Revert revno:3452.71.32 (Bug#12612184 fix).

Bug#12612184 RACE CONDITION AFTER BTR_CUR_PESSIMISTIC_UPDATE()

The fix introduced potentially more severe crash recovery problems
than the bug causes. Revert the fix for now.

storage/innobase/btr/btr0btr.c

@@ -1948,7 +1948,7 @@ btr_node_ptr_delete(
 	ut_a(err == DB_SUCCESS);
 
 	if (!compressed) {
-		btr_cur_compress_if_useful(&cursor, FALSE, mtr);
+		btr_cur_compress_if_useful(&cursor, mtr);
 	}
 }
 
@@ -1956,10 +1956,9 @@ btr_node_ptr_delete(
 If page is the only on its level, this function moves its records to the
 father page, thus reducing the tree height. */
 static
-page_t*
+void
 btr_lift_page_up(
 /*=============*/
-				/* out: father page */
 	dict_index_t*	index,	/* in: index tree */
 	page_t*		page,	/* in: page which is the only on its level;
 				must not be empty: use
@@ -2035,8 +2034,6 @@ btr_lift_page_up(
 	ibuf_reset_free_bits(index, father_page);
 	ut_ad(page_validate(father_page, index));
 	ut_ad(btr_check_node_ptr(index, father_page, mtr));
-
-	return(father_page);
 }
 
 /*****************************************************************
@@ -2053,13 +2050,11 @@ enough free extents so that the compression will always succeed if done! */
 void
 btr_compress(
 /*=========*/
-	btr_cur_t*	cursor,	/* in/out: cursor on the page to merge
+	btr_cur_t*	cursor,	/* in: cursor on the page to merge or lift;
-				or lift; the page must not be empty:
+				the page must not be empty: in record delete
-				when deleting records, use btr_discard_page()
+				use btr_discard_page if the page would become
-				if the page would become empty */
+				empty */
-	ibool		adjust,	/* in: TRUE if should adjust the
+	mtr_t*		mtr)	/* in: mtr */
-				cursor position even if compression occurs */
-	mtr_t*		mtr)	/* in/out: mini-transaction */
 {
 	dict_index_t*	index;
 	ulint		space;
@@ -2074,7 +2069,6 @@ btr_compress(
 	rec_t*		node_ptr;
 	ulint		data_size;
 	ulint		n_recs;
-	ulint		nth_rec = 0; /* remove bogus warning */
 	ulint		max_ins_size;
 	ulint		max_ins_size_reorg;
 	ulint		comp;
@@ -2082,7 +2076,6 @@ btr_compress(
 	page = btr_cur_get_page(cursor);
 	index = btr_cur_get_index(cursor);
 	comp = page_is_comp(page);
-
 	ut_a((ibool)!!comp == dict_table_is_comp(index->table));
 
 	ut_ad(mtr_memo_contains(mtr, dict_index_get_lock(index),
@@ -2104,10 +2097,6 @@ btr_compress(
 	father_page = buf_frame_align(node_ptr);
 	ut_a(comp == page_is_comp(father_page));
 
-	if (adjust) {
-		nth_rec = page_rec_get_n_recs_before(btr_cur_get_rec(cursor));
-	}
-
 	/* Decide the page to which we try to merge and which will inherit
 	the locks */
 
@@ -2132,8 +2121,9 @@ btr_compress(
 	} else {
 		/* The page is the only one on the level, lift the records
 		to the father */
-		merge_page = btr_lift_page_up(index, page, mtr);
+		btr_lift_page_up(index, page, mtr);
-		goto func_exit;
+
+		return;
 	}
 
 	n_recs = page_get_n_recs(page);
@@ -2209,10 +2199,6 @@ btr_compress(
 					 index, mtr);
 
 		lock_update_merge_left(merge_page, orig_pred, page);
-
-		if (adjust) {
-			nth_rec += page_rec_get_n_recs_before(orig_pred);
-		}
 	} else {
 		orig_succ = page_rec_get_next(
 			page_get_infimum_rec(merge_page));
@@ -2233,12 +2219,6 @@ btr_compress(
 	btr_page_free(index, page, mtr);
 
 	ut_ad(btr_check_node_ptr(index, merge_page, mtr));
-
-func_exit:
-	if (adjust) {
-		btr_cur_position(index, page_rec_get_nth(merge_page, nth_rec),
-				 cursor);
-	}
 }
 
 /*****************************************************************

storage/innobase/btr/btr0cur.c

@@ -1790,9 +1790,7 @@ btr_cur_pessimistic_update(
 				/* out: DB_SUCCESS or error code */
 	ulint		flags,	/* in: undo logging, locking, and rollback
 				flags */
-	btr_cur_t*	cursor,	/* in/out: cursor on the record to update;
+	btr_cur_t*	cursor,	/* in: cursor on the record to update */
-				cursor may become invalid if *big_rec == NULL
-				|| !(flags & BTR_KEEP_POS_FLAG) */
 	big_rec_t**	big_rec,/* out: big rec vector whose fields have to
 				be stored externally by the caller, or NULL */
 	upd_t*		update,	/* in: update vector; this is allowed also
@@ -1927,10 +1925,6 @@ btr_cur_pessimistic_update(
 			err = DB_TOO_BIG_RECORD;
 			goto return_after_reservations;
 		}
-
-		ut_ad(index->type & DICT_CLUSTERED);
-		ut_ad(btr_page_get_level(page, mtr) == 0);
-		ut_ad(flags & BTR_KEEP_POS_FLAG);
 	}
 
 	page_cursor = btr_cur_get_page_cur(cursor);
@@ -1957,8 +1951,6 @@ btr_cur_pessimistic_update(
 	ut_a(rec || optim_err != DB_UNDERFLOW);
 
 	if (rec) {
-		page_cursor->rec = rec;
-
 		lock_rec_restore_from_page_infimum(rec, page);
 		rec_set_field_extern_bits(rec, index,
 					  ext_vect, n_ext_vect, mtr);
@@ -1972,30 +1964,12 @@ btr_cur_pessimistic_update(
 			btr_cur_unmark_extern_fields(rec, mtr, offsets);
 		}
 
-		btr_cur_compress_if_useful(
+		btr_cur_compress_if_useful(cursor, mtr);
-			cursor,
-			big_rec_vec != NULL && (flags & BTR_KEEP_POS_FLAG),
-			mtr);
 
 		err = DB_SUCCESS;
 		goto return_after_reservations;
 	}
 
-	if (big_rec_vec) {
-		ut_ad(index->type & DICT_CLUSTERED);
-		ut_ad(btr_page_get_level(page, mtr) == 0);
-		ut_ad(flags & BTR_KEEP_POS_FLAG);
-
-		/* btr_page_split_and_insert() in
-		btr_cur_pessimistic_insert() invokes
-		mtr_memo_release(mtr, index->lock, MTR_MEMO_X_LOCK).
-		We must keep the index->lock when we created a
-		big_rec, so that row_upd_clust_rec() can store the
-		big_rec in the same mini-transaction. */
-
-		mtr_x_lock(dict_index_get_lock(index), mtr);
-	}
-
 	if (page_cur_is_before_first(page_cursor)) {
 		/* The record to be updated was positioned as the first user
 		record on its page */
@@ -2016,7 +1990,6 @@ btr_cur_pessimistic_update(
 	ut_a(rec);
 	ut_a(err == DB_SUCCESS);
 	ut_a(dummy_big_rec == NULL);
-	page_cursor->rec = rec;
 
 	rec_set_field_extern_bits(rec, index, ext_vect, n_ext_vect, mtr);
 	offsets = rec_get_offsets(rec, index, offsets, ULINT_UNDEFINED, &heap);
@@ -2051,43 +2024,6 @@ return_after_reservations:
 	return(err);
 }
 
