MDEV-14126: Remove page_is_root()

The predicate page_is_root(), which was added in MariaDB Server 10.2.2, is based on a wrong assumption. Under some circumstances, InnoDB can transform B-trees into a degenerate state where a non-leaf page has no sibling pages. Because of this, we cannot assume that a page that has no siblings is the root page. This bug will be tracked as MDEV-19022. Because of the bug that may affect many InnoDB data files, we must remove and replace the wrong predicate. Using the wrong predicate can cause corruption. A leaf page is not allowed to be empty except if it is the root page, and the entire table is empty.

Author: dr-m

storage/innobase/btr/btr0btr.cc

@@ -2,7 +2,7 @@
 
 Copyright (c) 1994, 2016, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2012, Facebook Inc.
-Copyright (c) 2014, 2017, MariaDB Corporation.
+Copyright (c) 2014, 2019, 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
@@ -378,8 +378,7 @@ btr_root_adjust_on_import(
 page = buf_block_get_frame(block);
 page_zip = buf_block_get_page_zip(block);
 
-if (!page_is_root(page)) {
-
+if (!fil_page_index_page_check(page) || page_has_siblings(page)) {
 err = DB_CORRUPTION;
 
 } else if (dict_index_is_clust(index)) {
@@ -1161,11 +1160,11 @@ btr_free_root_check(
 buf_block_dbg_add_level(block, SYNC_TREE_NODE);
 
 if (fil_page_index_page_check(block->frame)
-&& index_id == btr_page_get_index_id(block->frame)) {
+ && index_id == btr_page_get_index_id(block->frame)) {
 /* This should be a root page.
 It should not be possible to reassign the same
 index_id for some other index in the tablespace. */
-ut_ad(page_is_root(block->frame));
+ut_ad(!page_has_siblings(block->frame));
 } else {
 block = NULL;
 }
@@ -1358,7 +1357,8 @@ btr_free_but_not_root(
 ibool	finished;
 mtr_tmtr;
 
-ut_ad(page_is_root(block->frame));
+ut_ad(fil_page_index_page_check(block->frame));
+ut_ad(!page_has_siblings(block->frame));
 leaf_loop:
 mtr_start(&mtr);
 mtr_set_log_mode(&mtr, log_mode);
@@ -1430,7 +1430,6 @@ btr_free_if_exists(
 return;
 }
 
-ut_ad(page_is_root(root->frame));
 btr_free_but_not_root(root, mtr->get_log_mode());
 mtr->set_named_space(page_id.space());
 btr_free_root(root, mtr);
@@ -1453,8 +1452,6 @@ btr_free(
 page_id, page_size, RW_X_LATCH, &mtr);
 
 if (block) {
-ut_ad(page_is_root(block->frame));
-
 btr_free_but_not_root(block, MTR_LOG_NO_REDO);
 btr_free_root(block, &mtr);
 }
@@ -1614,12 +1611,17 @@ btr_page_reorganize_low(
 
 ut_ad(mtr_is_block_fix(mtr, block, MTR_MEMO_PAGE_X_FIX, index->table));
 btr_assert_not_corrupted(block, index);
+ut_ad(fil_page_index_page_check(block->frame));
+ut_ad(index->is_dummy
+ || block->page.id.space() == index->space);
+ut_ad(index->is_dummy
+ || block->page.id.page_no() != index->page
+ || !page_has_siblings(page));
 #ifdef UNIV_ZIP_DEBUG
 ut_a(!page_zip || page_zip_validate(page_zip, page, index));
 #endif /* UNIV_ZIP_DEBUG */
 data_size1 = page_get_data_size(page);
 max_ins_size1 = page_get_max_insert_size_after_reorganize(page, 1);
-
 /* Turn logging off */
 mtr_log_tlog_mode = mtr_set_log_mode(mtr, MTR_LOG_NONE);
 
@@ -1677,7 +1679,7 @@ btr_page_reorganize_low(
  || page_get_max_trx_id(page) == 0
  || (dict_index_is_sec_or_ibuf(index)
  ? page_is_leaf(temp_page)
- : page_is_root(temp_page)));
+ : block->page.id.page_no() == index->page));
 
