Date:   Thu, 16 Dec 2021 11:23:50 +0000
From:   =?iso-8859-1?Q?Lu=EDs?= Henriques <lhenriques@suse.de>
To:     Lukas Czerner <lczerner@redhat.com>
Cc:     Jan Kara <jack@suse.cz>, "Darrick J. Wong" <djwong@kernel.org>,
        Theodore Ts'o <tytso@mit.edu>,
        Andreas Dilger <adilger.kernel@dilger.ca>,
        linux-ext4@vger.kernel.org, linux-kernel@vger.kernel.org,
        Jeroen van Wolffelaar <jeroen@wolffelaar.nl>
Subject: Re: [PATCH v2] ext4: set csum seed in tmp inode while migrating to
 extents
Message-ID: <YbshxgbAnt7rupQG@suse.de>
References: <20211214175058.19511-1-lhenriques@suse.de>
 <20211215004945.GD69182@magnolia>
 <20211215112852.GM14044@quack2.suse.cz>
 <20211215141237.lrymhbebgjunh4n2@work>
MIME-Version: 1.0
Content-Type: text/plain; charset=iso-8859-1
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <20211215141237.lrymhbebgjunh4n2@work>
Precedence: bulk

On Wed, Dec 15, 2021 at 03:12:37PM +0100, Lukas Czerner wrote:
> On Wed, Dec 15, 2021 at 12:28:52PM +0100, Jan Kara wrote:
> > On Tue 14-12-21 16:49:45, Darrick J. Wong wrote:
> > > On Tue, Dec 14, 2021 at 05:50:58PM +0000, Lu?s Henriques wrote:
> > > > When migrating to extents, the temporary inode will have it's own checksum
> > > > seed.  This means that, when swapping the inodes data, the inode checksums
> > > > will be incorrect.
> > > > 
> > > > This can be fixed by recalculating the extents checksums again.  Or simply
> > > > by copying the seed into the temporary inode.
> > > > 
> > > > Link: https://bugzilla.kernel.org/show_bug.cgi?id=213357
> > > > Reported-by: Jeroen van Wolffelaar <jeroen@wolffelaar.nl>
> > > > Signed-off-by: Lu?s Henriques <lhenriques@suse.de>
> > > > ---
> > > >  fs/ext4/migrate.c | 12 +++++++++++-
> > > >  1 file changed, 11 insertions(+), 1 deletion(-)
> > > > 
> > > > changes since v1:
> > > > 
> > > > * Dropped tmp_ei variable
> > > > * ->i_csum_seed is now initialised immediately after tmp_inode is created
> > > > * New comment about the seed initialization and stating that recovery
> > > >   needs to be fixed.
> > > > 
> > > > Cheers,
> > > > --
> > > > Lu?s
> > > > 
> > > > diff --git a/fs/ext4/migrate.c b/fs/ext4/migrate.c
> > > > index 7e0b4f81c6c0..36dfc88ce05b 100644
> > > > --- a/fs/ext4/migrate.c
> > > > +++ b/fs/ext4/migrate.c
> > > > @@ -459,6 +459,17 @@ int ext4_ext_migrate(struct inode *inode)
> > > >  		ext4_journal_stop(handle);
> > > >  		goto out_unlock;
> > > >  	}
> > > > +	/*
> > > > +	 * Use the correct seed for checksum (i.e. the seed from 'inode').  This
> > > > +	 * is so that the metadata blocks will have the correct checksum after
> > > > +	 * the migration.
> > > > +	 *
> > > > +	 * Note however that, if a crash occurs during the migration process,
> > > > +	 * the recovery process is broken because the tmp_inode checksums will
> > > > +	 * be wrong and the orphans cleanup will fail.
> > > 
> > > ...and then what does the user do?
> > 
> > Run fsck of course! And then recover from backups :) I know this is sad but
> > the situation is that our migration code just is not crash-safe (if we
> > crash we are going to free blocks that are still used by the migrated
> > inode) and Luis makes it work in case we do not crash (which should be
> > hopefully more common) and documents it does not work in case we crash.
> > So overall I'd call it a win.
> > 
> > But maybe we should just remove this online-migration functionality
> > completely from the kernel? That would be also a fine solution for me. I
> > was thinking whether we could somehow make the inode migration crash-safe
> > but I didn't think of anything which would not require on-disk format
> > change...
> 
> Since this is not something that anyone can honestly recommend doing
> without a prior backup and a word of warning I personaly would be in favor
> of removing it.

BTW, in case migration is kept in the kernel (even with the broken
recovery), I think it's worth turning this bug reproducer into an ext4
fstest.  I was planning to do so, but I'd rather wait to see if the effort
is worthwhile (i.e. if migration is kept or not).

Cheers,
--
Lu?s