Merge 10.2 into 10.3

mysql-test/suite/innodb/r/foreign_key.result

@@ -739,4 +739,15 @@ t2	CREATE TABLE `t2` (
 CREATE TABLE t2 (f1 INT NOT NULL)ENGINE=InnoDB;
 ERROR 42S01: Table 't2' already exists
 DROP TABLE t2, t1;
+#
+# MDEV-23685 SIGSEGV on ADD FOREIGN KEY after failed attempt
+# to create unique key on virtual column
+#
+CREATE TABLE t1 (pk INT PRIMARY KEY, a INT, b INT AS (a)) ENGINE=InnODB;
+INSERT INTO t1 (pk,a) VALUES (1,10),(2,10);
+ALTER TABLE t1 ADD UNIQUE INDEX ind9 (b), LOCK=SHARED;
+ERROR 23000: Duplicate entry '10' for key 'ind9'
+SET FOREIGN_KEY_CHECKS= 0;
+ALTER TABLE t1 ADD FOREIGN KEY (a) REFERENCES t1 (pk);
+DROP TABLE t1;
 # End of 10.2 tests

mysql-test/suite/innodb/t/foreign_key.test

@@ -720,6 +720,19 @@ SHOW CREATE TABLE t2;
 CREATE TABLE t2 (f1 INT NOT NULL)ENGINE=InnoDB;
 DROP TABLE t2, t1;
 
+--echo #
+--echo # MDEV-23685 SIGSEGV on ADD FOREIGN KEY after failed attempt
+--echo # to create unique key on virtual column
+--echo #
+CREATE TABLE t1 (pk INT PRIMARY KEY, a INT, b INT AS (a)) ENGINE=InnODB;
+
+INSERT INTO t1 (pk,a) VALUES (1,10),(2,10);
+--error ER_DUP_ENTRY
+ALTER TABLE t1 ADD UNIQUE INDEX ind9 (b), LOCK=SHARED;
+SET FOREIGN_KEY_CHECKS= 0;
+ALTER TABLE t1 ADD FOREIGN KEY (a) REFERENCES t1 (pk);
+DROP TABLE t1;
+
 --echo # End of 10.2 tests
 
 --source include/wait_until_count_sessions.inc

storage/innobase/btr/btr0btr.cc

@@ -1082,8 +1082,7 @@ btr_create(
 	if (type & DICT_IBUF) {
 		/* Allocate first the ibuf header page */
 		buf_block_t*	ibuf_hdr_block = fseg_create(
-			space, 0,
-			IBUF_HEADER + IBUF_TREE_SEG_HEADER, mtr);
+			space, IBUF_HEADER + IBUF_TREE_SEG_HEADER, mtr);
 
 		if (ibuf_hdr_block == NULL) {
 			return(FIL_NULL);
@@ -1114,7 +1113,7 @@ btr_create(
 		flst_init(block->frame + PAGE_HEADER + PAGE_BTR_IBUF_FREE_LIST,
 			  mtr);
 	} else {
-		block = fseg_create(space, 0,
+		block = fseg_create(space,
 				    PAGE_HEADER + PAGE_BTR_SEG_TOP, mtr);
 
 		if (block == NULL) {
@@ -1123,8 +1122,9 @@ btr_create(
 
 		buf_block_dbg_add_level(block, SYNC_TREE_NODE_NEW);
 
-		if (!fseg_create(space, block->page.id.page_no(),
-				 PAGE_HEADER + PAGE_BTR_SEG_LEAF, mtr)) {
+		if (!fseg_create(space,
+				 PAGE_HEADER + PAGE_BTR_SEG_LEAF, mtr,
+				 false, block)) {
 			/* Not enough space for new segment, free root
 			segment before return. */
 			btr_free_root(block, mtr,

storage/innobase/buf/buf0buf.cc

@@ -5626,14 +5626,13 @@ buf_page_create(
 	buf_frame_t*	frame;
 	buf_block_t*	block;
 	buf_block_t*	free_block	= NULL;
-	buf_pool_t*	buf_pool = buf_pool_get(page_id);
+	buf_pool_t*	buf_pool= buf_pool_get(page_id);
 	rw_lock_t*	hash_lock;
 
 	ut_ad(mtr->is_active());
 	ut_ad(page_id.space() != 0 || !page_size.is_compressed());
-
+loop:
 	free_block = buf_LRU_get_free_block(buf_pool);
-
 	buf_pool_mutex_enter(buf_pool);
 
 	hash_lock = buf_page_hash_lock_get(buf_pool, page_id);
@@ -5645,20 +5644,67 @@ buf_page_create(
 	    && buf_page_in_file(&block->page)
 	    && !buf_pool_watch_is_sentinel(buf_pool, &block->page)) {
 		ut_d(block->page.file_page_was_freed = FALSE);
-
+		buf_page_state page_state = buf_block_get_state(block);
 #ifdef BTR_CUR_HASH_ADAPT
-		bool drop_hash_entry =
-			(block->page.state == BUF_BLOCK_FILE_PAGE
-			 && block->index);
+		const dict_index_t *drop_hash_entry= NULL;
+#endif
+		switch (page_state) {
+		default:
+			ut_ad(0);
+			break;
+		case BUF_BLOCK_ZIP_PAGE:
+		case BUF_BLOCK_ZIP_DIRTY:
+			buf_block_init_low(free_block);
+			mutex_enter(&buf_pool->zip_mutex);
 