-/*****************************************************************
-Commits and restarts a mini-transaction so that it will retain an
-x-lock on index->lock and the cursor page. */
-
-void
-btr_cur_mtr_commit_and_start(
-/*=========================*/
-	btr_cur_t*	cursor,	/* in: cursor */
-	mtr_t*		mtr)	/* in/out: mini-transaction */
-{
-	buf_block_t*	block;
-
-	block = buf_block_align(btr_cur_get_rec(cursor));
-
-	ut_ad(mtr_memo_contains(mtr, dict_index_get_lock(cursor->index),
-				MTR_MEMO_X_LOCK));
-	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
-	/* Keep the locks across the mtr_commit(mtr). */
-	rw_lock_x_lock(dict_index_get_lock(cursor->index));
-	rw_lock_x_lock(&block->lock);
-	mutex_enter(&block->mutex);
-#ifdef UNIV_SYNC_DEBUG
-	buf_block_buf_fix_inc_debug(block, __FILE__, __LINE__);
-#else
-	buf_block_buf_fix_inc(block);
-#endif
-	mutex_exit(&block->mutex);
-	/* Write out the redo log. */
-	mtr_commit(mtr);
-	mtr_start(mtr);
-	/* Reassociate the locks with the mini-transaction.
-	They will be released on mtr_commit(mtr). */
-	mtr_memo_push(mtr, dict_index_get_lock(cursor->index),
-		      MTR_MEMO_X_LOCK);
-	mtr_memo_push(mtr, block, MTR_MEMO_PAGE_X_FIX);
-}
-
 /*==================== B-TREE DELETE MARK AND UNMARK ===============*/
 
 /********************************************************************
@@ -2455,6 +2391,30 @@ btr_cur_del_unmark_for_ibuf(
 
 /*==================== B-TREE RECORD REMOVE =========================*/
 
+/*****************************************************************
+Tries to compress a page of the tree on the leaf level. It is assumed
+that mtr holds an x-latch on the tree and on the cursor page. To avoid
+deadlocks, mtr must also own x-latches to brothers of page, if those
+brothers exist. NOTE: it is assumed that the caller has reserved enough
+free extents so that the compression will always succeed if done! */
+
+void
+btr_cur_compress(
+/*=============*/
+	btr_cur_t*	cursor,	/* in: cursor on the page to compress;
+				cursor does not stay valid */
+	mtr_t*		mtr)	/* in: mtr */
+{
+	ut_ad(mtr_memo_contains(mtr,
+				dict_index_get_lock(btr_cur_get_index(cursor)),
+				MTR_MEMO_X_LOCK));
+	ut_ad(mtr_memo_contains(mtr, buf_block_align(btr_cur_get_rec(cursor)),
+				MTR_MEMO_PAGE_X_FIX));
+	ut_ad(btr_page_get_level(btr_cur_get_page(cursor), mtr) == 0);
+
+	btr_compress(cursor, mtr);
+}
+
 /*****************************************************************
 Tries to compress a page of the tree if it seems useful. It is assumed
 that mtr holds an x-latch on the tree and on the cursor page. To avoid
@@ -2466,12 +2426,10 @@ ibool
 btr_cur_compress_if_useful(
 /*=======================*/
 				/* out: TRUE if compression occurred */
-	btr_cur_t*	cursor,	/* in/out: cursor on the page to compress;
+	btr_cur_t*	cursor,	/* in: cursor on the page to compress;
-				cursor does not stay valid if !adjust and
+				cursor does not stay valid if compression
-				compression occurs */
+				occurs */
-	ibool		adjust,	/* in: TRUE if should adjust the
+	mtr_t*		mtr)	/* in: mtr */
-				cursor position even if compression occurs */
-	mtr_t*		mtr)	/* in/out: mini-transaction */
 {
 	ut_ad(mtr_memo_contains(mtr,
 				dict_index_get_lock(btr_cur_get_index(cursor)),
@@ -2481,7 +2439,7 @@ btr_cur_compress_if_useful(
 
 	if (btr_cur_compress_recommendation(cursor, mtr)) {
 
-		btr_compress(cursor, adjust, mtr);
+		btr_compress(cursor, mtr);
 
 		return(TRUE);
 	}
@@ -2694,7 +2652,7 @@ return_after_reservations:
 	mem_heap_free(heap);
 
 	if (ret == FALSE) {
-		ret = btr_cur_compress_if_useful(cursor, FALSE, mtr);
+		ret = btr_cur_compress_if_useful(cursor, mtr);
 	}
 
 	if (n_extents > 0) {

storage/innobase/include/btr0btr.h

@@ -312,13 +312,11 @@ enough free extents so that the compression will always succeed if done! */
 void
 btr_compress(
 /*=========*/
-	btr_cur_t*	cursor,	/* in/out: cursor on the page to merge
+	btr_cur_t*	cursor,	/* in: cursor on the page to merge or lift;
-				or lift; the page must not be empty:
+				the page must not be empty: in record delete
-				when deleting records, use btr_discard_page()
+				use btr_discard_page if the page would become
-				if the page would become empty */
+				empty */
-	ibool		adjust,	/* in: TRUE if should adjust the
+	mtr_t*		mtr);	/* in: mtr */
-				cursor position even if compression occurs */
-	mtr_t*		mtr);	/* in/out: mini-transaction */
 /*****************************************************************
 Discards a page from a B-tree. This is used to remove the last record from
 a B-tree page: the whole page must be removed at the same time. This cannot

storage/innobase/include/btr0cur.h

@@ -23,9 +23,6 @@ Created 10/16/1994 Heikki Tuuri
 #define BTR_NO_LOCKING_FLAG	2	/* do no record lock checking */
 #define BTR_KEEP_SYS_FLAG	4	/* sys fields will be found from the
 					update vector or inserted entry */
-#define BTR_KEEP_POS_FLAG	8	/* btr_cur_pessimistic_update()
-					must keep cursor position when
-					moving columns to big_rec */
 
