Bug#12612184 Race condition after btr_cur_pessimistic_update()

btr_cur_compress_if_useful(), btr_compress(): Add the parameter ibool
adjust. If adjust=TRUE, adjust the cursor position after compressing
the page.

btr_lift_page_up(): Return a pointer to the father page.

BTR_KEEP_POS_FLAG: A new flag for btr_cur_pessimistic_update().

btr_cur_pessimistic_update(): If *big_rec != NULL and flags &
BTR_KEEP_POS_FLAG, keep the cursor positioned on the updated record.
Also, do not release the index tree x-lock if *big_rec != NULL.

btr_cur_mtr_commit_and_start(): Commits and restarts a
mini-transaction so that it will retain an x-lock on index->lock and
the page of the cursor. This is invoked when
btr_cur_pessimistic_update() returns *big_rec != NULL.

In all callers of btr_cur_pessimistic_update() that do not pass
BTR_KEEP_POS_FLAG, assert that *big_rec == NULL.

btr_cur_compress(): Unused function [in the built-in MySQL 5.1], remove.

page_rec_get_nth(): Return the nth record on the page (an inverse
function of page_rec_get_n_recs_before()). Refactored from
page_get_middle_rec().

page_get_middle_rec(): Invoke page_rec_get_nth().

page_cur_insert_rec_zip_reorg(): Make use of the page directory
shortcuts in page_rec_get_nth() instead of scanning the whole list of
records.

row_ins_clust_index_entry_by_modify(): Pass BTR_KEEP_POS_FLAG to
btr_cur_pessimistic_update().

row_ins_index_entry_low(): If row_ins_clust_index_entry_by_modify()
returns a big_rec, invoke btr_cur_mtr_commit_and_start() in order to
commit and start the mini-transaction without releasing the x-locks on
index->lock and the cursor page, and write the big_rec. Releasing the
page latch in mtr_commit() caused a race condition.

row_upd_clust_rec(): Pass BTR_KEEP_POS_FLAG to
btr_cur_pessimistic_update(). If it returns a big_rec, invoke
btr_cur_mtr_commit_and_start() in order to commit and start the
mini-transaction without releasing the x-locks on index->lock and the
cursor page, and write the big_rec. Releasing the page latch in
mtr_commit() caused a race condition.

sync_thread_add_level(): Add the parameter ibool relock. When TRUE,
bypass the latching order rules.

rw_lock_add_debug_info(): For nested X-lock requests, pass relock=TRUE
to sync_thread_add_level().

rb:678 approved by Jimmy Yang

storage/innobase/btr/btr0btr.c

@@ -1937,7 +1937,7 @@ btr_node_ptr_delete(
 	ut_a(err == DB_SUCCESS);
 
 	if (!compressed) {
-		btr_cur_compress_if_useful(&cursor, mtr);
+		btr_cur_compress_if_useful(&cursor, FALSE, mtr);
 	}
 }
 
@@ -1945,9 +1945,10 @@ 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
-void
+page_t*
 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
@@ -2023,6 +2024,8 @@ 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);
 }
 
 /*****************************************************************
@@ -2039,11 +2042,13 @@ enough free extents so that the compression will always succeed if done! */
 void
 btr_compress(
 /*=========*/
-	btr_cur_t*	cursor,	/* in: cursor on the page to merge or lift;
-				the page must not be empty: in record delete
-				use btr_discard_page if the page would become
-				empty */
-	mtr_t*		mtr)	/* in: mtr */
+	btr_cur_t*	cursor,	/* in/out: cursor on the page to merge
+				or lift; the page must not be empty:
+				when deleting records, use btr_discard_page()
+				if the page would become empty */
+	ibool		adjust,	/* in: TRUE if should adjust the
+				cursor position even if compression occurs */
+	mtr_t*		mtr)	/* in/out: mini-transaction */
 {
 	dict_index_t*	index;
 	ulint		space;
@@ -2058,6 +2063,7 @@ btr_compress(
 	rec_t*		node_ptr;
 	ulint		data_size;
 	ulint		n_recs;
+	ulint		nth_rec;
 	ulint		max_ins_size;
 	ulint		max_ins_size_reorg;
 	ulint		comp;
@@ -2065,6 +2071,7 @@ 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),
@@ -2086,6 +2093,10 @@ 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 */
 
