On 2020/1/18 0:24, Darrick J. Wong wrote:
Does the problem go away if you apply[1]? If I understand the race correctly, marking the extents unwritten and leaving them that way until after we've written the disk should eliminate the exposure vector...?:)
Thank you for your response. The patch [1] fixed the stale data exposure problem which was tested by generic/042. I'm afarid it's completely a diffrent case for generic/418. In generic/418, direct write wrote some data to disk successfully, while buffer read left the correspond page uptodate with 0 content in pagecache. Thanks! Yu Kuai
