE that this copy of the undo log record must
				be preserved as long as the partial row is
				used, as we do NOT copy the data in the
				record! */
	dict_index_t*	index,	/* in: clustered index */
	dtuple_t**	row,	/* out, own: partial row */
	row_ext_t**	ext,	/* out, own: prefix cache for
				externally stored columns */
	mem_heap_t*	heap)	/* in: memory heap from which the memory
				needed is allocated */
{
	const byte*	end_ptr;
	ulint		row_len;
	ulint		n_ext_cols;
	ulint*		ext_cols;

	ut_ad(index && ptr && row && ext && heap);

	row_len = dict_table_get_n_cols(index->table);
	n_ext_cols = 0;
	ext_cols = mem_heap_alloc(heap, row_len * sizeof *ext_cols);

	*row = dtuple_create(heap, row_len);

	dict_table_copy_types(*row, index->table);

	end_ptr = ptr + mach_read_from_2(ptr);
	ptr += 2;

	while (ptr != end_ptr) {
		dfield_t*	dfield;
		byte*		field;
		ulint		field_no;
		ulint		col_no;
		ulint		len;

		ptr = trx_undo_update_rec_get_field_no(ptr, &field_no);

		col_no = dict_index_get_nth_col_no(index, field_no);

		ptr = trx_undo_rec_get_col_val(ptr, &field, &len);

		if (len >= UNIV_EXTERN_STORAGE_FIELD) {
			ext_cols[n_ext_cols++] = col_no;
		}

		dfield = dtuple_get_nth_field(*row, col_no);

		dfield_set_data(dfield, field, len);
	}

	if (n_ext_cols) {
		*ext = row_ext_create(n_ext_cols, ext_cols,
				      dict_table_zip_size(index->table),
				      heap);
	} else {
		*ext = NULL;
	}

	return(ptr);
}

/***************************************************************************
Erases the unused undo log page end. */
static
void
trx_undo_erase_page_end(
/*====================*/
	page_t*	undo_page,	/* in: undo page whose end to erase */
	mtr_t*	mtr)		/* in: mtr */
{
	ulint	first_free;

	first_free = mach_read_from_2(undo_page + TRX_UNDO_PAGE_HDR
				      + TRX_UNDO_PAGE_FREE);
	memset(undo_page + first_free, 0xff,
	       (UNIV_PAGE_SIZE - FIL_PAGE_DATA_END) - first_free);

	mlog_write_initial_log_record(undo_page, MLOG_UNDO_ERASE_END, mtr);
}

/***************************************************************
Parses a redo log record of erasing of an undo page end. */

byte*
trx_undo_parse_erase_page_end(
/*==========================*/
			/* out: end of log record or NULL */
	byte*	ptr,	/* in: buffer */
	byte*	end_ptr __attribute__((unused)), /* in: buffer end */
	page_t*	page,	/* in: page or NULL */
	mtr_t*	mtr)	/* in: mtr or NULL */
{
	ut_ad(ptr && end_ptr);

	if (page == NULL) {

		return(ptr);
	}

	trx_undo_erase_page_end(page, mtr);

	return(ptr);
}

/***************************************************************************
Writes information to an undo log about an insert, update, or a delete marking
of a clustered index record. This information is used in a rollback of the
transaction and in consistent reads that must look to the history of this
transaction. */