@@ -2110,9 +2121,8 @@ btr_compress(
 	} else {
 		/* The page is the only one on the level, lift the records
 		to the father */
-		btr_lift_page_up(index, page, mtr);
-
-		return;
+		merge_page = btr_lift_page_up(index, page, mtr);
+		goto func_exit;
 	}
 
 	n_recs = page_get_n_recs(page);
@@ -2188,6 +2198,10 @@ 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));
@@ -2208,6 +2222,12 @@ 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

@@ -1791,7 +1791,9 @@ btr_cur_pessimistic_update(
 				/* out: DB_SUCCESS or error code */
 	ulint		flags,	/* in: undo logging, locking, and rollback
 				flags */
-	btr_cur_t*	cursor,	/* in: cursor on the record to update */
+	btr_cur_t*	cursor,	/* in/out: 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
@@ -1926,6 +1928,10 @@ 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);
@@ -1952,6 +1958,8 @@ 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);
@@ -1965,12 +1973,30 @@ btr_cur_pessimistic_update(
 			btr_cur_unmark_extern_fields(rec, mtr, offsets);
 		}
 
-		btr_cur_compress_if_useful(cursor, mtr);
+		btr_cur_compress_if_useful(
+			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 */
@@ -1991,6 +2017,7 @@ 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);
@@ -2025,6 +2052,43 @@ 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 ===============*/
 
 /********************************************************************
@@ -2392,30 +2456,6 @@ 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
@@ -2427,10 +2467,12 @@ ibool
 btr_cur_compress_if_useful(
 /*=======================*/
 				/* out: TRUE if compression occurred */
-	btr_cur_t*	cursor,	/* in: cursor on the page to compress;
-				cursor does not stay valid if compression
-				occurs */
-	mtr_t*		mtr)	/* in: mtr */
+	btr_cur_t*	cursor,	/* in/out: cursor on the page to compress;
+				cursor does not stay valid if !adjust and
+				compression occurs */
+	ibool		adjust,	/* in: TRUE if should adjust the
+				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)),
@@ -2440,7 +2482,7 @@ btr_cur_compress_if_useful(
 
 	if (btr_cur_compress_recommendation(cursor, mtr)) {
 
-		btr_compress(cursor, mtr);
+		btr_compress(cursor, adjust, mtr);
 
 		return(TRUE);
 	}
@@ -2653,7 +2695,7 @@ return_after_reservations:
 	mem_heap_free(heap);
 
 	if (ret == FALSE) {
-		ret = btr_cur_compress_if_useful(cursor, mtr);
+		ret = btr_cur_compress_if_useful(cursor, FALSE, mtr);
 	}
 
 	if (n_extents > 0) {

storage/innobase/include/btr0btr.h

@@ -312,11 +312,13 @@ enough free extents so that the compression will always succeed if done! */
 void
 btr_compress(
 /*=========*/
-	btr_cur_t*	cursor,	/* in: cursor on the page to merge or lift;
-				the page must not be empty: in record delete
-				use btr_discard_page if the page would become
-				empty */
-	mtr_t*		mtr);	/* in: mtr */
+	btr_cur_t*	cursor,	/* in/out: cursor on the page to merge
+				or lift; the page must not be empty:
+				when deleting records, use btr_discard_page()
+				if the page would become empty */
+	ibool		adjust,	/* in: TRUE if should adjust the
+				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

@@ -23,6 +23,9 @@ 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
@@ -237,7 +240,9 @@ btr_cur_pessimistic_update(
 				/* out: DB_SUCCESS or error code */
 	ulint		flags,	/* in: undo logging, locking, and rollback
 				flags */
-	btr_cur_t*	cursor,	/* in: cursor on the record to update */
+	btr_cur_t*	cursor,	/* in/out: 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
@@ -247,6 +252,15 @@ 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
@@ -286,19 +300,6 @@ 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
@@ -309,10 +310,12 @@ ibool
 btr_cur_compress_if_useful(
 /*=======================*/
 				/* out: TRUE if compression occurred */
-	btr_cur_t*	cursor,	/* in: cursor on the page to compress;
-				cursor does not stay valid if compression
-				occurs */
-	mtr_t*		mtr);	/* in: mtr */
+	btr_cur_t*	cursor,	/* in/out: cursor on the page to compress;
+				cursor does not stay valid if !adjust and
+				compression occurs */
+	ibool		adjust,	/* in: TRUE if should adjust the
+				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

