Re: [PATCH 4/4] jbd: fix error handling for checkpoint io (rebased)

[Jan Kara <jack@suse.cz>]

> >  
> > Index: linux-2.6.26-rc2/fs/ext3/super.c
> > ===================================================================
> > --- linux-2.6.26-rc2.orig/fs/ext3/super.c
> > +++ linux-2.6.26-rc2/fs/ext3/super.c
> > @@ -395,7 +395,10 @@ static void ext3_put_super (struct super
> >  	ext3_xattr_put_super(sb);
> >  	journal_destroy(sbi->s_journal);
> >  	if (!(sb->s_flags & MS_RDONLY)) {
> > -		EXT3_CLEAR_INCOMPAT_FEATURE(sb, EXT3_FEATURE_INCOMPAT_RECOVER);
> > +		if (!is_checkpoint_aborted(sbi->s_journal)) {
> > +			EXT3_CLEAR_INCOMPAT_FEATURE(sb,
> > +				EXT3_FEATURE_INCOMPAT_RECOVER);
> > +		}
> >  		es->s_state = cpu_to_le16(sbi->s_mount_state);
> >  		BUFFER_TRACE(sbi->s_sbh, "marking dirty");
> >  		mark_buffer_dirty(sbi->s_sbh);
> 
> Is this bit here really needed?  If we abort the journal the fs will be mounted
> read only and we should never get in here.  Is there a case where we could abort
> the journal and not be flipped to being read-only?  If there is such a case I
> would think that we should fix that by making the fs read-only, and not have
> this check.
  Actually, journal_abort() (which could be called from journal_destroy())
does nothing to the filesystem as such. So at this moment, ext3 can still
happily think everything is OK. We only detect aborted journal in
ext3_journal_start_sb() and call ext3_abort() in that case, which does all
that is needed...

									Honza
-- 
Jan Kara <jack@suse.cz>
SUSE Labs, CR