 #define BTR_CUR_ADAPT
 #define BTR_CUR_HASH_ADAPT
@@ -240,9 +237,7 @@ btr_cur_pessimistic_update(
 				/* out: DB_SUCCESS or error code */
 	ulint		flags,	/* in: undo logging, locking, and rollback
 				flags */
-	btr_cur_t*	cursor,	/* in/out: cursor on the record to update;
+	btr_cur_t*	cursor,	/* in: cursor on the record to update */
-				cursor may become invalid if *big_rec == NULL
-				|| !(flags & BTR_KEEP_POS_FLAG) */
 	big_rec_t**	big_rec,/* out: big rec vector whose fields have to
 				be stored externally by the caller, or NULL */
 	upd_t*		update,	/* in: update vector; this is allowed also
@@ -252,15 +247,6 @@ btr_cur_pessimistic_update(
 				updates */
 	que_thr_t*	thr,	/* in: query thread */
 	mtr_t*		mtr);	/* in: mtr */
-/*****************************************************************
-Commits and restarts a mini-transaction so that it will retain an
-x-lock on index->lock and the cursor page. */
-
-void
-btr_cur_mtr_commit_and_start(
-/*=========================*/
-	btr_cur_t*	cursor,	/* in: cursor */
-	mtr_t*		mtr);	/* in/out: mini-transaction */
 /***************************************************************
 Marks a clustered index record deleted. Writes an undo log record to
 undo log on this delete marking. Writes in the trx id field the id
@@ -300,6 +286,19 @@ btr_cur_del_unmark_for_ibuf(
 	rec_t*		rec,	/* in: record to delete unmark */
 	mtr_t*		mtr);	/* in: mtr */
 /*****************************************************************
+Tries to compress a page of the tree on the leaf level. It is assumed
+that mtr holds an x-latch on the tree and on the cursor page. To avoid
+deadlocks, mtr must also own x-latches to brothers of page, if those
+brothers exist. NOTE: it is assumed that the caller has reserved enough
+free extents so that the compression will always succeed if done! */
+
+void
+btr_cur_compress(
+/*=============*/
+	btr_cur_t*	cursor,	/* in: cursor on the page to compress;
+				cursor does not stay valid */
+	mtr_t*		mtr);	/* in: mtr */
+/*****************************************************************
 Tries to compress a page of the tree if it seems useful. It is assumed
 that mtr holds an x-latch on the tree and on the cursor page. To avoid
 deadlocks, mtr must also own x-latches to brothers of page, if those
@@ -310,12 +309,10 @@ ibool
 btr_cur_compress_if_useful(
 /*=======================*/
 				/* out: TRUE if compression occurred */
-	btr_cur_t*	cursor,	/* in/out: cursor on the page to compress;
+	btr_cur_t*	cursor,	/* in: cursor on the page to compress;
-				cursor does not stay valid if !adjust and
+				cursor does not stay valid if compression
-				compression occurs */
+				occurs */
-	ibool		adjust,	/* in: TRUE if should adjust the
+	mtr_t*		mtr);	/* in: mtr */
-				cursor position even if compression occurs */
-	mtr_t*		mtr);	/* in/out: mini-transaction */
 /***********************************************************
 Removes the record on which the tree cursor is positioned. It is assumed
 that the mtr has an x-latch on the page where the cursor is positioned,

storage/innobase/include/buf0buf.h

@@ -654,25 +654,6 @@ buf_page_address_fold(
 			/* out: the folded value */
 	ulint	space,	/* in: space id */
 	ulint	offset);/* in: offset of the page within space */
-#ifdef UNIV_SYNC_DEBUG
-/***********************************************************************
-Increments the bufferfix count. */
-UNIV_INLINE
-void
-buf_block_buf_fix_inc_debug(
-/*========================*/
-	buf_block_t*	block,	/* in: block to bufferfix */
-	const char*	file __attribute__ ((unused)),	/* in: file name */
-	ulint		line __attribute__ ((unused)));	/* in: line */
-#else /* UNIV_SYNC_DEBUG */
-/***********************************************************************
-Increments the bufferfix count. */
-UNIV_INLINE
-void
-buf_block_buf_fix_inc(
-/*==================*/
-	buf_block_t*	block);	/* in: block to bufferfix */
-#endif /* UNIV_SYNC_DEBUG */
 /**********************************************************************
 Returns the control block of a file page, NULL if not found. */
 UNIV_INLINE

storage/innobase/include/page0page.h

@@ -234,21 +234,10 @@ page_get_supremum_rec(
 /*==================*/
 			/* out: the last record in record list */
 	page_t*	page);	/* in: page which must have record(s) */
-/************************************************************//**
+/****************************************************************
-Returns the nth record of the record list.
+Returns the middle record of record list. If there are an even number
-This is the inverse function of page_rec_get_n_recs_before(). */
+of records in the list, returns the first record of upper half-list. */
 
-rec_t*
-page_rec_get_nth(
-/*=============*/
-			/* out: nth record */
-	page_t*	page,	/* in: page */
-	ulint	nth);	/* in: nth record */
-/*****************************************************************
-Returns the middle record of the records on the page. If there is an
-even number of records in the list, returns the first record of the
-upper half-list. */
-UNIV_INLINE
 rec_t*
 page_get_middle_rec(
 /*================*/
@@ -291,8 +280,7 @@ page_get_n_recs(
 	page_t*	page);	/* in: index page */
 /*******************************************************************
 Returns the number of records before the given record in chain.
-The number includes infimum and supremum records.
+The number includes infimum and supremum records. */
-This is the inverse function of page_rec_get_nth(). */
 
 ulint
 page_rec_get_n_recs_before(

storage/innobase/include/page0page.ic

@@ -340,22 +340,6 @@ page_rec_is_infimum(
 	return(page_rec_is_infimum_low(page_offset(rec)));
 }
 
-/*****************************************************************
-Returns the middle record of the records on the page. If there is an
-even number of records in the list, returns the first record of the
-upper half-list. */
-UNIV_INLINE
-rec_t*
-page_get_middle_rec(
-/*================*/
-			/* out: middle record */
-	page_t*	page)	/* in: page */
-{
-	ulint	middle = (page_get_n_recs(page) + 2) / 2;
-
-	return(page_rec_get_nth(page, middle));
-}
-
 /*****************************************************************
 Compares a data tuple to a physical record. Differs from the function
 cmp_dtuple_rec_with_match in the way that the record must reside on an

storage/innobase/page/page0page.c

@@ -1194,42 +1194,49 @@ page_dir_balance_slot(
 }
 
 /****************************************************************
-Returns the nth record of the record list. */
+Returns the middle record of the record list. If there are an even number
+of records in the list, returns the first record of the upper half-list. */
 
 rec_t*
-page_rec_get_nth(
+page_get_middle_rec(
-/*=============*/
+/*================*/
-			/* out: nth record */
+			/* out: middle record */
-	page_t*	page,	/* in: page */
+	page_t*	page)	/* in: page */
-	ulint	nth)	/* in: nth record */
 {
 	page_dir_slot_t*	slot;
+	ulint			middle;
 	ulint			i;
 	ulint			n_owned;
+	ulint			count;
 	rec_t*			rec;
 
-	ut_ad(nth < UNIV_PAGE_SIZE / (REC_N_NEW_EXTRA_BYTES + 1));
+	/* This many records we must leave behind */
+	middle = (page_get_n_recs(page) + 2) / 2;
+
+	count = 0;
 
 	for (i = 0;; i++) {
 
 		slot = page_dir_get_nth_slot(page, i);
 		n_owned = page_dir_slot_get_n_owned(slot);
 
-		if (n_owned > nth) {
+		if (count + n_owned > middle) {
 			break;
 		} else {
-			nth -= n_owned;
+			count += n_owned;
 		}
 	}
 
 	ut_ad(i > 0);
 	slot = page_dir_get_nth_slot(page, i - 1);
 	rec = page_dir_slot_get_rec(slot);
+	rec = page_rec_get_next(rec);
 