@@ -682,6 +682,25 @@ 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/buf0buf.ic

@@ -660,6 +660,6 @@ buf_page_dbg_add_level(
 	ulint		level __attribute__((unused))) /* in: latching order
 				level */
 {
-	sync_thread_add_level(&(buf_block_align(frame)->lock), level);
+	sync_thread_add_level(&(buf_block_align(frame)->lock), level, FALSE);
 }
 #endif /* UNIV_SYNC_DEBUG */

storage/innobase/include/page0page.h

@@ -234,10 +234,21 @@ page_get_supremum_rec(
 /*==================*/
 			/* out: the last record in record list */
 	page_t*	page);	/* in: page which must have record(s) */
-/****************************************************************
-Returns the middle record of record list. If there are an even number
-of records in the list, returns the first record of upper half-list. */
+/************************************************************//**
+Returns the nth record of the record list.
+This is the inverse function of page_rec_get_n_recs_before(). */
 
+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(
 /*================*/
@@ -280,7 +291,8 @@ 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

@@ -340,6 +340,22 @@ 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/include/sync0sync.h

@@ -198,8 +198,9 @@ void
 sync_thread_add_level(
 /*==================*/
 	void*	latch,	/* in: pointer to a mutex or an rw-lock */
-	ulint	level);	/* in: level in the latching order; if
+	ulint	level,	/* in: level in the latching order; if
 			SYNC_LEVEL_VARYING, nothing is done */
+	ibool	relock);/* in: TRUE if re-entering an x-lock */
 /**********************************************************************
 Removes a latch from the thread level array if it is found there. */
 

storage/innobase/page/page0page.c

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

storage/innobase/row/row0ins.c

@@ -259,6 +259,7 @@ 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);
@@ -329,8 +330,9 @@ row_ins_clust_index_entry_by_modify(
 
 			goto func_exit;
 		}
-		err = btr_cur_pessimistic_update(0, cursor, big_rec, update,
-						 0, thr, mtr);
+		err = btr_cur_pessimistic_update(
+			BTR_KEEP_POS_FLAG, cursor, big_rec, update,
+			0, thr, mtr);
 	}
 func_exit:
 	mem_heap_free(heap);
@@ -2083,6 +2085,41 @@ 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);
@@ -2119,7 +2156,6 @@ 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,
@@ -2130,7 +2166,7 @@ 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

@@ -119,6 +119,7 @@ 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);
@@ -471,6 +472,7 @@ 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

@@ -1580,32 +1580,48 @@ 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(BTR_NO_LOCKING_FLAG, btr_cur,
-					 &big_rec, node->update,
-					 node->cmpl_info, thr, mtr);
-	mtr_commit(mtr);
+	err = btr_cur_pessimistic_update(
+		BTR_NO_LOCKING_FLAG | BTR_KEEP_POS_FLAG, btr_cur,
+		&big_rec, node->update, node->cmpl_info, thr, mtr);
 
-	if (err == DB_SUCCESS && big_rec) {
+	if (big_rec) {
 		mem_heap_t*	heap		= NULL;
 		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
 		rec_t*		rec;
 		*offsets_ = (sizeof offsets_) / sizeof *offsets_;
 
-		mtr_start(mtr);
+		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(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);
 		}
-		mtr_commit(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);
 	}
 
+	mtr_commit(mtr);
+
 	if (big_rec) {
 		dtuple_big_rec_free(big_rec);
 	}

storage/innobase/sync/sync0rw.c

@@ -663,7 +663,9 @@ rw_lock_add_debug_info(
 	rw_lock_debug_mutex_exit();
 
 	if ((pass == 0) && (lock_type != RW_LOCK_WAIT_EX)) {
-		sync_thread_add_level(lock, lock->level);
+		sync_thread_add_level(lock, lock->level,
+				      lock_type == RW_LOCK_EX
+				      && lock->writer_count > 1);
 	}
 }
 

storage/innobase/sync/sync0sync.c

@@ -641,7 +641,7 @@ mutex_set_debug_info(
 	ut_ad(mutex);
 	ut_ad(file_name);
 
