Received-SPF: pass (google.com: domain of linux-ext4-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Tue, 30 Mar 2021 13:15:08 -0400
From:   "Theodore Ts'o" <tytso@mit.edu>
To:     "Darrick J. Wong" <djwong@kernel.org>
Cc:     harshad shirwadkar <harshadshirwadkar@gmail.com>,
        Leah Rumancik <leah.rumancik@gmail.com>,
        Ext4 Developers List <linux-ext4@vger.kernel.org>
Subject: Re: [PATCH 2/2] ext4: add ioctl EXT4_FLUSH_JOURNAL
Message-ID: <YGNcnJpqUqrwbMAH@mit.edu>
References: <20210325181220.1118705-1-leah.rumancik@gmail.com>
 <20210325181220.1118705-2-leah.rumancik@gmail.com>
 <20210326012146.GB22091@magnolia>
 <YGHtC4vEWcRervi1@google.com>
 <CAD+ocbx-DZprEvzSQ2+rVFz6CPJy0iYbG60SGZHpi+ZX8ecqhg@mail.gmail.com>
 <20210330163223.GD22091@magnolia>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20210330163223.GD22091@magnolia>
Precedence: bulk

On Tue, Mar 30, 2021 at 09:32:23AM -0700, Darrick J. Wong wrote:
> Why not make discarding the journal part of FITRIM then?

Unfortunately, the fstrim_range structure doesn't have a place for a
flags field, and FITRIM works by specifying a range of LBA's:

struct fstrim_range {
	__u64 start;
	__u64 len;
	__u64 minlen;
};

I suppose we could do something where some combination of start/len
means "also checkpoint and discard the journal", but that seems rather
kludgy.

> It occurred to me overnight that another way to look at this ioctl
> proposal is that it checkpoints the filesystem and has a flag to discard
> the journal blocks too.  Given that we're now only two days away from
> my traditional bootfs[1] drum-banging day, and there's real user
> demand[2] for bootloaders to be able to force a journal checkpoint,

How about if we have an ioctl which is "checkpoint journal", which can
be file system independent (e.g., defined in include/uapi/linux/fs.h)
which takes a u32 flags field, where we define a flag bit to mean
"also discard the unused part of the journal after the checkpoint"?

It seems that would also solve your bootfs() use case.

      	      	  	      	     - Ted