 /* If innodb_log_compressed_pages is ON, page reorganize should log the
 compressed page image.*/
@@ -1894,6 +1896,8 @@ btr_page_empty(
 
 ut_ad(mtr_is_block_fix(mtr, block, MTR_MEMO_PAGE_X_FIX, index->table));
 ut_ad(page_zip == buf_block_get_page_zip(block));
+ut_ad(!index->is_dummy);
+ut_ad(index->space == block->page.id.space());
 #ifdef UNIV_ZIP_DEBUG
 ut_a(!page_zip || page_zip_validate(page_zip, page, index));
 #endif /* UNIV_ZIP_DEBUG */
@@ -1906,7 +1910,8 @@ btr_page_empty(
 /* Preserve PAGE_ROOT_AUTO_INC when creating a clustered index
 root page. */
 const ib_uint64_tautoinc
-= dict_index_is_clust(index) && page_is_root(page)
+= dict_index_is_clust(index)
+&& index->page == block->page.id.page_no()
 ? page_get_autoinc(page)
 : 0;
 
@@ -4225,6 +4230,8 @@ btr_discard_only_page_on_level(
 ulint	page_level = 0;
 trx_id_tmax_trx_id;
 
+ut_ad(!index->is_dummy);
+
 /* Save the PAGE_MAX_TRX_ID from the leaf page. */
 max_trx_id = page_get_max_trx_id(buf_block_get_frame(block));
 
@@ -4235,11 +4242,10 @@ btr_discard_only_page_on_level(
 
 ut_a(page_get_n_recs(page) == 1);
 ut_a(page_level == btr_page_get_level(page, mtr));
-ut_a(btr_page_get_prev(page, mtr) == FIL_NULL);
-ut_a(btr_page_get_next(page, mtr) == FIL_NULL);
-
-ut_ad(mtr_is_block_fix(
-mtr, block, MTR_MEMO_PAGE_X_FIX, index->table));
+ut_a(!page_has_siblings(page));
+ut_ad(fil_page_index_page_check(page));
+ut_ad(block->page.id.space() == index->space);
+ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
 btr_search_drop_page_hash_index(block);
 
 if (dict_index_is_spatial(index)) {
@@ -4265,6 +4271,7 @@ btr_discard_only_page_on_level(
 
 /* block is the root page, which must be empty, except
 for the node pointer to the (now discarded) block(s). */
+ut_ad(!page_has_siblings(block->frame));
 
 #ifdef UNIV_BTR_DEBUG
 if (!dict_index_is_ibuf(index)) {

storage/innobase/btr/btr0cur.cc

@@ -3,7 +3,7 @@
 Copyright (c) 1994, 2018, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2008, Google Inc.
 Copyright (c) 2012, Facebook Inc.
-Copyright (c) 2015, 2018, MariaDB Corporation.
+Copyright (c) 2015, 2019, MariaDB Corporation.
 
 Portions of this file contain modifications contributed and copyrighted by
 Google, Inc. Those modifications are gratefully acknowledged and are described
@@ -5113,14 +5113,14 @@ btr_cur_optimistic_delete_func(
 ut_ad(flags == 0 || flags == BTR_CREATE_FLAG);
 ut_ad(mtr_memo_contains(mtr, btr_cur_get_block(cursor),
 MTR_MEMO_PAGE_X_FIX));
-ut_ad(mtr_is_block_fix(mtr, btr_cur_get_block(cursor),
- MTR_MEMO_PAGE_X_FIX, cursor->index->table));
 ut_ad(mtr->is_named_space(cursor->index->space));
+ut_ad(!cursor->index->is_dummy);
 
 /* This is intended only for leaf page deletions */
 
 block = btr_cur_get_block(cursor);
 
+ut_ad(block->page.id.space() == cursor->index->space);
 ut_ad(page_is_leaf(buf_block_get_frame(block)));
 ut_ad(!dict_index_is_online_ddl(cursor->index)
  || dict_index_is_clust(cursor->index)
@@ -5242,8 +5242,10 @@ btr_cur_pessimistic_delete(
 ut_ad(mtr_memo_contains_flagged(mtr, dict_index_get_lock(index),
 MTR_MEMO_X_LOCK
 | MTR_MEMO_SX_LOCK));
-ut_ad(mtr_is_block_fix(mtr, block, MTR_MEMO_PAGE_X_FIX, index->table));
+ut_ad(mtr_memo_contains(mtr, block, MTR_MEMO_PAGE_X_FIX));
 ut_ad(mtr->is_named_space(index->space));
+ut_ad(!index->is_dummy);
+ut_ad(block->page.id.space() == index->space);
 
 if (!has_reserved_extents) {
 /* First reserve enough free space for the file segments

storage/innobase/btr/btr0defragment.cc

@@ -1,7 +1,7 @@
 /*****************************************************************************
 