-	sync_thread_add_level(mutex, mutex->level);
+	sync_thread_add_level(mutex, mutex->level, FALSE);
 
 	mutex->file_name = file_name;
 	mutex->line	 = line;
@@ -1011,8 +1011,9 @@ void
 sync_thread_add_level(
 /*==================*/
 	void*	latch,	/* in: pointer to a mutex or an rw-lock */
-	ulint	level)	/* in: level in the latching order; if
+	ulint	level,	/* in: level in the latching order; if
 			SYNC_LEVEL_VARYING, nothing is done */
+	ibool	relock)	/* in: TRUE if re-entering an x-lock */
 {
 	sync_level_t*	array;
 	sync_level_t*	slot;
@@ -1060,6 +1061,10 @@ sync_thread_add_level(
 
 	array = thread_slot->levels;
 
+	if (relock) {
+		goto levels_ok;
+	}
+
 	/* NOTE that there is a problem with _NODE and _LEAF levels: if the
 	B-tree height changes, then a leaf can change to an internal node
 	or the other way around. We do not know at present if this can cause
@@ -1209,6 +1214,7 @@ sync_thread_add_level(
 		ut_error;
 	}
 
+levels_ok:
 	for (i = 0; i < SYNC_THREAD_N_LEVELS; i++) {
 
 		slot = sync_thread_levels_get_nth(array, i);

storage/innodb_plugin/ChangeLog

@@ -1,3 +1,13 @@
+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

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1994, 2010, Innobase Oy. All Rights Reserved.
+Copyright (c) 1994, 2011, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -2272,7 +2272,7 @@ btr_attach_half_pages(
 /*==================*/
 	dict_index_t*	index,		/*!< in: the index tree */
 	buf_block_t*	block,		/*!< in/out: page to be split */
-	rec_t*		split_rec,	/*!< in: first record on upper
+	const rec_t*	split_rec,	/*!< in: first record on upper
 					half page */
 	buf_block_t*	new_block,	/*!< in/out: the new half page */
 	ulint		direction,	/*!< in: FSP_UP or FSP_DOWN */
@@ -2964,15 +2964,16 @@ btr_node_ptr_delete(
 	ut_a(err == DB_SUCCESS);
 
 	if (!compressed) {
-		btr_cur_compress_if_useful(&cursor, mtr);
+		btr_cur_compress_if_useful(&cursor, FALSE, 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
-void
+buf_block_t*
 btr_lift_page_up(
 /*=============*/
 	dict_index_t*	index,	/*!< in: index tree */
@@ -3089,6 +3090,8 @@ btr_lift_page_up(
 	}
 	ut_ad(page_validate(father_page, index));
 	ut_ad(btr_check_node_ptr(index, father_block, mtr));
+
+	return(father_block);
 }
 
 /*************************************************************//**
@@ -3105,11 +3108,13 @@ UNIV_INTERN
 ibool
 btr_compress(
 /*=========*/
-	btr_cur_t*	cursor,	/*!< in: cursor on the page to merge or lift;
-				the page must not be empty: in record delete
-				use btr_discard_page if the page would become
-				empty */
-	mtr_t*		mtr)	/*!< in: mtr */
+	btr_cur_t*	cursor,	/*!< in/out: cursor on the page to merge
+				or lift; the page must not be empty:
+				when deleting records, use btr_discard_page()
+				if the page would become empty */
+	ibool		adjust,	/*!< in: TRUE if should adjust the
+				cursor position even if compression occurs */
+	mtr_t*		mtr)	/*!< in/out: mini-transaction */
 {
 	dict_index_t*	index;
 	ulint		space;
@@ -3127,12 +3132,14 @@ btr_compress(
 	ulint*		offsets;
 	ulint		data_size;
 	ulint		n_recs;
+	ulint		nth_rec;
 	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),
@@ -3153,6 +3160,10 @@ 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 */
 
