On Wed, Sep 23, 2020 at 02:24:35PM +0200, Jan Kara wrote: > On Wed 23-09-20 06:29:12, 常凤楠 wrote: > > The attachment is new patch. > > Thanks! > > > I have fix the logic in JBD2_REVOKE_BLOCK and JBD2_COMMIT_BLOCK case. > > If the revoke block is the first block after valid transaction, I set the > > flag like descriptor block ,and check it in commit block. If the commit > > block is the first block after valid transaction, I use ri_commit_block > > to judge whether this commit block is next to another commit block, if so > > it is illegal. I did't use time to judge commit block, because of the > > possibility of time calibration, I think use ri_commit_block is more > > reliable. > > If the time is unreliable, your logic with ri_commit_block will detect only > a small minority of the cases. The majority of cases will fail with > checksum error. So I still don't think the ri_commit_block logic is really > worth the additional code. Other than that the current version looks OK to > me. So please do next submission with full changelog, Signed-off-by line, > and the coding style issues fixed (you can use scripts/checkpatch.pl for > patch verification). Thanks! Hi changfengnan, Have you had a chance to send a revised patch with the changes requested by Jan? Thanks, - Ted
