[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <238e40d0-606f-bb9e-c18c-542df87c1e31@huawei.com>
Date: Wed, 21 Aug 2024 09:44:40 +0800
From: Zhihao Cheng <chengzhihao1@...wei.com>
To: yangerkun <yangerkun@...weicloud.com>, <tytso@....edu>,
<adilger.kernel@...ger.ca>, <jack@...e.cz>
CC: <linux-ext4@...r.kernel.org>, <yangerkun@...wei.com>
Subject: Re: [PATCH 2/2] ext4: dax: keep orphan list before truncate overflow
allocated blocks
在 2024/8/20 22:06, yangerkun 写道:
> From: yangerkun <yangerkun@...wei.com>
>
> Any extended write for ext4 requires the inode to be placed on the
> orphan list before the actual write. In addition, the inode can be
> actually removed from the orphan list only after all writes are
> completed. Otherwise, those overcommitted blocks (If the allocated
> blocks are not written due to certain reasons, the inode size does not
> exceed the offset of these blocks) The leak status is always retained,
> and fsck reports an alarm for this scenario.
>
> Currently, the dio and buffer IO comply with this logic. However, the
> dax write will removed the inode from orphan list since
> ext4_handle_inode_extension is unconditionally called during extend
> write. Fix it with this patch. We open the code from
> ext4_handle_inode_extension since we want to keep the blocks valid
> has been allocated and write success.
>
> Signed-off-by: yangerkun <yangerkun@...wei.com>
> ---
> fs/ext4/file.c | 35 +++++++++++++++++++++++++++++++----
> 1 file changed, 31 insertions(+), 4 deletions(-)
Reviewed-by: Zhihao Cheng <chengzhihao1@...wei.com>
>
> diff --git a/fs/ext4/file.c b/fs/ext4/file.c
> index be061bb64067..fd8597eef75e 100644
> --- a/fs/ext4/file.c
> +++ b/fs/ext4/file.c
> @@ -628,11 +628,12 @@ static ssize_t ext4_dio_write_iter(struct kiocb *iocb, struct iov_iter *from)
> static ssize_t
> ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
> {
> - ssize_t ret;
> + ssize_t ret, written;
> size_t count;
> loff_t offset;
> handle_t *handle;
> bool extend = false;
> + bool need_trunc = true;
> struct inode *inode = file_inode(iocb->ki_filp);
>
> if (iocb->ki_flags & IOCB_NOWAIT) {
> @@ -668,10 +669,36 @@ ext4_dax_write_iter(struct kiocb *iocb, struct iov_iter *from)
>
> ret = dax_iomap_rw(iocb, from, &ext4_iomap_ops);
>
> - if (extend) {
> - ret = ext4_handle_inode_extension(inode, offset, ret);
> - ext4_inode_extension_cleanup(inode, ret < (ssize_t)count);
> + if (!extend)
> + goto out;
> +
> + if (ret <= 0)
> + goto err_trunc;
> +
> + written = ret;
> + handle = ext4_journal_start(inode, EXT4_HT_INODE, 2);
> + if (IS_ERR(handle)) {
> + ret = PTR_ERR(handle);
> + goto err_trunc;
> }
> +
> + if (ext4_update_inode_size(inode, offset + written)) {
> + ret = ext4_mark_inode_dirty(handle, inode);
> + if (unlikely(ret)) {
> + ext4_journal_stop(handle);
> + goto err_trunc;
> + }
> + }
> +
> + if (written == count)
> + need_trunc = false;
> +
> + if (inode->i_nlink)
> + ext4_orphan_del(handle, inode);
> + ext4_journal_stop(handle);
> + ret = written;
> +err_trunc:
> + ext4_inode_extension_cleanup(inode, need_trunc);
> out:
> inode_unlock(inode);
> if (ret > 0)
>
Powered by blists - more mailing lists