@@ -3179,9 +3190,9 @@ 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);
-		mem_heap_free(heap);
-		return(TRUE);
+
+		merge_block = btr_lift_page_up(index, block, mtr);
+		goto func_exit;
 	}
 
 	n_recs = page_get_n_recs(page);
@@ -3263,6 +3274,10 @@ 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
@@ -3327,7 +3342,6 @@ 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
@@ -3379,6 +3393,16 @@ 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

@@ -2088,7 +2088,9 @@ btr_cur_pessimistic_update(
 /*=======================*/
 	ulint		flags,	/*!< in: undo logging, locking, and rollback
 				flags */
-	btr_cur_t*	cursor,	/*!< in: cursor on the record to update */
+	btr_cur_t*	cursor,	/*!< in/out: 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 */
@@ -2227,7 +2229,7 @@ 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));
-	offsets = rec_get_offsets(rec, index, offsets, ULINT_UNDEFINED, heap);
+	ut_ad(rec_offs_validate(rec, index, offsets));
 	n_ext += btr_push_update_extern_fields(new_entry, update, *heap);
 
 	if (UNIV_LIKELY_NULL(page_zip)) {
@@ -2250,6 +2252,10 @@ 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,
@@ -2277,6 +2283,8 @@ 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);
 
@@ -2290,7 +2298,10 @@ make_external:
 						     rec, index, offsets, mtr);
 		}
 
-		btr_cur_compress_if_useful(cursor, mtr);
+		btr_cur_compress_if_useful(
+			cursor,
+			big_rec_vec != NULL && (flags & BTR_KEEP_POS_FLAG),
+			mtr);
 
 		if (page_zip && !dict_index_is_clust(index)
 		    && page_is_leaf(page)) {
@@ -2310,6 +2321,21 @@ 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);
@@ -2325,6 +2351,7 @@ 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.
@@ -2383,6 +2410,39 @@ 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 ===============*/
 
 /****************************************************************//**
@@ -2762,10 +2822,12 @@ UNIV_INTERN
 ibool
 btr_cur_compress_if_useful(
 /*=======================*/
-	btr_cur_t*	cursor,	/*!< in: cursor on the page to compress;
-				cursor does not stay valid if compression
-				occurs */
-	mtr_t*		mtr)	/*!< in: mtr */
+	btr_cur_t*	cursor,	/*!< in/out: cursor on the page to compress;
+				cursor does not stay valid if !adjust and
+				compression occurs */
+	ibool		adjust,	/*!< in: TRUE if should adjust the
+				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)),
@@ -2774,7 +2836,7 @@ btr_cur_compress_if_useful(
 				MTR_MEMO_PAGE_X_FIX));
 
 	return(btr_cur_compress_recommendation(cursor, mtr)
-	       && btr_compress(cursor, mtr));
+	       && btr_compress(cursor, adjust, mtr));
 }
 
 /*******************************************************//**
@@ -3016,7 +3078,7 @@ return_after_reservations:
 	mem_heap_free(heap);
 
 	if (ret == FALSE) {
-		ret = btr_cur_compress_if_useful(cursor, mtr);
+		ret = btr_cur_compress_if_useful(cursor, FALSE, mtr);
 	}
 
 	if (n_extents > 0) {

storage/innodb_plugin/include/btr0btr.h

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1994, 2010, Innobase Oy. All Rights Reserved.
+Copyright (c) 1994, 2011, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -470,11 +470,14 @@ UNIV_INTERN
 ibool
 btr_compress(
 /*=========*/
-	btr_cur_t*	cursor,	/*!< in: cursor on the page to merge or lift;
-				the page must not be empty: in record delete
-				use btr_discard_page if the page would become
-				empty */
-	mtr_t*		mtr);	/*!< in: mtr */
+	btr_cur_t*	cursor,	/*!< in/out: cursor on the page to merge
+				or lift; the page must not be empty:
+				when deleting records, use btr_discard_page()
+				if the page would become empty */
+	ibool		adjust,	/*!< in: TRUE if should adjust the
+				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