-		if (drop_hash_entry) {
-			mutex_enter(&block->mutex);
-			/* Avoid a hang if I/O is going on. Release
-			the buffer pool mutex and page hash lock
-			and wait for I/O to complete */
-			while (buf_block_get_io_fix(block) != BUF_IO_NONE) {
+			buf_page_mutex_enter(free_block);
+			if (buf_page_get_io_fix(&block->page) != BUF_IO_NONE) {
+				mutex_exit(&buf_pool->zip_mutex);
+				rw_lock_x_unlock(hash_lock);
+				buf_LRU_block_free_non_file_page(free_block);
+				buf_pool_mutex_exit(buf_pool);
+				buf_page_mutex_exit(free_block);
+
+				goto loop;
+			}
+
+			rw_lock_x_lock(&free_block->lock);
+
+			buf_relocate(&block->page, &free_block->page);
+			if (page_state == BUF_BLOCK_ZIP_DIRTY) {
+				ut_ad(block->page.in_flush_list);
+				ut_ad(block->page.oldest_modification > 0);
+				buf_flush_relocate_on_flush_list(
+					&block->page, &free_block->page);
+			} else {
+				ut_ad(block->page.oldest_modification == 0);
+				ut_ad(!block->page.in_flush_list);
+#ifdef UNIV_DEBUG
+				UT_LIST_REMOVE(
+					buf_pool->zip_clean, &block->page);
+#endif
+			}
+
+			free_block->page.state = BUF_BLOCK_FILE_PAGE;
+			mutex_exit(&buf_pool->zip_mutex);
+			free_block->lock_hash_val = lock_rec_hash(
+					page_id.space(), page_id.page_no());
+			buf_unzip_LRU_add_block(free_block, false);
+			buf_page_free_descriptor(&block->page);
+			block = free_block;
 			buf_block_fix(block);
-				mutex_exit(&block->mutex);
+			buf_page_mutex_exit(free_block);
+			free_block = NULL;
+			break;
+		case BUF_BLOCK_FILE_PAGE:
+			buf_block_fix(block);
+			const int32_t num_fix_count =
+				mtr->get_fix_count(block) + 1;
+			buf_page_mutex_enter(block);
+			while (buf_block_get_io_fix(block) != BUF_IO_NONE
+			       || (num_fix_count
+				   != block->page.buf_fix_count)) {
+				buf_page_mutex_exit(block);
 				buf_pool_mutex_exit(buf_pool);
 				rw_lock_x_unlock(hash_lock);
 
@@ -5666,33 +5712,39 @@ buf_page_create(
 
 				buf_pool_mutex_enter(buf_pool);
 				rw_lock_x_lock(hash_lock);
-				mutex_enter(&block->mutex);
-				buf_block_unfix(block);
+				buf_page_mutex_enter(block);
 			}
+
 			rw_lock_x_lock(&block->lock);
-			mutex_exit(&block->mutex);
-		}
+			buf_page_mutex_exit(block);
+#ifdef BTR_CUR_HASH_ADAPT
+			drop_hash_entry = block->index;
 #endif
+			break;
+		}
 		/* Page can be found in buf_pool */
 		buf_pool_mutex_exit(buf_pool);
 		rw_lock_x_unlock(hash_lock);
 
+		if (free_block) {
 			buf_block_free(free_block);
+		}
 #ifdef BTR_CUR_HASH_ADAPT
 		if (drop_hash_entry) {
 			btr_search_drop_page_hash_index(block);
-			rw_lock_x_unlock(&block->lock);
 		}
 #endif /* BTR_CUR_HASH_ADAPT */
 
-		if (!recv_recovery_is_on()) {
-			return buf_page_get_with_no_latch(page_id, page_size,
-							  mtr);
+#ifdef UNIV_DEBUG
+		if (!fsp_is_system_temporary(page_id.space())) {
+			rw_lock_s_lock_nowait(
+				&block->debug_latch,
+				__FILE__, __LINE__);
 		}
+#endif /* UNIV_DEBUG */
+
+		mtr_memo_push(mtr, block, MTR_MEMO_PAGE_X_FIX);
 
-		mutex_exit(&recv_sys->mutex);
-		block = buf_page_get_with_no_latch(page_id, page_size, mtr);
-		mutex_enter(&recv_sys->mutex);
 		return block;
 	}
 
@@ -5707,6 +5759,8 @@ buf_page_create(
 
 	buf_page_init(buf_pool, page_id, page_size, block);
 