ulint
trx_undo_report_row_operation(
/*==========================*/
					/* out: DB_SUCCESS or error code */
	ulint		flags,		/* in: if BTR_NO_UNDO_LOG_FLAG bit is
					set, does nothing */
	ulint		op_type,	/* in: TRX_UNDO_INSERT_OP or
					TRX_UNDO_MODIFY_OP */
	que_thr_t*	thr,		/* in: query thread */
	dict_index_t*	index,		/* in: clustered index */
	const dtuple_t*	clust_entry,	/* in: in the case of an insert,
					index entry to insert into the
					clustered index, otherwise NULL */
	const upd_t*	update,		/* in: in the case of an update,
					the update vector, otherwise NULL */
	ulint		cmpl_info,	/* in: compiler info on secondary
					index updates */
	const rec_t*	rec,		/* in: in case of an update or delete
					marking, the record in the clustered
					index, otherwise NULL */
	dulint*		roll_ptr)	/* out: rollback pointer to the
					inserted undo log record,
					ut_dulint_zero if BTR_NO_UNDO_LOG
					flag was specified */
{
	trx_t*		trx;
	trx_undo_t*	undo;
	ulint		page_no;
	trx_rseg_t*	rseg;
	mtr_t		mtr;
	ulint		err		= DB_SUCCESS;
	mem_heap_t*	heap		= NULL;
	ulint		offsets_[REC_OFFS_NORMAL_SIZE];
	ulint*		offsets		= offsets_;
	*offsets_ = (sizeof offsets_) / sizeof *offsets_;

	ut_a(dict_index_is_clust(index));

	if (flags & BTR_NO_UNDO_LOG_FLAG) {

		*roll_ptr = ut_dulint_zero;

		return(DB_SUCCESS);
	}

	ut_ad(thr);
	ut_ad((op_type != TRX_UNDO_INSERT_OP)
	      || (clust_entry && !update && !rec));

	trx = thr_get_trx(thr);
	rseg = trx->rseg;

	mutex_enter(&(trx->undo_mutex));

	/* If the undo log is not assigned yet, assign one */

	if (op_type == TRX_UNDO_INSERT_OP) {

		if (trx->insert_undo == NULL) {

			err = trx_undo_assign_undo(trx, TRX_UNDO_INSERT);
		}

		undo = trx->insert_undo;

		if (UNIV_UNLIKELY(!undo)) {
			/* Did not succeed */
			mutex_exit(&(trx->undo_mutex));

			return(err);
		}
	} else {
		ut_ad(op_type == TRX_UNDO_MODIFY_OP);

		if (trx->update_undo == NULL) {

			err = trx_undo_assign_undo(trx, TRX_UNDO_UPDATE);

		}

		undo = trx->update_undo;

		if (UNIV_UNLIKELY(!undo)) {
			/* Did not succeed */
			mutex_exit(&(trx->undo_mutex));
			return(err);
		}

		offsets = rec_get_offsets(rec, index, offsets,
					  ULINT_UNDEFINED, &heap);
	}

	page_no = undo->last_page_no;

	mtr_start(&mtr);

	for (;;) {
		buf_block_t*	undo_block;
		page_t*		undo_page;
		ulint		offset;

		undo_block = buf_page_get_gen(undo->space, undo->zip_size,
					      page_no, RW_X_LATCH,
					      undo->guess_block, BUF_GET,
					      __FILE__, __LINE__, &mtr);
#ifdef UNIV_SYNC_DEBUG
		buf_block_dbg_add_level(undo_block, SYNC_TRX_UNDO_PAGE);
#endif /* UNIV_SYNC_DEBUG */
		undo_page = buf_block_get_frame(undo_block);

		if (op_type == TRX_UNDO_INSERT_OP) {
			offset = trx_undo_page_report_insert(
				undo_page, trx, index, clust_entry, &mtr);
		} else {
			offset = trx_undo_page_report_modify(
				undo_page, trx, index, rec, offsets, update,
				cmpl_info, &mtr);
		}

		if (UNIV_UNLIKELY(offset == 0)) {
			/* The record did not fit on the page. We erase the
			end segment of the undo log page and write a log
			record of it: this is to ensure that in the debug
			version the replicate page constructed using the log
			records stays identical to the original page */

			trx_undo_erase_page_end(undo_page, &mtr);
			mtr_commit(&mtr);
		} else {
			/* Success */

			mtr_commit(&mtr);

			undo->empty = FALSE;
			undo->top_page_no = page_no;
			undo->top_offset  = offset;
			undo->top_undo_no = trx->undo_no;
			undo->guess_block = undo_block;

			UT_DULINT_INC(trx->undo_no);

			mutex_exit(&trx->undo_mutex);