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1994, 2010, Innobase Oy. All Rights Reserved.
+Copyright (c) 1994, 2011, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -36,6 +36,9 @@ 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"
@@ -309,7 +312,9 @@ btr_cur_pessimistic_update(
 /*=======================*/
 	ulint		flags,	/*!< in: undo logging, locking, and rollback
 				flags */
-	btr_cur_t*	cursor,	/*!< in: cursor on the record to update */
+	btr_cur_t*	cursor,	/*!< in/out: 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 */
@@ -321,6 +326,16 @@ 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
@@ -376,10 +391,13 @@ UNIV_INTERN
 ibool
 btr_cur_compress_if_useful(
 /*=======================*/
-	btr_cur_t*	cursor,	/*!< in: cursor on the page to compress;
+	btr_cur_t*	cursor,	/*!< in/out: cursor on the page to compress;
 				cursor does not stay valid if compression
 				occurs */
-	mtr_t*		mtr);	/*!< in: mtr */
+	ibool		adjust,	/*!< in: TRUE if should adjust the
+				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

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 1994, 2011, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -139,7 +139,7 @@ btr_cur_compress_recommendation(
 	btr_cur_t*	cursor,	/*!< in: btr cursor */
 	mtr_t*		mtr)	/*!< in: mtr */
 {
-	page_t*		page;
+	const 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

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1995, 2010, Innobase Oy. All Rights Reserved.
+Copyright (c) 1995, 2011, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -467,6 +467,31 @@ 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

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1995, 2010, Innobase Oy. All Rights Reserved.
+Copyright (c) 1995, 2011, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2008, Google Inc.
 
 Portions of this file contain modifications contributed and copyrighted by
@@ -871,19 +871,6 @@ 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. */
@@ -1071,7 +1058,7 @@ buf_block_dbg_add_level(
 				where we have acquired latch */
 	ulint		level)	/*!< in: latching order level */
 {
-	sync_thread_add_level(&block->lock, level);
+	sync_thread_add_level(&block->lock, level, FALSE);
 }
 #endif /* UNIV_SYNC_DEBUG */
 #endif /* !UNIV_HOTBACKUP */

storage/innodb_plugin/include/page0cur.ic

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 1994, 2011, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -27,6 +27,8 @@ Created 10/4/1994 Heikki Tuuri
 #include "buf0types.h"
 
 #ifdef UNIV_DEBUG
+# include "rem0cmp.h"
+
 /*********************************************************//**
 Gets pointer to the page frame where the cursor is positioned.
 @return	page */
@@ -268,6 +270,7 @@ page_cur_tuple_insert(
 					      index, rec, offsets, mtr);
 	}
 
+	ut_ad(!rec || !cmp_dtuple_rec(tuple, rec, offsets));
 	mem_heap_free(heap);
 	return(rec);
 }

storage/innodb_plugin/include/page0page.h

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 1994, 2011, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -284,16 +284,42 @@ 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 middle record of record list. If there are an even number
-of records in the list, returns the first record of upper half-list.
-@return	middle record */
+Returns the nth record of the record list.
+This is the inverse function of page_rec_get_n_recs_before().
+@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
 rec_t*
 page_get_middle_rec(
 /*================*/
-	page_t*	page);	/*!< in: page */
-#ifndef UNIV_HOTBACKUP
+	page_t*	page)	/*!< in: page */
+	__attribute__((nonnull, warn_unused_result));
 /*************************************************************//**
 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
@@ -348,6 +374,7 @@ 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

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 1994, 2011, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -420,7 +420,37 @@ 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/include/sync0rw.ic

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 1995, 2011, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2008, Google Inc.
 
 Portions of this file contain modifications contributed and copyrighted by
@@ -603,16 +603,16 @@ rw_lock_x_unlock_direct(
 
 	ut_ad((lock->lock_word % X_LOCK_DECR) == 0);
 
-#ifdef UNIV_SYNC_DEBUG
-	rw_lock_remove_debug_info(lock, 0, RW_LOCK_EX);
-#endif
-
 	if (lock->lock_word == 0) {
 		lock->recursive = FALSE;
 		UNIV_MEM_INVALID(&lock->writer_thread,
 				 sizeof lock->writer_thread);
 	}
 
+#ifdef UNIV_SYNC_DEBUG
+	rw_lock_remove_debug_info(lock, 0, RW_LOCK_EX);
+#endif
+
 	lock->lock_word += X_LOCK_DECR;
 
 	ut_ad(!lock->waiters);

storage/innodb_plugin/include/sync0sync.h

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1995, 2010, Innobase Oy. All Rights Reserved.
+Copyright (c) 1995, 2011, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2008, Google Inc.
 
