Date:   Mon, 14 Dec 2020 10:44:29 -0800
From:   harshad shirwadkar <harshadshirwadkar@...il.com>
To:     Haotian Li <lihaotian9@...wei.com>
Cc:     Ext4 Developers List <linux-ext4@...r.kernel.org>,
        "Theodore Y. Ts'o" <tytso@....edu>,
        "liuzhiqiang (I)" <liuzhiqiang26@...wei.com>,
        linfeilong <linfeilong@...wei.com>, tytso@...m.mit.edu,
        liangyun2@...wei.com
Subject: Re: [PATCH] e2fsck: Avoid changes on recovery flags when
 jbd2_journal_recover() failed

Hi Haotian,

Yeah perhaps these are the only recoverable errors. I also think that
we can't surely say that these errors are recoverable always. That's
because in some setups, these errors may still be unrecoverable (for
example, if the machine is running under low memory). I still feel
that we should ask the user about whether they want to continue or
not. The reason is that firstly if we don't allow running e2fsck in
these cases, I wonder what would the user do with their file system -
they can't mount / can't run fsck, right? Secondly, not doing that
would be a regression. I wonder if some setups would have chosen to
ignore journal recovery if there are errors during journal recovery
and with this fix they may start seeing that their file systems aren't
getting repaired.

I'm wondering if you saw any a situation in your setup where exiting
e2fsck helped? If possible, could you share what kind of errors were
seen in journal recovery and what was the expected behavior? Maybe
that would help us decide on the right behavior.

Thanks,
Harshad

On Sun, Dec 13, 2020 at 5:27 PM Haotian Li <lihaotian9@...wei.com> wrote:
>
> Hi Harshad,
>
> Thanks for your review. I think you are right, so I try to find
> all the recoverable err_codes in journal recovery. But I have no
> idea to distinguish all the err_codes. Only the following three
> err_codes I think may be recoverable. -ENOMEM,EXT2_ET_NO_MEMORY
> ,-EIO. In these cases, I think we probably don't need ask user if
> they want to continue or not, only tell them why journal recover
> failed and exit instead. Because, the reason cause these cases
> may not disk errors, we need try to avoid the changes on the disk.
> What do you think?
>
> Thanks,
> Haotian
>
> 在 2020/12/12 6:07, harshad shirwadkar 写道:
> > Hi Haotian,
> >
> > Thanks for your patch. I noticed that the following test fails:
> >
> > $ make -j 64
> > ...
> > 365 tests succeeded     1 tests failed
> > Tests failed: j_corrupt_revoke_rcount
> > make: *** [Makefile:397: test_post] Error 1
> >
> > This test fails because the test expects e2fsck to continue even if
> > the journal superblock is corrupt and with your patch e2fsck exits
> > immediately. This brings up a higher level question - if we abort on
> > errors when recovery fails during fsck, how would that problem get
> > fixed if we don't run fsck? In this particular example, the journal
> > superblock is corrupt and that is an unrecoverable error. I wonder if
> > instead we should check for certain specific transient errors such as
> > -ENOMEM and only then exit? I suspect even in those cases we probably
> > should ask the user if they would like to continue or not. What do you
> > think?
> >
> > Thanks,
> > Harshad
> >
> >
> > On Fri, Dec 11, 2020 at 4:19 AM Haotian Li <lihaotian9@...wei.com> wrote:
> >>
> >> jbd2_journal_revocer() may fail when some error occers
> >> such as ENOMEM. However, jsb->s_start is still cleared
> >> by func e2fsck_journal_release(). This may break
> >> consistency between metadata and data in disk. Sometimes,
> >> failure in jbd2_journal_revocer() is temporary but retry
> >> e2fsck will skip the journal recovery when the temporary
> >> problem is fixed.
> >>
> >> To fix this case, we use "fatal_error" instead "goto errout"
> >> when recover journal failed. We think if journal recovery
> >> fails, we need send error message to user and reserve the
> >> recovery flags to recover the journal when try e2fsck again.
> >>
> >> Reported-by: Liangyun <liangyun2@...wei.com>
> >> Signed-off-by: Haotian Li <lihaotian9@...wei.com>
> >> Signed-off-by: Zhiqiang Liu <liuzhiqiang26@...wei.com>
> >> ---
> >>  e2fsck/journal.c | 9 +++++++--
> >>  1 file changed, 7 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/e2fsck/journal.c b/e2fsck/journal.c
> >> index 7d9f1b40..546beafd 100644
> >> --- a/e2fsck/journal.c
> >> +++ b/e2fsck/journal.c
> >> @@ -952,8 +952,13 @@ static errcode_t recover_ext3_journal(e2fsck_t ctx)
> >>                 goto errout;
> >>
> >>         retval = -jbd2_journal_recover(journal);
> >> -       if (retval)
> >> -               goto errout;
> >> +       if (retval && retval != EFSBADCRC && retval != EFSCORRUPTED) {
> >> +               ctx->fs->flags &= ~EXT2_FLAG_VALID;
> >> +               com_err(ctx->program_name, 0,
> >> +                                       _("Journal recovery failed "
> >> +                                         "on %s\n"), ctx->device_name);
> >> +               fatal_error(ctx, 0);
> >> +       }
> >>
> >>         if (journal->j_failed_commit) {
> >>                 pctx.ino = journal->j_failed_commit;
> >> --
> >> 2.19.1
> >>
> > .
> >

Powered by blists - more mailing lists