-	do {
+	/* There are now count records behind rec */
+
+	for (i = 0; i < middle - count; i++) {
 		rec = page_rec_get_next(rec);
-		ut_ad(rec);
+	}
-	} while (nth--);
 
 	return(rec);
 }

storage/innobase/row/row0ins.c

@@ -259,7 +259,6 @@ row_ins_sec_index_entry_by_modify(
 		err = btr_cur_pessimistic_update(BTR_KEEP_SYS_FLAG, cursor,
 						 &dummy_big_rec, update,
 						 0, thr, mtr);
-		ut_a(!dummy_big_rec);
 	}
 func_exit:
 	mem_heap_free(heap);
@@ -330,9 +329,8 @@ row_ins_clust_index_entry_by_modify(
 
 			goto func_exit;
 		}
-		err = btr_cur_pessimistic_update(
+		err = btr_cur_pessimistic_update(0, cursor, big_rec, update,
-			BTR_KEEP_POS_FLAG, cursor, big_rec, update,
+						 0, thr, mtr);
-			0, thr, mtr);
 	}
 func_exit:
 	mem_heap_free(heap);
@@ -2086,41 +2084,6 @@ row_ins_index_entry_low(
 			err = row_ins_clust_index_entry_by_modify(
 				mode, &cursor, &big_rec, entry,
 				ext_vec, n_ext_vec, thr, &mtr);
-
-			if (big_rec) {
-				ut_a(err == DB_SUCCESS);
-				/* Write out the externally stored
-				columns while still x-latching
-				index->lock and block->lock. We have
-				to mtr_commit(mtr) first, so that the
-				redo log will be written in the
-				correct order. Otherwise, we would run
-				into trouble on crash recovery if mtr
-				freed B-tree pages on which some of
-				the big_rec fields will be written. */
-				btr_cur_mtr_commit_and_start(&cursor, &mtr);
-
-				rec = btr_cur_get_rec(&cursor);
-				offsets = rec_get_offsets(rec, index, offsets,
-							  ULINT_UNDEFINED,
-							  &heap);
-
-				err = btr_store_big_rec_extern_fields(
-					index, rec, offsets, big_rec, &mtr);
-				/* If writing big_rec fails (for
-				example, because of DB_OUT_OF_FILE_SPACE),
-				the record will be corrupted. Even if
-				we did not update any externally
-				stored columns, our update could cause
-				the record to grow so that a
-				non-updated column was selected for
-				external storage. This non-update
-				would not have been written to the
-				undo log, and thus the record cannot
-				be rolled back. */
-				ut_a(err == DB_SUCCESS);
-				goto stored_big_rec;
-			}
 		} else {
 			err = row_ins_sec_index_entry_by_modify(
 				mode, &cursor, entry, thr, &mtr);
@@ -2157,6 +2120,7 @@ function_exit:
 	mtr_commit(&mtr);
 
 	if (big_rec) {
+		rec_t*		rec;
 		mtr_start(&mtr);
 
 		btr_cur_search_to_nth_level(index, 0, entry, PAGE_CUR_LE,
@@ -2167,7 +2131,7 @@ function_exit:
 
 		err = btr_store_big_rec_extern_fields(index, rec,
 						      offsets, big_rec, &mtr);
-stored_big_rec:
+
 		if (modify) {
 			dtuple_big_rec_free(big_rec);
 		} else {

storage/innobase/row/row0umod.c

@@ -119,7 +119,6 @@ row_undo_mod_clust_low(
 			| BTR_KEEP_SYS_FLAG,
 			btr_cur, &dummy_big_rec, node->update,
 			node->cmpl_info, thr, mtr);
-		ut_ad(!dummy_big_rec);
 	}
 
 	return(err);
@@ -472,7 +471,6 @@ row_undo_mod_del_unmark_sec_and_undo_update(
 				BTR_KEEP_SYS_FLAG | BTR_NO_LOCKING_FLAG,
 				btr_cur, &dummy_big_rec,
 				update, 0, thr, &mtr);
-			ut_ad(!dummy_big_rec);
 		}
 
 		mem_heap_free(heap);

storage/innobase/row/row0upd.c

@@ -1580,48 +1580,32 @@ row_upd_clust_rec(
 	ut_ad(!rec_get_deleted_flag(btr_pcur_get_rec(pcur),
 				    dict_table_is_comp(index->table)));
 
-	err = btr_cur_pessimistic_update(
+	err = btr_cur_pessimistic_update(BTR_NO_LOCKING_FLAG, btr_cur,
-		BTR_NO_LOCKING_FLAG | BTR_KEEP_POS_FLAG, btr_cur,
+					 &big_rec, node->update,
-		&big_rec, node->update, node->cmpl_info, thr, mtr);
+					 node->cmpl_info, thr, mtr);
+	mtr_commit(mtr);
 
-	if (big_rec) {
+	if (err == DB_SUCCESS && big_rec) {
 		mem_heap_t*	heap		= NULL;
 		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
 		rec_t*		rec;
 		*offsets_ = (sizeof offsets_) / sizeof *offsets_;
 
-		ut_a(err == DB_SUCCESS);
+		mtr_start(mtr);
-		/* Write out the externally stored columns while still
-		x-latching index->lock and block->lock. We have to
-		mtr_commit(mtr) first, so that the redo log will be
-		written in the correct order. Otherwise, we would run
-		into trouble on crash recovery if mtr freed B-tree
-		pages on which some of the big_rec fields will be
-		written. */
-		btr_cur_mtr_commit_and_start(btr_cur, mtr);
 
+		ut_a(btr_pcur_restore_position(BTR_MODIFY_TREE, pcur, mtr));
 		rec = btr_cur_get_rec(btr_cur);
 		err = btr_store_big_rec_extern_fields(
 			index, rec,
 			rec_get_offsets(rec, index, offsets_,
 					ULINT_UNDEFINED, &heap),
-			big_rec, mtr);
+			 big_rec, mtr);
 		if (UNIV_LIKELY_NULL(heap)) {
 			mem_heap_free(heap);
 		}
-		/* If writing big_rec fails (for example, because of
+		mtr_commit(mtr);
-		DB_OUT_OF_FILE_SPACE), the record will be corrupted.
-		Even if we did not update any externally stored
-		columns, our update could cause the record to grow so
-		that a non-updated column was selected for external
-		storage. This non-update would not have been written
-		to the undo log, and thus the record cannot be rolled
-		back. */
-		ut_a(err == DB_SUCCESS);
 	}
 
-	mtr_commit(mtr);
-
 	if (big_rec) {
 		dtuple_big_rec_free(big_rec);
 	}

storage/innodb_plugin/ChangeLog

@@ -101,16 +101,6 @@
 	* page/page0zip.c, rem/rem0rec.c:
 	Fix Bug#61191 question about page_zip_available()
 
-2011-06-16	The InnoDB Team
-
-	* btr/btr0btr.c, btr/btr0cur.c, include/btr0btr.h, include/btr0cur.h,
-	include/btr0cur.ic, include/buf0buf.h, include/buf0buf.ic,
-	include/page0cur.ic, include/page0page.h, include/page0page.ic,
-	include/sync0rw.ic, include/sync0sync.h, page/page0cur.c,
-	page/page0page.c, row/row0ins.c, row/row0upd.c,
-	sync/sync0rw.c, sync/sync0sync.c:
-	Fix Bug#12612184 Race condition after btr_cur_pessimistic_update()
-
 2011-06-09	The InnoDB Team
 	* btr/btr0cur.c, include/rem0rec.h, include/rem0rec.ic,
 	* row/row0row.c, row/row0vers.c, trx/trx0rec.c:

storage/innodb_plugin/btr/btr0btr.c

@@ -3001,16 +3001,15 @@ btr_node_ptr_delete(
 	ut_a(err == DB_SUCCESS);
 
 	if (!compressed) {
-		btr_cur_compress_if_useful(&cursor, FALSE, mtr);
+		btr_cur_compress_if_useful(&cursor, mtr);
 	}
 }
 
 /*************************************************************//**
 If page is the only on its level, this function moves its records to the
-father page, thus reducing the tree height.
+father page, thus reducing the tree height. */
-@return father block */
 static
-buf_block_t*
+void
 btr_lift_page_up(
 /*=============*/
 	dict_index_t*	index,	/*!< in: index tree */
@@ -3127,8 +3126,6 @@ btr_lift_page_up(
 	}
 	ut_ad(page_validate(father_page, index));
 	ut_ad(btr_check_node_ptr(index, father_block, mtr));
-
-	return(father_block);
 }
 