 Portions of this file contain modifications contributed and copyrighted by
@@ -219,8 +219,10 @@ void
 sync_thread_add_level(
 /*==================*/
 	void*	latch,	/*!< in: pointer to a mutex or an rw-lock */
-	ulint	level);	/*!< in: level in the latching order; if
+	ulint	level,	/*!< in: level in the latching order; if
 			SYNC_LEVEL_VARYING, nothing is done */
+	ibool	relock)	/*!< in: TRUE if re-entering an x-lock */
+	__attribute__((nonnull));
 /******************************************************************//**
 Removes a latch from the thread level array if it is found there.
 @return TRUE if found in the array; it is no error if the latch is

storage/innodb_plugin/page/page0cur.c

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1994, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 1994, 2011, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -1180,14 +1180,15 @@ 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. */
-		rec = page + PAGE_NEW_INFIMUM;
-
-		while (--pos) {
-			rec = page + rec_get_next_offs(rec, TRUE);
+		if (pos > 1) {
+			rec = page_rec_get_nth(page, pos - 1);
+		} else {
+			rec = page + PAGE_NEW_INFIMUM;
 		}
 
 		*current_rec = rec;
@@ -1283,6 +1284,12 @@ 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

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1994, 2010, Innobase Oy. All Rights Reserved.
+Copyright (c) 1994, 2011, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -1487,55 +1487,54 @@ page_dir_balance_slot(
 	}
 }
 
-#ifndef UNIV_HOTBACKUP
 /************************************************************//**
-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.
-@return	middle record */
+Returns the nth record of the record list.
+This is the inverse function of page_rec_get_n_recs_before().
+@return	nth record */
 UNIV_INTERN
-rec_t*
-page_get_middle_rec(
-/*================*/
-	page_t*	page)	/*!< in: page */
+const rec_t*
+page_rec_get_nth_const(
+/*===================*/
+	const page_t*	page,	/*!< in: page */
+	ulint		nth)	/*!< in: nth record */
 {
-	page_dir_slot_t*	slot;
-	ulint			middle;
+	const page_dir_slot_t*	slot;
 	ulint			i;
 	ulint			n_owned;
-	ulint			count;
-	rec_t*			rec;
+	const rec_t*		rec;
 
-	/* This many records we must leave behind */
-	middle = (page_get_n_recs(page) + PAGE_HEAP_NO_USER_LOW) / 2;
-
-	count = 0;
+	ut_ad(nth < UNIV_PAGE_SIZE / (REC_N_NEW_EXTRA_BYTES + 1));
 
 	for (i = 0;; i++) {
 
 		slot = page_dir_get_nth_slot(page, i);
 		n_owned = page_dir_slot_get_n_owned(slot);
 
-		if (count + n_owned > middle) {
+		if (n_owned > nth) {
 			break;
 		} else {
-			count += n_owned;
+			nth -= n_owned;
 		}
 	}
 
 	ut_ad(i > 0);
 	slot = page_dir_get_nth_slot(page, i - 1);
-	rec = (rec_t*) page_dir_slot_get_rec(slot);
-	rec = page_rec_get_next(rec);
+	rec = page_dir_slot_get_rec(slot);
 
-	/* There are now count records behind rec */
-
-	for (i = 0; i < middle - count; i++) {
-		rec = page_rec_get_next(rec);
+	if (page_is_comp(page)) {
+		do {
+			rec = page_rec_get_next_low(rec, TRUE);
+			ut_ad(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.
@@ -1597,6 +1596,7 @@ 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

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1996, 2010, Innobase Oy. All Rights Reserved.
+Copyright (c) 1996, 2011, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -345,9 +345,9 @@ row_ins_clust_index_entry_by_modify(
 			return(DB_LOCK_TABLE_FULL);
 