			*roll_ptr = trx_undo_build_roll_ptr(
				op_type == TRX_UNDO_INSERT_OP,
				rseg->id, page_no, offset);
			if (UNIV_LIKELY_NULL(heap)) {
				mem_heap_free(heap);
			}
			return(DB_SUCCESS);
		}

		ut_ad(page_no == undo->last_page_no);

		/* We have to extend the undo log by one page */

		mtr_start(&mtr);

		/* When we add a page to an undo log, this is analogous to
		a pessimistic insert in a B-tree, and we must reserve the
		counterpart of the tree latch, which is the rseg mutex. */

		mutex_enter(&(rseg->mutex));

		page_no = trx_undo_add_page(trx, undo, &mtr);

		mutex_exit(&(rseg->mutex));

		if (UNIV_UNLIKELY(page_no == FIL_NULL)) {
			/* Did not succeed: out of space */

			mutex_exit(&(trx->undo_mutex));
			mtr_commit(&mtr);
			if (UNIV_LIKELY_NULL(heap)) {
				mem_heap_free(heap);
			}
			return(DB_OUT_OF_FILE_SPACE);
		}
	}
}

/***************************************************************************
Writes information to an undo log about dictionary operation e.g.
rename_table, create_table, create_index, drop table. This information
is used in a rollback of the transaction. */

ulint
trx_undo_report_dict_operation(
/*===========================*/
					/* out: DB_SUCCESS or error code */
	ulint		op_type,	/* in: TRX_UNDO_TABLE_CREATE_OP,
					TRX_UNDO_TABLE_RENAME_OP,
					TRX_UNDO_TABLE_DROP_OP, or
					TRX_UNDO_INDEX_CREATE_OP */
	trx_t*		trx,		/* in: trx */
	dict_index_t*	index,		/* in:
					if TRX_UNDO_INDEX_CREATE_OP
					index to be created*/
	const char*	table_name,	/* in: table name or NULL, used in
					create table, rename table and
					drop table*/
	const char*	old_table_name,	/* in: old table name or NULL.
					used in rename table */
	const char*	tmp_table_name, /* in: the intermediate name used
					for renaming */
	dulint*		roll_ptr)	/* out: rollback pointer to the
					inserted undo log record */
{
	trx_undo_t*	undo;
	ulint		page_no;
	trx_rseg_t*	rseg;
	mtr_t		mtr;

	ut_ad(trx);

#ifdef UNIV_DEBUG
	switch (op_type) {
	case TRX_UNDO_TABLE_RENAME_OP:
		ut_ad(old_table_name);
	case TRX_UNDO_TABLE_DROP_OP:
	case TRX_UNDO_TABLE_CREATE_OP:
		ut_ad(table_name);
		break;

	case TRX_UNDO_INDEX_CREATE_OP:
		ut_ad(index);
		break;
	default:
		ut_error;
	}
#endif

	rseg = trx->rseg;

	mutex_enter(&(trx->undo_mutex));

	/* If the undo log is not assigned yet, assign one */

	if (trx->insert_undo == NULL) {

		trx_undo_assign_undo(trx, TRX_UNDO_INSERT);
	}

	undo = trx->insert_undo;

	if (undo == NULL) {
		/* Did not succeed: out of space */
		mutex_exit(&(trx->undo_mutex));

		return(DB_OUT_OF_FILE_SPACE);
	}

	page_no = undo->last_page_no;

	mtr_start(&mtr);

	for (;;) {
		buf_block_t*	undo_block;
		ulint		offset;

		undo_block = buf_page_get_gen(undo->space, undo->zip_size,
					     page_no, RW_X_LATCH,
					     undo->guess_block, BUF_GET,
					     __FILE__, __LINE__, &mtr);

