Received-SPF: pass (google.com: domain of linux-ext4-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Tue, 16 Aug 2022 13:21:24 +0200
From:   Jan Kara <jack@suse.cz>
To:     Eric Biggers <ebiggers@kernel.org>
Cc:     Lukas Czerner <lczerner@redhat.com>, linux-ext4@vger.kernel.org,
        tytso@mit.edu, jlayton@kernel.org, jack@suse.cz,
        linux-fsdevel@vger.kernel.org, david@fromorbit.com,
        Christoph Hellwig <hch@infradead.org>
Subject: Re: [PATCH v3 2/3] fs: record I_DIRTY_TIME even if inode already has
 I_DIRTY_INODE
Message-ID: <20220816112124.taqvli527475gwv4@quack3>
References: <20220812123727.46397-1-lczerner@redhat.com>
 <20220812123727.46397-2-lczerner@redhat.com>
 <YvaYC+LRFqQJT0U9@sol.localdomain>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <YvaYC+LRFqQJT0U9@sol.localdomain>
Precedence: bulk

On Fri 12-08-22 11:12:27, Eric Biggers wrote:
> On Fri, Aug 12, 2022 at 02:37:26PM +0200, Lukas Czerner wrote:
> > diff --git a/Documentation/filesystems/vfs.rst b/Documentation/filesystems/vfs.rst
> > index 6cd6953e175b..5d72b6ba4e63 100644
> > --- a/Documentation/filesystems/vfs.rst
> > +++ b/Documentation/filesystems/vfs.rst
> > @@ -274,6 +274,8 @@ or bottom half).
> >  	This is specifically for the inode itself being marked dirty,
> >  	not its data.  If the update needs to be persisted by fdatasync(),
> >  	then I_DIRTY_DATASYNC will be set in the flags argument.
> > +	If the inode has dirty timestamp and lazytime is enabled
> > +	I_DIRTY_TIME will be set in the flags.
> 
> The new sentence is not always true, since with this patch if
> __mark_inode_dirty(I_DIRTY_INODE) is called twice on an inode that has
> I_DIRTY_TIME, the second call will no longer include I_DIRTY_TIME -- even though
> the inode still has dirty timestamps.  Please be super clear about what the
> flags actually mean -- I'm still struggling to understand this patch...

Let me chime in here because I was the one who suggested the solution to
Lukas. There are two different things (which is why this is confusing I
guess):

1) I_DIRTY_TIME in the inode->i_state should mean: struct inode has times
updated after we last called ->dirty_inode() callback. Hence
inode_is_dirtytime_only() as well as the chunk:
                /* I_DIRTY_INODE supersedes I_DIRTY_TIME. */
                flags &= ~I_DIRTY_TIME;
you mention in the previous email are compatible with this meaning AFAICT.

2) I_DIRTY_TIME flag passed to ->dirty_inode() callback. This is admittedly
bit of a hack. Currently XFS relies on the fact that the only time its
->dirty_inode() callback needs to do anything is when VFS decides it is
time to writeback timestamps and XFS detects this situation by checking for
I_DIRTY_TIME in inode->i_state. Now to fix the race, we need to first clear
I_DIRTY_TIME in inode->i_state and only then call the ->dirty_inode()
callback (otherwise timestamp update can get lost). So the solution I've
suggested was to propagate the information "timestamp update needed" to XFS
through I_DIRTY_TIME in flags passed to ->dirty_inode().

I hope things are clearer now.

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