 		}
-		err = btr_cur_pessimistic_update(0, cursor,
-						 heap, big_rec, update,
-						 0, thr, mtr);
+		err = btr_cur_pessimistic_update(
+			BTR_KEEP_POS_FLAG, cursor, heap, big_rec, update,
+			0, thr, mtr);
 	}
 
 	return(err);
@@ -1986,6 +1986,7 @@ 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;
@@ -2089,6 +2090,42 @@ 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(
@@ -2117,8 +2154,6 @@ 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,
@@ -2132,6 +2167,7 @@ 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

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1996, 2009, Innobase Oy. All Rights Reserved.
+Copyright (c) 1996, 2011, Oracle and/or its affiliates. All Rights Reserved.
 
 This program is free software; you can redistribute it and/or modify it under
 the terms of the GNU General Public License as published by the Free Software
@@ -1969,28 +1969,43 @@ 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(BTR_NO_LOCKING_FLAG, btr_cur,
-					 &heap, &big_rec, node->update,
-					 node->cmpl_info, thr, mtr);
-	mtr_commit(mtr);
-
-	if (err == DB_SUCCESS && big_rec) {
+	err = btr_cur_pessimistic_update(
+		BTR_NO_LOCKING_FLAG | BTR_KEEP_POS_FLAG, btr_cur,
+		&heap, &big_rec, node->update, node->cmpl_info, thr, mtr);
+	if (big_rec) {
 		ulint		offsets_[REC_OFFS_NORMAL_SIZE];
 		rec_t*		rec;
 		rec_offs_init(offsets_);
 
-		mtr_start(mtr);
+		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(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);
-		mtr_commit(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);
 	}
 
+	mtr_commit(mtr);
+
 	if (UNIV_LIKELY_NULL(heap)) {
 		mem_heap_free(heap);
 	}

storage/innodb_plugin/sync/sync0rw.c

@@ -766,7 +766,9 @@ rw_lock_add_debug_info(
 	rw_lock_debug_mutex_exit();
 
 	if ((pass == 0) && (lock_type != RW_LOCK_WAIT_EX)) {
-		sync_thread_add_level(lock, lock->level);
+		sync_thread_add_level(lock, lock->level,
+				      lock_type == RW_LOCK_EX
+				      && lock->lock_word < 0);
 	}
 }
 

storage/innodb_plugin/sync/sync0sync.c

@@ -1,6 +1,6 @@
 /*****************************************************************************
 
-Copyright (c) 1995, 2010, Innobase Oy. All Rights Reserved.
+Copyright (c) 1995, 2011, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2008, Google Inc.
 
 Portions of this file contain modifications contributed and copyrighted by
@@ -656,7 +656,7 @@ mutex_set_debug_info(
 	ut_ad(mutex);
 	ut_ad(file_name);
 
-	sync_thread_add_level(mutex, mutex->level);
+	sync_thread_add_level(mutex, mutex->level, FALSE);
 
 	mutex->file_name = file_name;
 	mutex->line	 = line;
@@ -1083,8 +1083,9 @@ void
 sync_thread_add_level(
 /*==================*/
 	void*	latch,	/*!< in: pointer to a mutex or an rw-lock */
-	ulint	level)	/*!< in: level in the latching order; if
+	ulint	level,	/*!< in: level in the latching order; if
 			SYNC_LEVEL_VARYING, nothing is done */
+	ibool	relock)	/*!< in: TRUE if re-entering an x-lock */
 {
 	sync_level_t*	array;
 	sync_level_t*	slot;
@@ -1132,6 +1133,10 @@ sync_thread_add_level(
 
 	array = thread_slot->levels;
 
+	if (relock) {
+		goto levels_ok;
+	}
+
 	/* NOTE that there is a problem with _NODE and _LEAF levels: if the
 	B-tree height changes, then a leaf can change to an internal node
 	or the other way around. We do not know at present if this can cause
@@ -1269,6 +1274,7 @@ sync_thread_add_level(
 		ut_error;
 	}
 
+levels_ok:
 	for (i = 0; i < SYNC_THREAD_N_LEVELS; i++) {
 
 		slot = sync_thread_levels_get_nth(array, i);