#ifdef UNIV_SYNC_DEBUG
		buf_block_dbg_add_level(undo_block, SYNC_TRX_UNDO_PAGE);
#endif /* UNIV_SYNC_DEBUG */

		switch (op_type) {

		case TRX_UNDO_TABLE_CREATE_OP:
			offset = trx_undo_page_report_table_create(
				undo_block->frame, trx, table_name, &mtr);
			break;

		case TRX_UNDO_INDEX_CREATE_OP:
			offset = trx_undo_page_report_index_create(
				undo_block->frame, trx, index, &mtr);
			break;

		case TRX_UNDO_TABLE_RENAME_OP:
			offset = trx_undo_page_report_table_rename(
				undo_block->frame, trx, table_name,
				old_table_name, tmp_table_name, &mtr);
			break;

		case TRX_UNDO_TABLE_DROP_OP:
			offset = trx_undo_page_report_table_drop(
				undo_block->frame, trx, table_name, &mtr);
			break;

		default:
			ut_print_timestamp(stderr);
			fprintf(stderr,
				"  InnoDB: [Error]: Undefined op_type = %lu "
				"at trx_undo_report_dict_operation\n",
				(ulong) op_type);

			mutex_enter(&kernel_mutex);
			trx_print(stderr, trx, 1024);
			mutex_exit(&kernel_mutex);

			return(DB_ERROR);
		}

		if (UNIV_UNLIKELY(offset == 0)) {
			/* The record did not fit on the page. We erase the
			end segment of the undo log page and write a log
			record of it: this is to ensure that in the debug
			version the replicate page constructed using the log
			records stays identical to the original page */

			trx_undo_erase_page_end(undo_block->frame, &mtr);
			mtr_commit(&mtr);
		} else {
			/* Success */
			mtr_commit(&mtr);

			undo->empty = FALSE;
			undo->top_page_no = page_no;
			undo->top_offset  = offset;
			undo->top_undo_no = trx->undo_no;
			undo->guess_block = undo_block;

			UT_DULINT_INC(trx->undo_no);

			mutex_exit(&(trx->undo_mutex));

			*roll_ptr = trx_undo_build_roll_ptr(TRUE, rseg->id,
							    page_no, offset);

			return(DB_SUCCESS);
		}

		ut_ad(page_no == undo->last_page_no);

		/* We have to extend the undo log by one page */

		mtr_start(&mtr);

		/* When we add a page to an undo log, this is analogous to
		a pessimistic insert in a B-tree, and we must reserve the
		counterpart of the tree latch, which is the rseg mutex. */

		mutex_enter(&(rseg->mutex));

		page_no = trx_undo_add_page(trx, undo, &mtr);

		mutex_exit(&(rseg->mutex));

		if (page_no == FIL_NULL) {
			/* Did not succeed: out of space */

			mutex_exit(&(trx->undo_mutex));
			mtr_commit(&mtr);

			return(DB_OUT_OF_FILE_SPACE);
		}
	}
}

/*============== BUILDING PREVIOUS VERSION OF A RECORD ===============*/

/**********************************************************************
Copies an undo record to heap. This function can be called if we know that
the undo log record exists. */

trx_undo_rec_t*
trx_undo_get_undo_rec_low(
/*======================*/
					/* out, own: copy of the record */
	dulint		roll_ptr,	/* in: roll pointer to record */
	mem_heap_t*	heap)		/* in: memory heap where copied */
{
	trx_undo_rec_t*	undo_rec;
	ulint		rseg_id;
	ulint		page_no;
	ulint		offset;
	page_t*		undo_page;
	trx_rseg_t*	rseg;
	ibool		is_insert;
	mtr_t		mtr;

	trx_undo_decode_roll_ptr(roll_ptr, &is_insert, &rseg_id, &page_no,
				 &offset);
	rseg = trx_rseg_get_on_id(rseg_id);

	mtr_start(&mtr);

	undo_page = trx_undo_page_get_s_latched(rseg->space, rseg->zip_size,
						page_no, &mtr);

	undo_rec = trx_undo_rec_copy(undo_page + offset, heap);

	mtr_commit(&mtr);

	return(undo_rec);
}

/**********************************************************************
Copies an undo record to heap. */