 /*************************************************************//**
@@ -3145,13 +3142,11 @@ UNIV_INTERN
 ibool
 btr_compress(
 /*=========*/
-	btr_cur_t*	cursor,	/*!< in/out: cursor on the page to merge
+	btr_cur_t*	cursor,	/*!< in: cursor on the page to merge or lift;
-				or lift; the page must not be empty:
+				the page must not be empty: in record delete
-				when deleting records, use btr_discard_page()
+				use btr_discard_page if the page would become
-				if the page would become empty */
+				empty */
-	ibool		adjust,	/*!< in: TRUE if should adjust the
+	mtr_t*		mtr)	/*!< in: mtr */
-				cursor position even if compression occurs */
-	mtr_t*		mtr)	/*!< in/out: mini-transaction */
 {
 	dict_index_t*	index;
 	ulint		space;
@@ -3169,14 +3164,12 @@ btr_compress(
 	ulint*		offsets;
 	ulint		data_size;
 	ulint		n_recs;
-	ulint		nth_rec = 0; /* remove bogus warning */
 	ulint		max_ins_size;
 	ulint		max_ins_size_reorg;
 
 	block = btr_cur_get_block(cursor);
 	page = btr_cur_get_page(cursor);
 	index = btr_cur_get_index(cursor);
-
 	ut_a((ibool) !!page_is_comp(page) == dict_table_is_comp(index->table));
 
 	ut_ad(mtr_memo_contains(mtr, dict_index_get_lock(index),
@@ -3197,10 +3190,6 @@ btr_compress(
 	offsets = btr_page_get_father_block(NULL, heap, index, block, mtr,
 					    &father_cursor);
 
-	if (adjust) {
-		nth_rec = page_rec_get_n_recs_before(btr_cur_get_rec(cursor));
-	}
-
 	/* Decide the page to which we try to merge and which will inherit
 	the locks */
 
@@ -3227,9 +3216,9 @@ btr_compress(
 	} else {
 		/* The page is the only one on the level, lift the records
 		to the father */
-
+		btr_lift_page_up(index, block, mtr);
-		merge_block = btr_lift_page_up(index, block, mtr);
+		mem_heap_free(heap);
-		goto func_exit;
+		return(TRUE);
 	}
 
 	n_recs = page_get_n_recs(page);
@@ -3311,10 +3300,6 @@ err_exit:
 
 		btr_node_ptr_delete(index, block, mtr);
 		lock_update_merge_left(merge_block, orig_pred, block);
-
-		if (adjust) {
-			nth_rec += page_rec_get_n_recs_before(orig_pred);
-		}
 	} else {
 		rec_t*		orig_succ;
 #ifdef UNIV_BTR_DEBUG
@@ -3379,6 +3364,7 @@ err_exit:
 	}
 
 	btr_blob_dbg_remove(page, index, "btr_compress");
+	mem_heap_free(heap);
 
 	if (!dict_index_is_clust(index) && page_is_leaf(merge_page)) {
 		/* Update the free bits of the B-tree page in the
@@ -3430,16 +3416,6 @@ err_exit:
 	btr_page_free(index, block, mtr);
 
 	ut_ad(btr_check_node_ptr(index, merge_block, mtr));
-func_exit:
-	mem_heap_free(heap);
-
-	if (adjust) {
-		btr_cur_position(
-			index,
-			page_rec_get_nth(merge_block->frame, nth_rec),
-			merge_block, cursor);
-	}
-
 	return(TRUE);
 }
 

storage/innodb_plugin/btr/btr0cur.c

@@ -2099,9 +2099,7 @@ btr_cur_pessimistic_update(
 /*=======================*/
 	ulint		flags,	/*!< in: undo logging, locking, and rollback
 				flags */
-	btr_cur_t*	cursor,	/*!< in/out: cursor on the record to update;
+	btr_cur_t*	cursor,	/*!< in: cursor on the record to update */
-				cursor may become invalid if *big_rec == NULL
-				|| !(flags & BTR_KEEP_POS_FLAG) */
 	mem_heap_t**	heap,	/*!< in/out: pointer to memory heap, or NULL */
 	big_rec_t**	big_rec,/*!< out: big rec vector whose fields have to
 				be stored externally by the caller, or NULL */
@@ -2240,7 +2238,7 @@ btr_cur_pessimistic_update(
 	record to be inserted: we have to remember which fields were such */
 
 	ut_ad(!page_is_comp(page) || !rec_get_node_ptr_flag(rec));
-	ut_ad(rec_offs_validate(rec, index, offsets));
+	offsets = rec_get_offsets(rec, index, offsets, ULINT_UNDEFINED, heap);
 	n_ext += btr_push_update_extern_fields(new_entry, update, *heap);
 
 	if (UNIV_LIKELY_NULL(page_zip)) {
@@ -2263,10 +2261,6 @@ make_external:
 			err = DB_TOO_BIG_RECORD;
 			goto return_after_reservations;
 		}
-
-		ut_ad(page_is_leaf(page));
-		ut_ad(dict_index_is_clust(index));
-		ut_ad(flags & BTR_KEEP_POS_FLAG);
 	}
 
 	/* Store state of explicit locks on rec on the page infimum record,
@@ -2294,8 +2288,6 @@ make_external:
 	rec = btr_cur_insert_if_possible(cursor, new_entry, n_ext, mtr);
 
 	if (rec) {
-		page_cursor->rec = rec;
-
 		lock_rec_restore_from_page_infimum(btr_cur_get_block(cursor),
 						   rec, block);
 
@@ -2309,10 +2301,7 @@ make_external:
 						     rec, index, offsets, mtr);
 		}
 
-		btr_cur_compress_if_useful(
+		btr_cur_compress_if_useful(cursor, mtr);
-			cursor,
-			big_rec_vec != NULL && (flags & BTR_KEEP_POS_FLAG),
-			mtr);
 
 		if (page_zip && !dict_index_is_clust(index)
 		    && page_is_leaf(page)) {
@@ -2332,21 +2321,6 @@ make_external:
 		}
 	}
 
-	if (big_rec_vec) {
-		ut_ad(page_is_leaf(page));
-		ut_ad(dict_index_is_clust(index));
-		ut_ad(flags & BTR_KEEP_POS_FLAG);
-
-		/* btr_page_split_and_insert() in
-		btr_cur_pessimistic_insert() invokes
-		mtr_memo_release(mtr, index->lock, MTR_MEMO_X_LOCK).
-		We must keep the index->lock when we created a
-		big_rec, so that row_upd_clust_rec() can store the
-		big_rec in the same mini-transaction. */
-
-		mtr_x_lock(dict_index_get_lock(index), mtr);
-	}
-
 	/* Was the record to be updated positioned as the first user
 	record on its page? */
 	was_first = page_cur_is_before_first(page_cursor);
@@ -2362,7 +2336,6 @@ make_external:
 	ut_a(rec);
 	ut_a(err == DB_SUCCESS);
 	ut_a(dummy_big_rec == NULL);
-	page_cursor->rec = rec;
 
 	if (dict_index_is_sec_or_ibuf(index)) {
 		/* Update PAGE_MAX_TRX_ID in the index page header.
@@ -2421,39 +2394,6 @@ return_after_reservations:
 	return(err);
 }
 