 Copyright (C) 2013, 2014 Facebook, Inc. All Rights Reserved.
-Copyright (C) 2014, 2017, MariaDB Corporation.
+Copyright (C) 2014, 2019, 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
@@ -181,7 +181,8 @@ btr_defragment_add_index(
 return NULL;
 }
 
-ut_ad(page_is_root(page));
+ut_ad(fil_page_index_page_check(page));
+ut_ad(!page_has_siblings(page));
 
 if (page_is_leaf(page)) {
 // Index root is a leaf page, no need to defragment.

storage/innobase/include/page0page.h

@@ -1,6 +1,6 @@
 /*****************************************************************************
 Copyright (c) 1994, 2016, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2013, 2018, MariaDB Corporation.
+Copyright (c) 2013, 2019, 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
@@ -684,14 +684,19 @@ ulint
 page_rec_get_heap_no(
 /*=================*/
 const rec_t*rec);/*!< in: the physical record */
-/** Determine whether a page is an index root page.
+
+/** Determine whether a page has any siblings.
 @param[in]	page	page frame
-@return true if the page is a root page of an index */
-UNIV_INLINE
-bool
-page_is_root(
-const page_t*page)
-MY_ATTRIBUTE((warn_unused_result));
+@return true if the page has any siblings */
+inline bool page_has_siblings(const page_t* page)
+{
+compile_time_assert(!(FIL_PAGE_PREV % 8));
+compile_time_assert(FIL_PAGE_NEXT == FIL_PAGE_PREV + 4);
+compile_time_assert(FIL_NULL == 0xffffffff);
+return *reinterpret_cast<const uint64_t*>(page + FIL_PAGE_PREV)
+!= ~uint64_t(0);
+}
+
 /************************************************************//**
 Gets the pointer to the next record on the page.
 @return pointer to next record */

storage/innobase/include/page0page.ic

@@ -1,7 +1,7 @@
 /*****************************************************************************
 
 Copyright (c) 1994, 2015, Oracle and/or its affiliates. All Rights Reserved.
-Copyright (c) 2016, 2017, MariaDB Corporation.
+Copyright (c) 2016, 2019, 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
@@ -91,7 +91,8 @@ UNIV_INLINE
 ib_uint64_t
 page_get_autoinc(const page_t* page)
 {
-ut_ad(page_is_root(page));
+ut_ad(fil_page_index_page_check(page));
+ut_ad(!page_has_siblings(page));
 return(mach_read_from_8(PAGE_HEADER + PAGE_ROOT_AUTO_INC + page));
 }
 
@@ -277,31 +278,6 @@ page_rec_get_heap_no(
 }
 }
 
-/** Determine whether a page is an index root page.
-@param[in]	page	page frame
-@return true if the page is a root page of an index */
-UNIV_INLINE
-bool
-page_is_root(
-const page_t*	page)
-{
-#if FIL_PAGE_PREV % 8
-# error FIL_PAGE_PREV must be 64-bit aligned
-#endif
-#if FIL_PAGE_NEXT != FIL_PAGE_PREV + 4
-# error FIL_PAGE_NEXT must be adjacent to FIL_PAGE_PREV
-#endif
-#if FIL_NULL != 0xffffffff
-# error FIL_NULL != 0xffffffff
-#endif
-/* Check that this is an index page and both the PREV and NEXT
-pointers are FIL_NULL, because the root page does not have any
-siblings. */
-return(fil_page_index_page_check(page)
- && *reinterpret_cast<const ib_uint64_t*>(page + FIL_PAGE_PREV)
- == IB_UINT64_MAX);
-}
-
 /** Determine whether an index page record is a user record.
 @param[in]	rec	record in an index page
 @return true if a user record */

storage/innobase/page/page0cur.cc

@@ -2,7 +2,7 @@
 
 Copyright (c) 1994, 2016, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2012, Facebook Inc.
-Copyright (c) 2018, MariaDB Corporation.
+Copyright (c) 2018, 2019, 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
@@ -1990,6 +1990,8 @@ page_parse_copy_rec_list_to_created_page(
 return(rec_end);
 }
 