+	rw_lock_x_lock(&block->lock);
+
 	rw_lock_x_unlock(hash_lock);
 
 	/* The block must be put to the LRU list */
@@ -5724,7 +5778,6 @@ buf_page_create(
 		by IO-fixing and X-latching the block. */
 
 		buf_page_set_io_fix(&block->page, BUF_IO_READ);
-		rw_lock_x_lock(&block->lock);
 
 		buf_page_mutex_exit(block);
 		/* buf_pool->mutex may be released and reacquired by
@@ -5746,12 +5799,11 @@ buf_page_create(
 		buf_unzip_LRU_add_block(block, FALSE);
 
 		buf_page_set_io_fix(&block->page, BUF_IO_NONE);
-		rw_lock_x_unlock(&block->lock);
 	}
 
 	buf_pool_mutex_exit(buf_pool);
 
-	mtr_memo_push(mtr, block, MTR_MEMO_BUF_FIX);
+	mtr_memo_push(mtr, block, MTR_MEMO_PAGE_X_FIX);
 
 	buf_page_set_accessed(&block->page);
 

storage/innobase/buf/buf0dblwr.cc

@@ -170,6 +170,7 @@ buf_dblwr_create()
 {
 	buf_block_t*	block2;
 	buf_block_t*	new_block;
+	buf_block_t*	trx_sys_block;
 	byte*	doublewrite;
 	byte*	fseg_header;
 	ulint	page_no;
@@ -205,9 +206,13 @@ buf_dblwr_create()
 		}
 	}
 
-	block2 = fseg_create(fil_system.sys_space, TRX_SYS_PAGE_NO,
-			     TRX_SYS_DOUBLEWRITE
-			     + TRX_SYS_DOUBLEWRITE_FSEG, &mtr);
+	trx_sys_block = buf_page_get(
+		page_id_t(TRX_SYS_SPACE, TRX_SYS_PAGE_NO),
+		univ_page_size, RW_X_LATCH, &mtr);
+
+	block2 = fseg_create(fil_system.sys_space,
+			     TRX_SYS_DOUBLEWRITE + TRX_SYS_DOUBLEWRITE_FSEG,
+			     &mtr, false, trx_sys_block);
 
 	if (block2 == NULL) {
 too_small:

storage/innobase/dict/dict0boot.cc

 /*****************************************************************************
 
 Copyright (c) 1996, 2017, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2016, 2019, MariaDB Corporation.
+Copyright (c) 2016, 2020, MariaDB Corporation.
 
 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
@@ -180,7 +180,7 @@ dict_hdr_create(
 
 	/* Create the dictionary header file block in a new, allocated file
 	segment in the system tablespace */
-	block = fseg_create(fil_system.sys_space, 0,
+	block = fseg_create(fil_system.sys_space,
 			    DICT_HDR + DICT_HDR_FSEG_HEADER, mtr);
 
 	ut_a(DICT_HDR_PAGE_NO == block->page.id.page_no());

storage/innobase/dict/dict0dict.cc

@@ -6514,7 +6514,11 @@ dict_foreign_qualify_index(
 		return(false);
 	}
 
-	if (index->type & (DICT_SPATIAL | DICT_FTS)) {
+	if (index->type & (DICT_SPATIAL | DICT_FTS | DICT_CORRUPT)) {
+		return false;
+	}
+
+	if (index->online_status >= ONLINE_INDEX_ABORTED) {
 		return false;
 	}
 

storage/innobase/fsp/fsp0fsp.cc

@@ -117,7 +117,6 @@ to minimize file space fragmentation.
 @param[in]	direction		if the new page is needed because of
 an index page split, and records are inserted there in order, into which
 direction they go alphabetically: FSP_DOWN, FSP_UP, FSP_NO_DIR
-@param[in]	rw_latch		RW_SX_LATCH, RW_X_LATCH
 @param[in,out]	mtr			mini-transaction
 @param[in,out]	init_mtr		mtr or another mini-transaction in
 which the page should be initialized. If init_mtr != mtr, but the page is
@@ -136,7 +135,6 @@ fseg_alloc_free_page_low(
 	fseg_inode_t*		seg_inode,
 	ulint			hint,
 	byte			direction,
-	rw_lock_type_t		rw_latch,
 	mtr_t*			mtr,
 	mtr_t*			init_mtr
 #ifdef UNIV_DEBUG
@@ -210,7 +208,9 @@ xdes_set_bit(
 	ulint	bit_index;
 	ulint	descr_byte;
 
-	ut_ad(mtr_memo_contains_page(mtr, descr, MTR_MEMO_PAGE_SX_FIX));
+	ut_ad(mtr_memo_contains_page(
+			mtr, descr,
+			MTR_MEMO_PAGE_SX_FIX | MTR_MEMO_PAGE_X_FIX));
 	ut_ad((bit == XDES_FREE_BIT) || (bit == XDES_CLEAN_BIT));
 	ut_ad(offset < FSP_EXTENT_SIZE);
 