-/**************************************************************//**
-Commits and restarts a mini-transaction so that it will retain an
-x-lock on index->lock and the cursor page. */
-UNIV_INTERN
-void
-btr_cur_mtr_commit_and_start(
-/*=========================*/
-	btr_cur_t*	cursor,	/*!< in: cursor */
-	mtr_t*		mtr)	/*!< in/out: mini-transaction */
-{
-	buf_block_t*	block;
-
-	block = btr_cur_get_block(cursor);
-
-	ut_ad(mtr_memo_contains(mtr, dict_index_get_lock(cursor->index),
-				MTR_MEMO_X_LOCK));
-	ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
-	/* Keep the locks across the mtr_commit(mtr). */
-	rw_lock_x_lock(dict_index_get_lock(cursor->index));
-	rw_lock_x_lock(&block->lock);
-	mutex_enter(&block->mutex);
-	buf_block_buf_fix_inc(block, __FILE__, __LINE__);
-	mutex_exit(&block->mutex);
-	/* Write out the redo log. */
-	mtr_commit(mtr);
-	mtr_start(mtr);
-	/* Reassociate the locks with the mini-transaction.
-	They will be released on mtr_commit(mtr). */
-	mtr_memo_push(mtr, dict_index_get_lock(cursor->index),
-		      MTR_MEMO_X_LOCK);
-	mtr_memo_push(mtr, block, MTR_MEMO_PAGE_X_FIX);
-}
-
 /*==================== B-TREE DELETE MARK AND UNMARK ===============*/
 
 /****************************************************************//**
@@ -2829,12 +2769,10 @@ UNIV_INTERN
 ibool
 btr_cur_compress_if_useful(
 /*=======================*/
-	btr_cur_t*	cursor,	/*!< in/out: cursor on the page to compress;
+	btr_cur_t*	cursor,	/*!< in: cursor on the page to compress;
-				cursor does not stay valid if !adjust and
+				cursor does not stay valid if compression
-				compression occurs */
+				occurs */
-	ibool		adjust,	/*!< in: TRUE if should adjust the
+	mtr_t*		mtr)	/*!< in: mtr */
-				cursor position even if compression occurs */
-	mtr_t*		mtr)	/*!< in/out: mini-transaction */
 {
 	ut_ad(mtr_memo_contains(mtr,
 				dict_index_get_lock(btr_cur_get_index(cursor)),
@@ -2843,7 +2781,7 @@ btr_cur_compress_if_useful(
 				MTR_MEMO_PAGE_X_FIX));
 
 	return(btr_cur_compress_recommendation(cursor, mtr)
-	       && btr_compress(cursor, adjust, mtr));
+	       && btr_compress(cursor, mtr));
 }
 
 /*******************************************************//**
@@ -3085,7 +3023,7 @@ return_after_reservations:
 	mem_heap_free(heap);
 
 	if (ret == FALSE) {
-		ret = btr_cur_compress_if_useful(cursor, FALSE, mtr);
+		ret = btr_cur_compress_if_useful(cursor, mtr);
 	}
 
 	if (n_extents > 0) {

storage/innodb_plugin/include/btr0btr.h

@@ -488,14 +488,11 @@ UNIV_INTERN
 ibool
 btr_compress(
 /*=========*/
-	btr_cur_t*	cursor,	/*!< in/out: cursor on the page to merge
+	btr_cur_t*	cursor,	/*!< in: cursor on the page to merge or lift;
-				or lift; the page must not be empty:
+				the page must not be empty: in record delete
-				when deleting records, use btr_discard_page()
+				use btr_discard_page if the page would become
-				if the page would become empty */
+				empty */
-	ibool		adjust,	/*!< in: TRUE if should adjust the
+	mtr_t*		mtr);	/*!< in: mtr */
-				cursor position even if compression occurs */
-	mtr_t*		mtr)	/*!< in/out: mini-transaction */
-	__attribute__((nonnull));
 /*************************************************************//**
 Discards a page from a B-tree. This is used to remove the last record from
 a B-tree page: the whole page must be removed at the same time. This cannot

storage/innodb_plugin/include/btr0cur.h

@@ -36,9 +36,6 @@ Created 10/16/1994 Heikki Tuuri
 #define BTR_NO_LOCKING_FLAG	2	/* do no record lock checking */
 #define BTR_KEEP_SYS_FLAG	4	/* sys fields will be found from the
 					update vector or inserted entry */
-#define BTR_KEEP_POS_FLAG	8	/* btr_cur_pessimistic_update()
-					must keep cursor position when
-					moving columns to big_rec */
 
 #ifndef UNIV_HOTBACKUP
 #include "que0types.h"
@@ -312,9 +309,7 @@ btr_cur_pessimistic_update(
 /*=======================*/
 	ulint		flags,	/*!< in: undo logging, locking, and rollback
 				flags */
-	btr_cur_t*	cursor,	/*!< in/out: cursor on the record to update;
+	btr_cur_t*	cursor,	/*!< in: cursor on the record to update */
-				cursor may become invalid if *big_rec == NULL
-				|| !(flags & BTR_KEEP_POS_FLAG) */
 	mem_heap_t**	heap,	/*!< in/out: pointer to memory heap, or NULL */
 	big_rec_t**	big_rec,/*!< out: big rec vector whose fields have to
 				be stored externally by the caller, or NULL */
@@ -326,16 +321,6 @@ btr_cur_pessimistic_update(
 	que_thr_t*	thr,	/*!< in: query thread */
 	mtr_t*		mtr);	/*!< in: mtr; must be committed before
 				latching any further pages */
-/*****************************************************************
-Commits and restarts a mini-transaction so that it will retain an
-x-lock on index->lock and the cursor page. */
-UNIV_INTERN
-void
-btr_cur_mtr_commit_and_start(
-/*=========================*/
-	btr_cur_t*	cursor,	/*!< in: cursor */
-	mtr_t*		mtr)	/*!< in/out: mini-transaction */
-	__attribute__((nonnull));
 /***********************************************************//**
 Marks a clustered index record deleted. Writes an undo log record to
 undo log on this delete marking. Writes in the trx id field the id
@@ -391,13 +376,10 @@ UNIV_INTERN
 ibool
 btr_cur_compress_if_useful(
 /*=======================*/
-	btr_cur_t*	cursor,	/*!< in/out: cursor on the page to compress;
+	btr_cur_t*	cursor,	/*!< in: cursor on the page to compress;
 				cursor does not stay valid if compression
 				occurs */
-	ibool		adjust,	/*!< in: TRUE if should adjust the
+	mtr_t*		mtr);	/*!< in: mtr */
-				cursor position even if compression occurs */
-	mtr_t*		mtr)	/*!< in/out: mini-transaction */
-	__attribute__((nonnull));
 /*******************************************************//**
 Removes the record on which the tree cursor is positioned. It is assumed
 that the mtr has an x-latch on the page where the cursor is positioned,

storage/innodb_plugin/include/btr0cur.ic

@@ -139,7 +139,7 @@ btr_cur_compress_recommendation(
 	btr_cur_t*	cursor,	/*!< in: btr cursor */
 	mtr_t*		mtr)	/*!< in: mtr */
 {
-	const page_t*		page;
+	page_t*		page;
 
 	ut_ad(mtr_memo_contains(mtr, btr_cur_get_block(cursor),
 				MTR_MEMO_PAGE_X_FIX));