ulint
trx_undo_get_undo_rec(
/*==================*/
					/* out: DB_SUCCESS, or
					DB_MISSING_HISTORY if the undo log
					has been truncated and we cannot
					fetch the old version; NOTE: the
					caller must have latches on the
					clustered index page and purge_view */
	dulint		roll_ptr,	/* in: roll pointer to record */
	dulint		trx_id,		/* in: id of the trx that generated
					the roll pointer: it points to an
					undo log of this transaction */
	trx_undo_rec_t** undo_rec,	/* out, own: copy of the record */
	mem_heap_t*	heap)		/* in: memory heap where copied */
{
#ifdef UNIV_SYNC_DEBUG
	ut_ad(rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
#endif /* UNIV_SYNC_DEBUG */

	if (!trx_purge_update_undo_must_exist(trx_id)) {

		/* It may be that the necessary undo log has already been
		deleted */

		return(DB_MISSING_HISTORY);
	}

	*undo_rec = trx_undo_get_undo_rec_low(roll_ptr, heap);

	return(DB_SUCCESS);
}

/***********************************************************************
Build a previous version of a clustered index record. This function checks
that the caller has a latch on the index page of the clustered index record
and an s-latch on the purge_view. This guarantees that the stack of versions
is locked. */

ulint
trx_undo_prev_version_build(
/*========================*/
				/* out: DB_SUCCESS, or DB_MISSING_HISTORY if
				the previous version is not >= purge_view,
				which means that it may have been removed,
				DB_ERROR if corrupted record */
	rec_t*		index_rec,/* in: clustered index record in the
				index tree */
	mtr_t*		index_mtr __attribute__((unused)),
				/* in: mtr which contains the latch to
				index_rec page and purge_view */
	rec_t*		rec,	/* in: version of a clustered index record */
	dict_index_t*	index,	/* in: clustered index */
	ulint*		offsets,/* in: rec_get_offsets(rec, index) */
	mem_heap_t*	heap,	/* in: memory heap from which the memory
				needed is allocated */
	rec_t**		old_vers)/* out, own: previous version, or NULL if
				rec is the first inserted version, or if
				history data has been deleted */
{
	trx_undo_rec_t*	undo_rec;
	dtuple_t*	entry;
	dulint		rec_trx_id;
	ulint		type;
	dulint		undo_no;
	dulint		table_id;
	dulint		trx_id;
	dulint		roll_ptr;
	dulint		old_roll_ptr;
	upd_t*		update;
	byte*		ptr;
	ulint		info_bits;
	ulint		cmpl_info;
	ibool		dummy_extern;
	byte*		buf;
	ulint		err;
#ifdef UNIV_SYNC_DEBUG
	ut_ad(rw_lock_own(&(purge_sys->latch), RW_LOCK_SHARED));
#endif /* UNIV_SYNC_DEBUG */
	ut_ad(mtr_memo_contains_page(index_mtr, index_rec, MTR_MEMO_PAGE_S_FIX)
	      || mtr_memo_contains_page(index_mtr, index_rec,
					MTR_MEMO_PAGE_X_FIX));
	ut_ad(rec_offs_validate(rec, index, offsets));

	if (!dict_index_is_clust(index)) {
		fprintf(stderr, "InnoDB: Error: trying to access"
			" update undo rec for non-clustered index %s\n"
			"InnoDB: Submit a detailed bug report to"
			" http://bugs.mysql.com\n"
			"InnoDB: index record ", index->name);
		rec_print(stderr, index_rec, index);
		fputs("\n"
		      "InnoDB: record version ", stderr);
		rec_print_new(stderr, rec, offsets);
		putc('\n', stderr);
		return(DB_ERROR);
	}

	roll_ptr = row_get_rec_roll_ptr(rec, index, offsets);
	old_roll_ptr = roll_ptr;

	*old_vers = NULL;

	if (trx_undo_roll_ptr_is_insert(roll_ptr)) {

		/* The record rec is the first inserted version */

		return(DB_SUCCESS);
	}

	rec_trx_id = row_get_rec_trx_id(rec, index, offsets);

	err = trx_undo_get_undo_rec(roll_ptr, rec_trx_id, &undo_rec, heap);

	if (err != DB_SUCCESS) {

		return(err);
	}

	ptr = trx_undo_rec_get_pars(undo_rec, &type, &cmpl_info,
				    &dummy_extern, &undo_no, &table_id);