@@ -337,7 +337,9 @@ xdes_set_state(
 	ut_ad(descr && mtr);
 	ut_ad(state >= XDES_FREE);
 	ut_ad(state <= XDES_FSEG);
-	ut_ad(mtr_memo_contains_page(mtr, descr, MTR_MEMO_PAGE_SX_FIX));
+	ut_ad(mtr_memo_contains_page(
+			mtr, descr,
+			MTR_MEMO_PAGE_SX_FIX | MTR_MEMO_PAGE_X_FIX));
 
 	mlog_write_ulint(descr + XDES_STATE, state, MLOG_4BYTES, mtr);
 }
@@ -374,7 +376,9 @@ xdes_init(
 	ulint	i;
 
 	ut_ad(descr && mtr);
-	ut_ad(mtr_memo_contains_page(mtr, descr, MTR_MEMO_PAGE_SX_FIX));
+	ut_ad(mtr_memo_contains_page(
+			mtr, descr,
+			MTR_MEMO_PAGE_SX_FIX | MTR_MEMO_PAGE_X_FIX));
 	ut_ad((XDES_SIZE - XDES_BITMAP) % 4 == 0);
 
 	for (i = XDES_BITMAP; i < XDES_SIZE; i += 4) {
@@ -409,7 +413,8 @@ xdes_get_descriptor_with_space_hdr(
 	ulint	descr_page_no;
 	page_t*	descr_page;
 	ut_ad(mtr_memo_contains(mtr, space, MTR_MEMO_SPACE_X_LOCK));
-	ut_ad(mtr_memo_contains_page(mtr, sp_header, MTR_MEMO_PAGE_SX_FIX));
+	ut_ad(mtr_memo_contains_page(mtr, sp_header, MTR_MEMO_PAGE_SX_FIX)
+	      || mtr_memo_contains_page(mtr, sp_header, MTR_MEMO_PAGE_X_FIX));
 	ut_ad(page_offset(sp_header) == FSP_HEADER_OFFSET);
 	/* Read free limit and space size */
 	limit = mach_read_from_4(sp_header + FSP_FREE_LIMIT);
@@ -659,7 +664,6 @@ void fsp_header_init(fil_space_t* space, ulint size, mtr_t* mtr)
 
 	mtr_x_lock_space(space, mtr);
 	buf_block_t* block = buf_page_create(page_id, page_size, mtr);
-	buf_page_get(page_id, page_size, RW_SX_LATCH, mtr);
 	buf_block_dbg_add_level(block, SYNC_FSP_PAGE);
 
 	space->size_in_header = size;
@@ -1002,9 +1006,6 @@ fsp_fill_free_list(
 				block = buf_page_create(
 					page_id, page_size, mtr);
 
-				buf_page_get(
-					page_id, page_size, RW_SX_LATCH, mtr);
-
 				buf_block_dbg_add_level(block, SYNC_FSP_PAGE);
 
 				fsp_init_file_page(space, block, mtr);
@@ -1022,9 +1023,6 @@ fsp_fill_free_list(
 				block = buf_page_create(
 					page_id, page_size, mtr);
 
-				buf_page_get(
-					page_id, page_size, RW_SX_LATCH, mtr);
-
 				buf_block_dbg_add_level(block, SYNC_FSP_PAGE);
 
 				fsp_init_file_page(space, block, mtr);
@@ -1170,7 +1168,6 @@ x-latched only by mtr, and freed in mtr in that case.
 @param[in,out]	space		tablespace
 @param[in]	offset		page number of the allocated page
 @param[in]	page_size	page size of the allocated page
-@param[in]	rw_latch	RW_SX_LATCH, RW_X_LATCH
 @param[in,out]	mtr		mini-transaction of the allocation
 @param[in,out]	init_mtr	mini-transaction for initializing the page
 @return block, initialized if init_mtr==mtr
@@ -1181,7 +1178,6 @@ fsp_page_create(
 	fil_space_t*		space,
 	page_no_t		offset,
 	const page_size_t&	page_size,
-	rw_lock_type_t		rw_latch,
 	mtr_t*			mtr,
 	mtr_t*			init_mtr)
 {
@@ -1191,26 +1187,10 @@ fsp_page_create(
 						page_size, init_mtr);
 
 	ut_d(bool latched = mtr_memo_contains_flagged(mtr, block,
-						      MTR_MEMO_PAGE_X_FIX
-						      | MTR_MEMO_PAGE_SX_FIX));
-
-	ut_ad(rw_latch == RW_X_LATCH || rw_latch == RW_SX_LATCH);
-
-	/* Mimic buf_page_get(), but avoid the buf_pool->page_hash lookup. */
-	if (rw_latch == RW_X_LATCH) {
-		rw_lock_x_lock(&block->lock);
-	} else {
-		rw_lock_sx_lock(&block->lock);
-	}
-
-	buf_block_buf_fix_inc(block, __FILE__, __LINE__);
-	mtr_memo_push(init_mtr, block, rw_latch == RW_X_LATCH
-		      ? MTR_MEMO_PAGE_X_FIX : MTR_MEMO_PAGE_SX_FIX);
+						      MTR_MEMO_PAGE_X_FIX));
 
 	if (init_mtr == mtr
-	    || (rw_latch == RW_X_LATCH
-		? rw_lock_get_x_lock_count(&block->lock) == 1
-		: rw_lock_get_sx_lock_count(&block->lock) == 1)) {
+	    || rw_lock_get_x_lock_count(&block->lock) == 1) {
 