storage/innodb_plugin/include/buf0buf.h

@@ -468,31 +468,6 @@ buf_block_get_modify_clock(
 #else /* !UNIV_HOTBACKUP */
 # define buf_block_modify_clock_inc(block) ((void) 0)
 #endif /* !UNIV_HOTBACKUP */
-/*******************************************************************//**
-Increments the bufferfix count. */
-UNIV_INLINE
-void
-buf_block_buf_fix_inc_func(
-/*=======================*/
-#ifdef UNIV_SYNC_DEBUG
-	const char*	file,	/*!< in: file name */
-	ulint		line,	/*!< in: line */
-#endif /* UNIV_SYNC_DEBUG */
-	buf_block_t*	block)	/*!< in/out: block to bufferfix */
-	__attribute__((nonnull));
-#ifdef UNIV_SYNC_DEBUG
-/** Increments the bufferfix count.
-@param b	in/out: block to bufferfix
-@param f	in: file name where requested
-@param l	in: line number where requested */
-# define buf_block_buf_fix_inc(b,f,l) buf_block_buf_fix_inc_func(f,l,b)
-#else /* UNIV_SYNC_DEBUG */
-/** Increments the bufferfix count.
-@param b	in/out: block to bufferfix
-@param f	in: file name where requested
-@param l	in: line number where requested */
-# define buf_block_buf_fix_inc(b,f,l) buf_block_buf_fix_inc_func(b)
-#endif /* UNIV_SYNC_DEBUG */
 /********************************************************************//**
 Calculates a page checksum which is stored to the page when it is written
 to a file. Note that we must be careful to calculate the same value

storage/innodb_plugin/include/buf0buf.ic

@@ -916,6 +916,19 @@ buf_block_buf_fix_inc_func(
 
 	block->page.buf_fix_count++;
 }
+#ifdef UNIV_SYNC_DEBUG
+/** Increments the bufferfix count.
+@param b	in/out: block to bufferfix
+@param f	in: file name where requested
+@param l	in: line number where requested */
+# define buf_block_buf_fix_inc(b,f,l) buf_block_buf_fix_inc_func(f,l,b)
+#else /* UNIV_SYNC_DEBUG */
+/** Increments the bufferfix count.
+@param b	in/out: block to bufferfix
+@param f	in: file name where requested
+@param l	in: line number where requested */
+# define buf_block_buf_fix_inc(b,f,l) buf_block_buf_fix_inc_func(b)
+#endif /* UNIV_SYNC_DEBUG */
 
 /*******************************************************************//**
 Decrements the bufferfix count. */

storage/innodb_plugin/include/page0page.h

@@ -281,42 +281,16 @@ page_get_supremum_offset(
 	const page_t*	page);	/*!< in: page which must have record(s) */
 #define page_get_infimum_rec(page) ((page) + page_get_infimum_offset(page))
 #define page_get_supremum_rec(page) ((page) + page_get_supremum_offset(page))
-
 /************************************************************//**
-Returns the nth record of the record list.
+Returns the middle record of record list. If there are an even number
-This is the inverse function of page_rec_get_n_recs_before().
+of records in the list, returns the first record of upper half-list.
-@return	nth record */
-UNIV_INTERN
-const rec_t*
-page_rec_get_nth_const(
-/*===================*/
-	const page_t*	page,	/*!< in: page */
-	ulint		nth)	/*!< in: nth record */
-	__attribute__((nonnull, warn_unused_result));
-/************************************************************//**
-Returns the nth record of the record list.
-This is the inverse function of page_rec_get_n_recs_before().
-@return	nth record */
-UNIV_INLINE
-rec_t*
-page_rec_get_nth(
-/*=============*/
-	page_t*	page,	/*< in: page */
-	ulint	nth)	/*!< in: nth record */
-	__attribute__((nonnull, warn_unused_result));
-
-#ifndef UNIV_HOTBACKUP
-/************************************************************//**
-Returns the middle record of the records on the page. If there is an
-even number of records in the list, returns the first record of the
-upper half-list.
 @return	middle record */
-UNIV_INLINE
+UNIV_INTERN
 rec_t*
 page_get_middle_rec(
 /*================*/
-	page_t*	page)	/*!< in: page */
+	page_t*	page);	/*!< in: page */
-	__attribute__((nonnull, warn_unused_result));
+#ifndef UNIV_HOTBACKUP
 /*************************************************************//**
 Compares a data tuple to a physical record. Differs from the function
 cmp_dtuple_rec_with_match in the way that the record must reside on an
@@ -371,7 +345,6 @@ page_get_n_recs(
 /***************************************************************//**
 Returns the number of records before the given record in chain.
 The number includes infimum and supremum records.
-This is the inverse function of page_rec_get_nth().
 @return	number of records */
 UNIV_INTERN
 ulint

storage/innodb_plugin/include/page0page.ic

@@ -420,37 +420,7 @@ page_rec_is_infimum(
 	return(page_rec_is_infimum_low(page_offset(rec)));
 }
 
-/************************************************************//**
-Returns the nth record of the record list.
-This is the inverse function of page_rec_get_n_recs_before().
-@return	nth record */
-UNIV_INLINE
-rec_t*
-page_rec_get_nth(
-/*=============*/
-	page_t*	page,	/*!< in: page */
-	ulint	nth)	/*!< in: nth record */
-{
-	return((rec_t*) page_rec_get_nth_const(page, nth));
-}
-
 #ifndef UNIV_HOTBACKUP
-/************************************************************//**
-Returns the middle record of the records on the page. If there is an
-even number of records in the list, returns the first record of the
-upper half-list.
-@return	middle record */
-UNIV_INLINE
-rec_t*
-page_get_middle_rec(
-/*================*/
-	page_t*	page)	/*!< in: page */
-{
-	ulint	middle = (page_get_n_recs(page) + PAGE_HEAP_NO_USER_LOW) / 2;
-
-	return(page_rec_get_nth(page, middle));
-}
-
 /*************************************************************//**
 Compares a data tuple to a physical record. Differs from the function
 cmp_dtuple_rec_with_match in the way that the record must reside on an

storage/innodb_plugin/page/page0cur.c

@@ -1180,15 +1180,14 @@ page_cur_insert_rec_zip_reorg(
 	/* Before trying to reorganize the page,
 	store the number of preceding records on the page. */
 	pos = page_rec_get_n_recs_before(rec);
-	ut_ad(pos > 0);
 
 	if (page_zip_reorganize(block, index, mtr)) {
 		/* The page was reorganized: Find rec by seeking to pos,
 		and update *current_rec. */
-		if (pos > 1) {
+		rec = page + PAGE_NEW_INFIMUM;
-			rec = page_rec_get_nth(page, pos - 1);
+
-		} else {
+		while (--pos) {
-			rec = page + PAGE_NEW_INFIMUM;
+			rec = page + rec_get_next_offs(rec, TRUE);
 		}
 
 		*current_rec = rec;
@@ -1284,12 +1283,6 @@ page_cur_insert_rec_zip(
 			insert_rec = page_cur_insert_rec_zip_reorg(
 				current_rec, block, index, insert_rec,
 				page, page_zip, mtr);
-#ifdef UNIV_DEBUG
-			if (insert_rec) {
-				rec_offs_make_valid(
-					insert_rec, index, offsets);
-			}
-#endif /* UNIV_DEBUG */
 		}
 
 		return(insert_rec);

storage/innodb_plugin/page/page0page.c

@@ -1475,54 +1475,55 @@ page_dir_balance_slot(
 	}
 }
 
+#ifndef UNIV_HOTBACKUP
 /************************************************************//**
-Returns the nth record of the record list.
+Returns the middle record of the record list. If there are an even number
-This is the inverse function of page_rec_get_n_recs_before().
+of records in the list, returns the first record of the upper half-list.
-@return	nth record */
+@return	middle record */
 UNIV_INTERN
-const rec_t*
+rec_t*
-page_rec_get_nth_const(
+page_get_middle_rec(
-/*===================*/
+/*================*/
-	const page_t*	page,	/*!< in: page */
+	page_t*	page)	/*!< in: page */
-	ulint		nth)	/*!< in: nth record */
 {
-	const page_dir_slot_t*	slot;
+	page_dir_slot_t*	slot;
+	ulint			middle;
 	ulint			i;
 	ulint			n_owned;
-	const rec_t*		rec;
+	ulint			count;
+	rec_t*			rec;
 