+ut_ad(fil_page_index_page_check(block->frame));
+
 while (ptr < rec_end) {
 ptr = page_cur_parse_insert_rec(TRUE, ptr, end_ptr,
 block, index, mtr);
@@ -2045,6 +2047,7 @@ page_copy_rec_list_end_to_created_page(
 ut_ad(page_dir_get_n_heap(new_page) == PAGE_HEAP_NO_USER_LOW);
 ut_ad(page_align(rec) != new_page);
 ut_ad(page_rec_is_comp(rec) == page_is_comp(new_page));
+ut_ad(fil_page_index_page_check(new_page));
 
 if (page_rec_is_infimum(rec)) {
 

storage/innobase/page/page0page.cc

@@ -2,7 +2,7 @@
 
 Copyright (c) 1994, 2016, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2012, Facebook Inc.
-Copyright (c) 2017, 2018, MariaDB Corporation.
+Copyright (c) 2017, 2019, 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
@@ -526,6 +526,8 @@ page_create_empty(
 page_zip_des_t*	page_zip= buf_block_get_page_zip(block);
 
 ut_ad(fil_page_index_page_check(page));
+ut_ad(!index->is_dummy);
+ut_ad(block->page.id.space() == index->space);
 
 /* Multiple transactions cannot simultaneously operate on the
 same temp-table in parallel.
@@ -536,7 +538,7 @@ page_create_empty(
  && page_is_leaf(page)) {
 max_trx_id = page_get_max_trx_id(page);
 ut_ad(max_trx_id);
-} else if (page_is_root(page)) {
+} else if (block->page.id.page_no() == index->page) {
 /* Preserve PAGE_ROOT_AUTO_INC. */
 max_trx_id = page_get_max_trx_id(page);
 } else {

storage/innobase/page/page0zip.cc

@@ -2,7 +2,7 @@
 
 Copyright (c) 2005, 2016, Oracle and/or its affiliates. All Rights Reserved.
 Copyright (c) 2012, Facebook Inc.
-Copyright (c) 2014, 2018, MariaDB Corporation.
+Copyright (c) 2014, 2019, 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
@@ -4727,7 +4727,7 @@ page_zip_reorganize(
 clustered index root pages. */
 ut_ad(page_get_max_trx_id(page) == 0
  || (dict_index_is_clust(index)
- ? page_is_root(temp_page)
+ ? !page_has_siblings(temp_page)
  : page_is_leaf(temp_page)));
 
 /* Restore logging. */

storage/innobase/row/row0import.cc

@@ -668,7 +668,7 @@ dberr_t FetchIndexRootPages::operator()(buf_block_t* block) UNIV_NOTHROW
 return set_current_xdes(block->page.id.page_no(), page);
 } else if (fil_page_index_page_check(page)
  && !is_free(block->page.id.page_no())
- && page_is_root(page)) {
+ && !page_has_siblings(page)) {
 
 index_id_tid = btr_page_get_index_id(page);
 
@@ -1834,7 +1834,7 @@ PageConverter::update_index_page(
 page, m_page_zip_ptr, m_index->m_srv_index->id, 0);
 
 if (dict_index_is_clust(m_index->m_srv_index)) {
-if (page_is_root(page)) {
+if (block->page.id.page_no() == m_index->m_srv_index->page) {
 /* Preserve the PAGE_ROOT_AUTO_INC. */
 } else {
 /* Clear PAGE_MAX_TRX_ID so that it can be
@@ -1854,7 +1854,7 @@ PageConverter::update_index_page(
 if (page_is_empty(page)) {
 
 /* Only a root page can be empty. */
-if (!page_is_root(page)) {
+if (page_has_siblings(page)) {
 // TODO: We should relax this and skip secondary
 // indexes. Mark them as corrupt because they can
 // always be rebuilt.