 		/* Initialize the page, unless it was already
 		SX-latched in mtr. (In this case, we would want to
@@ -1227,7 +1207,6 @@ The page is marked as used.
 @param[in,out]	space		tablespace
 @param[in]	page_size	page size
 @param[in]	hint		hint of which page would be desirable
-@param[in]	rw_latch	RW_SX_LATCH, RW_X_LATCH
 @param[in,out]	mtr		mini-transaction
 @param[in,out]	init_mtr	mini-transaction in which the page should be
 initialized (may be the same as mtr)
@@ -1241,7 +1220,6 @@ fsp_alloc_free_page(
 	fil_space_t*		space,
 	const page_size_t&	page_size,
 	ulint			hint,
-	rw_lock_type_t		rw_latch,
 	mtr_t*			mtr,
 	mtr_t*			init_mtr)
 {
@@ -1333,8 +1311,7 @@ fsp_alloc_free_page(
 	}
 
 	fsp_alloc_from_free_frag(header, descr, free, mtr);
-	return(fsp_page_create(space, page_no, page_size, rw_latch,
-			       mtr, init_mtr));
+	return(fsp_page_create(space, page_no, page_size, mtr, init_mtr));
 }
 
 /** Frees a single page of a space.
@@ -1571,8 +1548,7 @@ fsp_alloc_seg_inode_page(
 
 	const page_size_t	page_size(space->flags);
 
-	block = fsp_alloc_free_page(
-		space, page_size, 0, RW_SX_LATCH, mtr, mtr);
+	block = fsp_alloc_free_page(space, page_size, 0, mtr, mtr);
 
 	if (block == NULL) {
 
@@ -1580,7 +1556,7 @@ fsp_alloc_seg_inode_page(
 	}
 
 	buf_block_dbg_add_level(block, SYNC_FSP_PAGE);
-	ut_ad(rw_lock_get_sx_lock_count(&block->lock) == 1);
+	ut_ad(rw_lock_get_x_lock_count(&block->lock) == 1);
 
 	mlog_write_ulint(block->frame + FIL_PAGE_TYPE, FIL_PAGE_INODE,
 			 MLOG_2BYTES, mtr);
@@ -1889,31 +1865,22 @@ fseg_get_n_frag_pages(
 	return(count);
 }
 
-/**********************************************************************//**
-Creates a new segment.
-@return the block where the segment header is placed, x-latched, NULL
-if could not create segment because of lack of space */
+/** Create a new segment.
+@param space                tablespace
+@param byte_offset          byte offset of the created segment header
+@param mtr                  mini-transaction
+@param has_done_reservation whether fsp_reserve_free_extents() was invoked
+@param block                block where segment header is placed,
+                            or NULL to allocate an additional page for that
+@return the block where the segment header is placed, x-latched
+@retval NULL if could not create segment because of lack of space */
 buf_block_t*
-fseg_create(
-	fil_space_t* space, /*!< in,out: tablespace */
-	ulint	page,	/*!< in: page where the segment header is placed: if
-			this is != 0, the page must belong to another segment,
-			if this is 0, a new page will be allocated and it
-			will belong to the created segment */
-	ulint	byte_offset, /*!< in: byte offset of the created segment header
-			on the page */
-	mtr_t*	mtr,
-   	bool	has_done_reservation) /*!< in: whether the caller
-			has already done the reservation for the pages with
-			fsp_reserve_free_extents (at least 2 extents: one for
-			the inode and the other for the segment) then there is
-			no need to do the check for this individual
-			operation */
+fseg_create(fil_space_t *space, ulint byte_offset, mtr_t *mtr,
+            bool has_done_reservation, buf_block_t *block)
 {
 	fsp_header_t*	space_header;
 	fseg_inode_t*	inode;
 	ib_id_t		seg_id;
-	buf_block_t*	block	= 0; /* remove warning */
 	fseg_header_t*	header	= 0; /* remove warning */
 	ulint		n_reserved;
 	ulint		i;
@@ -1921,6 +1888,7 @@ fseg_create(
 	DBUG_ENTER("fseg_create");
 
 	ut_ad(mtr);
+	ut_ad(byte_offset >= FIL_PAGE_DATA);
 	ut_ad(byte_offset + FSEG_HEADER_SIZE
 	      <= srv_page_size - FIL_PAGE_DATA_END);
 
@@ -1928,14 +1896,11 @@ fseg_create(
 	const page_size_t	page_size(space->flags);
 	ut_d(space->modify_check(*mtr));
 