	ptr = trx_undo_update_rec_get_sys_cols(ptr, &trx_id, &roll_ptr,
					       &info_bits);
	ptr = trx_undo_rec_skip_row_ref(ptr, index);

	ptr = trx_undo_update_rec_get_update(ptr, index, type, trx_id,
					     roll_ptr, info_bits,
					     NULL, heap, &update);

	if (ut_dulint_cmp(table_id, index->table->id) != 0) {
		ptr = NULL;

		fprintf(stderr,
			"InnoDB: Error: trying to access update undo rec"
			" for table %s\n"
			"InnoDB: but the table id in the"
			" undo record is wrong\n"
			"InnoDB: Submit a detailed bug report"
			" to http://bugs.mysql.com\n"
			"InnoDB: Run also CHECK TABLE %s\n",
			index->table_name, index->table_name);
	}

	if (ptr == NULL) {
		/* The record was corrupted, return an error; these printfs
		should catch an elusive bug in row_vers_old_has_index_entry */

		fprintf(stderr,
			"InnoDB: table %s, index %s, n_uniq %lu\n"
			"InnoDB: undo rec address %p, type %lu cmpl_info %lu\n"
			"InnoDB: undo rec table id %lu %lu,"
			" index table id %lu %lu\n"
			"InnoDB: dump of 150 bytes in undo rec: ",
			index->table_name, index->name,
			(ulong) dict_index_get_n_unique(index),
			undo_rec, (ulong) type, (ulong) cmpl_info,
			(ulong) ut_dulint_get_high(table_id),
			(ulong) ut_dulint_get_low(table_id),
			(ulong) ut_dulint_get_high(index->table->id),
			(ulong) ut_dulint_get_low(index->table->id));
		ut_print_buf(stderr, undo_rec, 150);
		fputs("\n"
		      "InnoDB: index record ", stderr);
		rec_print(stderr, index_rec, index);
		fputs("\n"
		      "InnoDB: record version ", stderr);
		rec_print_new(stderr, rec, offsets);
		fprintf(stderr, "\n"
			"InnoDB: Record trx id %lu %lu, update rec"
			" trx id %lu %lu\n"
			"InnoDB: Roll ptr in rec %lu %lu, in update rec"
			" %lu %lu\n",
			(ulong) ut_dulint_get_high(rec_trx_id),
			(ulong) ut_dulint_get_low(rec_trx_id),
			(ulong) ut_dulint_get_high(trx_id),
			(ulong) ut_dulint_get_low(trx_id),
			(ulong) ut_dulint_get_high(old_roll_ptr),
			(ulong) ut_dulint_get_low(old_roll_ptr),
			(ulong) ut_dulint_get_high(roll_ptr),
			(ulong) ut_dulint_get_low(roll_ptr));

		trx_purge_sys_print();
		return(DB_ERROR);
	}

	if (row_upd_changes_field_size_or_external(index, offsets, update)) {
		ulint*	ext;
		ulint	n_ext;

		/* We have to set the appropriate extern storage bits in the
		old version of the record: the extern bits in rec for those
		fields that update does NOT update, as well as the the bits for
		those fields that update updates to become externally stored
		fields. Store the info to ext: */

		ext = mem_alloc(sizeof(ulint) * 2
				* rec_offs_n_fields(offsets));
		n_ext = btr_push_update_extern_fields(ext, offsets, update);
		entry = row_rec_to_index_entry(ROW_COPY_DATA, index, rec,
					       heap);
		row_upd_index_replace_new_col_vals(entry, index, update, heap);

		buf = mem_heap_alloc(heap, rec_get_converted_size(
					     index, entry, ext, n_ext));

		*old_vers = rec_convert_dtuple_to_rec(buf, index, entry,
						      ext, n_ext);

		mem_free(ext);
	} else {
		buf = mem_heap_alloc(heap, rec_offs_size(offsets));
		*old_vers = rec_copy(buf, rec, offsets);
		rec_offs_make_valid(*old_vers, index, offsets);
		row_upd_rec_in_place(*old_vers, index, offsets, update, NULL);
	}

	return(DB_SUCCESS);
}