diff options
| author | Marko Mäkelä <marko.makela@mariadb.com> | 2020-01-23 16:14:28 +0200 |
|---|---|---|
| committer | Marko Mäkelä <marko.makela@mariadb.com> | 2020-01-23 16:25:46 +0200 |
| commit | 1d12bff42c7dd73b2a0158d6780267467b9e0871 (patch) | |
| tree | 0145c0d7a84eee3147118759179eac6f4b52245c /storage | |
| parent | 7c166e68aaf51bd9c1b3152ba9674fcc261f3d84 (diff) | |
| download | mariadb-git-1d12bff42c7dd73b2a0158d6780267467b9e0871.tar.gz | |
MDEV-20775: page_zip_validate() failure due to AUTO_INCREMENT
cmake -DWITH_INNODB_EXTRA_DEBUG:BOOL=ON
was broken ever since commit 8777458a6eb73ac1d7d864ebac390ea7039e21c1
(MDEV-6076 Persistent AUTO_INCREMENT for InnoDB).
There is a race condition between page reads that call
page_zip_validate() (while holding clustered index root page S-latch)
and writes that update PAGE_ROOT_AUTO_INC
(with buf_block_t::lock SX-latch, compatible with S-latch).
page_zip_validate_low(): Skip the PAGE_ROOT_AUTO_INC field on
clustered index root pages in order to avoid false positives.
Diffstat (limited to 'storage')
| -rw-r--r-- | storage/innobase/page/page0zip.cc | 16 |
1 files changed, 14 insertions, 2 deletions
diff --git a/storage/innobase/page/page0zip.cc b/storage/innobase/page/page0zip.cc index 1c0c8f75f78..7a017913dfe 100644 --- a/storage/innobase/page/page0zip.cc +++ b/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, 2019, MariaDB Corporation. +Copyright (c) 2014, 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 @@ -3336,7 +3336,19 @@ page_zip_validate_low( FIL_PAGE_LSN - FIL_PAGE_PREV) || memcmp(page_zip->data + FIL_PAGE_TYPE, page + FIL_PAGE_TYPE, 2) || memcmp(page_zip->data + FIL_PAGE_DATA, page + FIL_PAGE_DATA, - PAGE_DATA - FIL_PAGE_DATA)) { + PAGE_ROOT_AUTO_INC) + /* The PAGE_ROOT_AUTO_INC can be updated while holding an SX-latch + on the clustered index root page (page number 3 in .ibd files). + That allows concurrent readers (holding buf_block_t::lock S-latch). + Because we do not know what type of a latch our caller is holding, + we will ignore the field on clustered index root pages in order + to avoid false positives. */ + || (page_get_page_no(page) != 3/* clustered index root page */ + && memcmp(&page_zip->data[FIL_PAGE_DATA + PAGE_ROOT_AUTO_INC], + &page[FIL_PAGE_DATA + PAGE_ROOT_AUTO_INC], 8)) + || memcmp(&page_zip->data[FIL_PAGE_DATA + PAGE_HEADER_PRIV_END], + &page[FIL_PAGE_DATA + PAGE_HEADER_PRIV_END], + PAGE_DATA - FIL_PAGE_DATA - PAGE_HEADER_PRIV_END)) { page_zip_fail(("page_zip_validate: page header\n")); page_zip_hexdump(page_zip, sizeof *page_zip); page_zip_hexdump(page_zip->data, page_zip_get_size(page_zip)); |