-	if (page != 0) {
-		block = buf_page_get(page_id_t(space->id, page), page_size,
-				     RW_SX_LATCH, mtr);
-
+	if (block) {
 		header = byte_offset + buf_block_get_frame(block);
 
-		const ulint	type = space->id == TRX_SYS_SPACE
-			&& page == TRX_SYS_PAGE_NO
+		const ulint type = block->page.id == page_id_t(TRX_SYS_SPACE,
+							       TRX_SYS_PAGE_NO)
 			? FIL_PAGE_TYPE_TRX_SYS
 			: FIL_PAGE_TYPE_SYS;
 
@@ -1976,9 +1941,9 @@ fseg_create(
 		fseg_set_nth_frag_page_no(inode, i, FIL_NULL, mtr);
 	}
 
-	if (page == 0) {
+	if (!block) {
 		block = fseg_alloc_free_page_low(space, page_size,
-						 inode, 0, FSP_UP, RW_SX_LATCH,
+						 inode, 0, FSP_UP,
 						 mtr, mtr
 #ifdef UNIV_DEBUG
 						 , has_done_reservation
@@ -1996,7 +1961,7 @@ fseg_create(
 			goto funct_exit;
 		}
 
-		ut_ad(rw_lock_get_sx_lock_count(&block->lock) == 1);
+		ut_ad(rw_lock_get_x_lock_count(&block->lock) == 1);
 
 		header = byte_offset + buf_block_get_frame(block);
 		mlog_write_ulint(buf_block_get_frame(block) + FIL_PAGE_TYPE,
@@ -2215,7 +2180,6 @@ minimize file space fragmentation.
 @param[in]	direction		if the new page is needed because of
 an index page split, and records are inserted there in order, into which
 direction they go alphabetically: FSP_DOWN, FSP_UP, FSP_NO_DIR
-@param[in]	rw_latch		RW_SX_LATCH, RW_X_LATCH
 @param[in,out]	mtr			mini-transaction
 @param[in,out]	init_mtr		mtr or another mini-transaction in
 which the page should be initialized. If init_mtr != mtr, but the page is
@@ -2234,7 +2198,6 @@ fseg_alloc_free_page_low(
 	fseg_inode_t*		seg_inode,
 	ulint			hint,
 	byte			direction,
-	rw_lock_type_t		rw_latch,
 	mtr_t*			mtr,
 	mtr_t*			init_mtr
 #ifdef UNIV_DEBUG
@@ -2377,7 +2340,7 @@ fseg_alloc_free_page_low(
 		/* 6. We allocate an individual page from the space
 		===================================================*/
 		buf_block_t* block = fsp_alloc_free_page(
-			space, page_size, hint, rw_latch, mtr, init_mtr);
+			space, page_size, hint, mtr, init_mtr);
 
 		ut_ad(!has_done_reservation || block != NULL);
 
@@ -2458,8 +2421,7 @@ fseg_alloc_free_page_low(
 		fseg_mark_page_used(seg_inode, ret_page, ret_descr, mtr);
 	}
 
-	return(fsp_page_create(space, ret_page, page_size, rw_latch,
-			       mtr, init_mtr));
+	return(fsp_page_create(space, ret_page, page_size, mtr, init_mtr));
 }
 
 /**********************************************************************//**
@@ -2514,7 +2476,7 @@ fseg_alloc_free_page_general(
 
 	block = fseg_alloc_free_page_low(space, page_size,
 					 inode, hint, direction,
-					 RW_X_LATCH, mtr, init_mtr
+					 mtr, init_mtr
 #ifdef UNIV_DEBUG
 					 , has_done_reservation
 #endif /* UNIV_DEBUG */
@@ -3209,7 +3171,7 @@ fseg_print_low(
 	ulint	page_no;
 	ib_id_t	seg_id;
 
-	ut_ad(mtr_memo_contains_page(mtr, inode, MTR_MEMO_PAGE_SX_FIX));
+	ut_ad(mtr_memo_contains_page(mtr, inode, MTR_MEMO_PAGE_X_FIX));
 	space = page_get_space_id(page_align(inode));
 	page_no = page_get_page_no(page_align(inode));
 

storage/innobase/include/dict0mem.h

@@ -1047,6 +1047,25 @@ struct dict_index_t{
 	/** @return whether the index includes virtual columns */
 	bool has_virtual() const { return type & DICT_VIRTUAL; }
 