-	ut_ad(nth < UNIV_PAGE_SIZE / (REC_N_NEW_EXTRA_BYTES + 1));
+	/* This many records we must leave behind */
+	middle = (page_get_n_recs(page) + PAGE_HEAP_NO_USER_LOW) / 2;
+
+	count = 0;
 
 	for (i = 0;; i++) {
 
 		slot = page_dir_get_nth_slot(page, i);
 		n_owned = page_dir_slot_get_n_owned(slot);
 
-		if (n_owned > nth) {
+		if (count + n_owned > middle) {
 			break;
 		} else {
-			nth -= n_owned;
+			count += n_owned;
 		}
 	}
 
 	ut_ad(i > 0);
 	slot = page_dir_get_nth_slot(page, i - 1);
-	rec = page_dir_slot_get_rec(slot);
+	rec = (rec_t*) page_dir_slot_get_rec(slot);
+	rec = page_rec_get_next(rec);
 
-	if (page_is_comp(page)) {
+	/* There are now count records behind rec */
-		do {
+
-			rec = page_rec_get_next_low(rec, TRUE);
+	for (i = 0; i < middle - count; i++) {
-			ut_ad(rec);
+		rec = page_rec_get_next(rec);
-		} while (nth--);
-	} else {
-		do {
-			rec = page_rec_get_next_low(rec, FALSE);
-			ut_ad(rec);
-		} while (nth--);
 	}
 
 	return(rec);
 }
+#endif /* !UNIV_HOTBACKUP */
 
 /***************************************************************//**
 Returns the number of records before the given record in chain.
@@ -1584,7 +1585,6 @@ page_rec_get_n_recs_before(
 	n--;
 
 	ut_ad(n >= 0);
-	ut_ad(n < UNIV_PAGE_SIZE / (REC_N_NEW_EXTRA_BYTES + 1));
 
 	return((ulint) n);
 }

storage/innodb_plugin/row/row0ins.c

@@ -345,9 +345,9 @@ row_ins_clust_index_entry_by_modify(
 			return(DB_LOCK_TABLE_FULL);
 
 		}
-		err = btr_cur_pessimistic_update(
+		err = btr_cur_pessimistic_update(0, cursor,
-			BTR_KEEP_POS_FLAG, cursor, heap, big_rec, update,
+						 heap, big_rec, update,
-			0, thr, mtr);
+						 0, thr, mtr);
 	}
 
 	return(err);
@@ -1989,7 +1989,6 @@ row_ins_index_entry_low(
 	ulint		modify = 0; /* remove warning */
 	rec_t*		insert_rec;
 	rec_t*		rec;
-	ulint*		offsets;
 	ulint		err;
 	ulint		n_unique;
 	big_rec_t*	big_rec			= NULL;
@@ -2093,42 +2092,6 @@ row_ins_index_entry_low(
 			err = row_ins_clust_index_entry_by_modify(
 				mode, &cursor, &heap, &big_rec, entry,
 				thr, &mtr);
-
-			if (big_rec) {
-				ut_a(err == DB_SUCCESS);
-				/* Write out the externally stored
-				columns while still x-latching
-				index->lock and block->lock. We have
-				to mtr_commit(mtr) first, so that the
-				redo log will be written in the
-				correct order. Otherwise, we would run
-				into trouble on crash recovery if mtr
-				freed B-tree pages on which some of
-				the big_rec fields will be written. */
-				btr_cur_mtr_commit_and_start(&cursor, &mtr);
-
-				rec = btr_cur_get_rec(&cursor);
-				offsets = rec_get_offsets(
-					rec, index, NULL,
-					ULINT_UNDEFINED, &heap);
-
-				err = btr_store_big_rec_extern_fields(
-					index, btr_cur_get_block(&cursor),
-					rec, offsets, &mtr, FALSE, big_rec);
-				/* If writing big_rec fails (for
-				example, because of DB_OUT_OF_FILE_SPACE),
-				the record will be corrupted. Even if
-				we did not update any externally
-				stored columns, our update could cause
-				the record to grow so that a
-				non-updated column was selected for
-				external storage. This non-update
-				would not have been written to the
-				undo log, and thus the record cannot
-				be rolled back. */
-				ut_a(err == DB_SUCCESS);
-				goto stored_big_rec;
-			}
 		} else {
 			ut_ad(!n_ext);
 			err = row_ins_sec_index_entry_by_modify(
@@ -2157,6 +2120,8 @@ function_exit:
 	mtr_commit(&mtr);
 
 	if (UNIV_LIKELY_NULL(big_rec)) {
+		rec_t*	rec;
+		ulint*	offsets;
 		mtr_start(&mtr);
 
 		btr_cur_search_to_nth_level(index, 0, entry, PAGE_CUR_LE,
@@ -2170,7 +2135,6 @@ function_exit:
 			index, btr_cur_get_block(&cursor),
 			rec, offsets, &mtr, FALSE, big_rec);
 
-stored_big_rec:
 		if (modify) {
 			dtuple_big_rec_free(big_rec);
 		} else {

storage/innodb_plugin/row/row0upd.c

@@ -1969,43 +1969,28 @@ row_upd_clust_rec(
 	ut_ad(!rec_get_deleted_flag(btr_pcur_get_rec(pcur),
 				    dict_table_is_comp(index->table)));
 
-	err = btr_cur_pessimistic_update(
+	err = btr_cur_pessimistic_update(BTR_NO_LOCKING_FLAG, btr_cur,
-		BTR_NO_LOCKING_FLAG | BTR_KEEP_POS_FLAG, btr_cur,
+					 &heap, &big_rec, node->update,
-		&heap, &big_rec, node->update, node->cmpl_info, thr, mtr);
+					 node->cmpl_info, thr, mtr);
-	if (big_rec) {
+	mtr_commit(mtr);
+
+	if (err == DB_SUCCESS && big_rec) {
 		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
 		rec_t*		rec;
 		rec_offs_init(offsets_);
 
-		ut_a(err == DB_SUCCESS);
+		mtr_start(mtr);
-		/* Write out the externally stored columns while still
-		x-latching index->lock and block->lock. We have to
-		mtr_commit(mtr) first, so that the redo log will be
-		written in the correct order. Otherwise, we would run
-		into trouble on crash recovery if mtr freed B-tree
-		pages on which some of the big_rec fields will be
-		written. */
-		btr_cur_mtr_commit_and_start(btr_cur, mtr);
 
+		ut_a(btr_pcur_restore_position(BTR_MODIFY_TREE, pcur, mtr));
 		rec = btr_cur_get_rec(btr_cur);
 		err = btr_store_big_rec_extern_fields(
 			index, btr_cur_get_block(btr_cur), rec,
 			rec_get_offsets(rec, index, offsets_,
 					ULINT_UNDEFINED, &heap),
 			mtr, TRUE, big_rec);
-		/* If writing big_rec fails (for example, because of
+		mtr_commit(mtr);
-		DB_OUT_OF_FILE_SPACE), the record will be corrupted.
-		Even if we did not update any externally stored
-		columns, our update could cause the record to grow so
-		that a non-updated column was selected for external
-		storage. This non-update would not have been written
-		to the undo log, and thus the record cannot be rolled
-		back. */
-		ut_a(err == DB_SUCCESS);
 	}
 
-	mtr_commit(mtr);
-
 	if (UNIV_LIKELY_NULL(heap)) {
 		mem_heap_free(heap);
 	}