From: Yongqiang Yang Subject: Re: [PATCH] ext4: ext4_split_extent shoult take care about extent zeroout Date: Thu, 31 Jan 2013 21:35:10 +0800 Message-ID: References: <1359617098-18451-1-git-send-email-dmonakhov@openvz.org> Mime-Version: 1.0 Content-Type: text/plain; charset=ISO-8859-1 Cc: Ext4 Developers List , "Theodore Ts'o" , Jan Kara To: Dmitry Monakhov Return-path: Received: from mail-ea0-f182.google.com ([209.85.215.182]:34213 "EHLO mail-ea0-f182.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754600Ab3AaNfc (ORCPT ); Thu, 31 Jan 2013 08:35:32 -0500 Received: by mail-ea0-f182.google.com with SMTP id a12so1271826eaa.13 for ; Thu, 31 Jan 2013 05:35:30 -0800 (PST) In-Reply-To: <1359617098-18451-1-git-send-email-dmonakhov@openvz.org> Sender: linux-ext4-owner@vger.kernel.org List-ID: Hi Dmitry, I am a little confusing. ext4_convert_unwritten_extents_endio() is used to convert uninit extent to init, it finds out uninit extent is the expected case. Am I missing something ? I will look into the code this weekend. In your description, extent [1000,128] is split to [1000,20] and [1020, 108], right? Thanks, Yongqiang. On Thu, Jan 31, 2013 at 3:24 PM, Dmitry Monakhov wrote: > We have to update extent's state after first ext4_split_extent_at otherwise this result > in following trace: > ->ext4_ext_handle_uninitialized_extents (ex=[1000:20:uninit], lblock 1000, max_blocks 10) > ->ext4_split_extent_at(ex=[1000,128], lblk 10010) /// First split > ->ext4_ext_split() -> ENOSPC > ->ext4_ext_zeroout > ->ext4_ext_dirty -> ex=[1000:20:init] > ->ext4_split_extent_at(ex=[1000,128], lblk 10000) /// Second split > if(split == ee_block) > if (split_flag & EXT4_EXT_MARK_UNINIT2) > ext4_ext_mark_uninitialized(ex); ex=[1000:20:uninit] /// The bug! > ->ext4_ext_dirty ->ex=[1000:20:uninit] > > At the end ext4_convert_unwritten_extents_endio() will findout large uninitialized > extent. > > TESTCASE: https://github.com/dmonakhov/xfstests/commit/1a1c4f337d4d198803436c63a56625b1a78d8a5e > > Signed-off-by: Dmitry Monakhov > --- > fs/ext4/extents.c | 14 ++++++++++---- > 1 files changed, 10 insertions(+), 4 deletions(-) > > diff --git a/fs/ext4/extents.c b/fs/ext4/extents.c > index 97cac01..7a3f679 100644 > --- a/fs/ext4/extents.c > +++ b/fs/ext4/extents.c > @@ -3091,18 +3091,24 @@ static int ext4_split_extent(handle_t *handle, > if (err) > goto out; > } > - > + /* Update path is required because previous ext4_split_extent_at() may > + * result in split of original leaf or extent zeroout. > + */ > ext4_ext_drop_refs(path); > path = ext4_ext_find_extent(inode, map->m_lblk, path); > if (IS_ERR(path)) > return PTR_ERR(path); > + depth = ext_depth(inode); > + ex = path[depth].p_ext; > + uninitialized = ext4_ext_is_uninitialized(ex); > > if (map->m_lblk >= ee_block) { > split_flag1 = split_flag & EXT4_EXT_MAY_ZEROOUT; > - if (uninitialized) > + if (uninitialized) { > split_flag1 |= EXT4_EXT_MARK_UNINIT1; > - if (split_flag & EXT4_EXT_MARK_UNINIT2) > - split_flag1 |= EXT4_EXT_MARK_UNINIT2; > + if (split_flag & EXT4_EXT_MARK_UNINIT2) > + split_flag1 |= EXT4_EXT_MARK_UNINIT2; > + } > err = ext4_split_extent_at(handle, inode, path, > map->m_lblk, split_flag1, flags); > if (err) > -- > 1.7.1 > -- Best Wishes Yongqiang Yang