+	/** @return the position of DB_TRX_ID */
+	uint16_t db_trx_id() const {
+		DBUG_ASSERT(is_primary());
+		DBUG_ASSERT(n_uniq);
+		return n_uniq;
+	}
+	/** @return the position of DB_ROLL_PTR */
+	uint16_t db_roll_ptr() const
+	{
+		return static_cast<uint16_t>(db_trx_id() + 1);
+	}
+
+	/** @return the offset of the metadata BLOB field,
+	or the first user field after the PRIMARY KEY,DB_TRX_ID,DB_ROLL_PTR */
+	uint16_t first_user_field() const
+	{
+		return static_cast<uint16_t>(db_trx_id() + 2);
+	}
+
 	/** @return whether the index is corrupted */
 	inline bool is_corrupted() const;
 

storage/innobase/include/fsp0fsp.h

@@ -383,26 +383,18 @@ fsp_header_init_fields(
 void fsp_header_init(fil_space_t* space, ulint size, mtr_t* mtr)
 	MY_ATTRIBUTE((nonnull));
 
-/**********************************************************************//**
-Creates a new segment.
-@return the block where the segment header is placed, x-latched, NULL
-if could not create segment because of lack of space */
+/** Create a new segment.
+@param space                tablespace
+@param byte_offset          byte offset of the created segment header
+@param mtr                  mini-transaction
+@param has_done_reservation whether fsp_reserve_free_extents() was invoked
+@param block                block where segment header is placed,
+                            or NULL to allocate an additional page for that
+@return the block where the segment header is placed, x-latched
+@retval NULL if could not create segment because of lack of space */
 buf_block_t*
-fseg_create(
-	fil_space_t* space, /*!< in,out: tablespace */
-	ulint	page,	/*!< in: page where the segment header is placed: if
-			this is != 0, the page must belong to another segment,
-			if this is 0, a new page will be allocated and it
-			will belong to the created segment */
-	ulint	byte_offset, /*!< in: byte offset of the created segment header
-			on the page */
-	mtr_t*	mtr,
-   	bool	has_done_reservation = false); /*!< in: whether the caller
-			has already done the reservation for the pages with
-			fsp_reserve_free_extents (at least 2 extents: one for
-			the inode and the other for the segment) then there is
-			no need to do the check for this individual
-			operation */
+fseg_create(fil_space_t *space, ulint byte_offset, mtr_t *mtr,
+            bool has_done_reservation= false, buf_block_t *block= NULL);
 
 /**********************************************************************//**
 Calculates the number of pages reserved by a segment, and how many pages are

storage/innobase/include/mtr0mtr.h

@@ -437,6 +437,10 @@ struct mtr_t {
 	static inline bool is_block_dirtied(const buf_block_t* block)
 		MY_ATTRIBUTE((warn_unused_result));
 
+	/** Get the buffer fix count for the block added by this mtr.
+	@param[in]	block	block to be checked
+	@return number of buffer count added by this mtr */
+	int32_t get_fix_count(const buf_block_t *block);
 private:
 	/** Prepare to write the mini-transaction log to the redo log buffer.
 	@return number of bytes to write in finish_write() */

storage/innobase/log/log0recv.cc

@@ -2323,7 +2323,6 @@ static buf_block_t* recv_recovery_create_page_low(const page_id_t page_id,
   {
     i.created= true;
     buf_block_dbg_add_level(block, SYNC_NO_ORDER_CHECK);
-    mtr.x_latch_at_savepoint(0, block);
     recv_recover_page(block, mtr, recv_addr, i.lsn);
     ut_ad(mtr.has_committed());
   }

storage/innobase/mtr/mtr0mtr.cc

@@ -308,6 +308,24 @@ struct DebugCheck {
 };
 #endif
 
+/** Find buffer fix count of the given block acquired by the
+mini-transaction */
+struct FindBlock
+{
+  int32_t num_fix;
+  const buf_block_t *const block;
+
+  FindBlock(const buf_block_t *block_buf): num_fix(0), block(block_buf) {}
+
+  bool operator()(const mtr_memo_slot_t* slot)
+  {
+    if (slot->object == block)
+      ut_d(if (slot->type != MTR_MEMO_MODIFY))
+      num_fix++;
+    return true;
+  }
+};
+
 /** Release a resource acquired by the mini-transaction. */
 struct ReleaseBlocks {
 	/** Release specific object */
@@ -736,6 +754,14 @@ inline lsn_t mtr_t::finish_write(ulint len)
 	return start_lsn;
 }
 
+int32_t mtr_t::get_fix_count(const buf_block_t *block)
+{
+  Iterate<FindBlock> iteration((FindBlock(block)));
+  if (m_memo.for_each_block(iteration))
+    return iteration.functor.num_fix;
+  return 0;
+}
+
 #ifdef UNIV_DEBUG
 /** Check if memo contains the given item.
 @return	true if contains */

storage/innobase/row/row0sel.cc

@@ -3216,15 +3216,41 @@ row_sel_build_prev_vers_for_mysql(
 	return(err);
 }
 
-/** Helper class to cache clust_rec and old_ver */
+/** Helper class to cache clust_rec and old_vers */
 class Row_sel_get_clust_rec_for_mysql
 {
   const rec_t *cached_clust_rec;
   rec_t *cached_old_vers;
+  lsn_t cached_lsn;
+  page_id_t cached_page_id;
+
+#ifdef UNIV_DEBUG
+  void check_eq(const dict_index_t *index, const rec_offs *offsets) const
+  {
+    rec_offs vers_offs[REC_OFFS_HEADER_SIZE + MAX_REF_PARTS];
+    rec_offs_init(vers_offs);
+    mem_heap_t *heap= NULL;
+
+    ut_ad(rec_offs_validate(cached_clust_rec, index, offsets));
+    ut_ad(index->first_user_field() <= rec_offs_n_fields(offsets));
+    ut_ad(vers_offs == rec_get_offsets(cached_old_vers, index, vers_offs, true,
+                                       index->db_trx_id(), &heap));
+    ut_ad(!heap);
+    for (unsigned n= index->db_trx_id(); n--; )
+    {
+      const dict_col_t *col= dict_index_get_nth_col(index, n);
+      ulint len1, len2;
+      const byte *b1= rec_get_nth_field(cached_clust_rec, offsets, n, &len1);
+      const byte *b2= rec_get_nth_field(cached_old_vers, vers_offs, n, &len2);
+      ut_ad(!cmp_data_data(col->mtype, col->prtype, b1, len1, b2, len2));
+    }
+  }
+#endif
 
 public:
   Row_sel_get_clust_rec_for_mysql() :
-	cached_clust_rec(NULL), cached_old_vers(NULL) {}
+    cached_clust_rec(NULL), cached_old_vers(NULL), cached_lsn(0),
+    cached_page_id(page_id_t(0,0)) {}
 
   dberr_t operator()(row_prebuilt_t *prebuilt, dict_index_t *sec_index,
                      const rec_t *rec, que_thr_t *thr, const rec_t **out_rec,
@@ -3429,8 +3455,18 @@ Row_sel_get_clust_rec_for_mysql::operator()(
 		    && !lock_clust_rec_cons_read_sees(
 			    clust_rec, clust_index, *offsets,
 			    &trx->read_view)) {
+			const buf_page_t& bpage = btr_pcur_get_block(
+				prebuilt->clust_pcur)->page;
+
+			lsn_t lsn = bpage.newest_modification;
+			if (!lsn) {
+				lsn = mach_read_from_8(
+					page_align(clust_rec) + FIL_PAGE_LSN);
+			}
 
-			if (clust_rec != cached_clust_rec) {
+			if (lsn != cached_lsn
+			    || bpage.id != cached_page_id
+			    || clust_rec != cached_clust_rec) {
 				/* The following call returns 'offsets' associated with
 				'old_vers' */
 				err = row_sel_build_prev_vers_for_mysql(
@@ -3442,6 +3478,8 @@ Row_sel_get_clust_rec_for_mysql::operator()(
 
 					goto err_exit;
 				}
+				cached_lsn = lsn;
+				cached_page_id = bpage.id;
 				cached_clust_rec = clust_rec;
 				cached_old_vers = old_vers;
 			} else {
@@ -3452,7 +3490,8 @@ Row_sel_get_clust_rec_for_mysql::operator()(
 				version of clust_rec and its old version
 				old_vers. Re-calculate the offsets for old_vers. */
 
-				if (old_vers != NULL) {
+				if (old_vers) {
+					ut_d(check_eq(clust_index, *offsets));
 					*offsets = rec_get_offsets(
 						old_vers, clust_index, *offsets,
 						true, ULINT_UNDEFINED, offset_heap);

storage/innobase/trx/trx0rseg.cc

@@ -303,7 +303,7 @@ trx_rseg_header_create(
 	ut_ad(!sys_header == (space == fil_system.temp_space));
 
 	/* Allocate a new file segment for the rollback segment */
-	block = fseg_create(space, 0, TRX_RSEG + TRX_RSEG_FSEG_HEADER, mtr);
+	block = fseg_create(space, TRX_RSEG + TRX_RSEG_FSEG_HEADER, mtr);
 
 	if (block == NULL) {
 		/* No space left */

storage/innobase/trx/trx0sys.cc

@@ -160,7 +160,7 @@ trx_sysf_create(
 	compile_time_assert(TRX_SYS_SPACE == 0);
 
 	/* Create the trx sys file block in a new allocated file segment */
-	block = fseg_create(fil_system.sys_space, 0,
+	block = fseg_create(fil_system.sys_space,
 			    TRX_SYS + TRX_SYS_FSEG_HEADER,
 			    mtr);
 	buf_block_dbg_add_level(block, SYNC_TRX_SYS_HEADER);

storage/innobase/trx/trx0undo.cc

@@ -526,7 +526,7 @@ trx_undo_seg_create(fil_space_t* space, trx_rsegf_t* rseg_hdr, ulint* id,
 	}
 
 	/* Allocate a new file segment for the undo log */
-	block = fseg_create(space, 0, TRX_UNDO_SEG_HDR + TRX_UNDO_FSEG_HEADER,
+	block = fseg_create(space, TRX_UNDO_SEG_HDR + TRX_UNDO_FSEG_HEADER,
 			    mtr, true);
 
 	space->release_free_extents(n_reserved);