Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:3 as permitted sender) client-ip=2620:137:e000::3:3;
Date:   Tue, 12 Dec 2023 07:19:02 -0800
From:   Christoph Hellwig <hch@infradead.org>
To:     Theodore Ts'o <tytso@mit.edu>
Cc:     Christoph Hellwig <hch@infradead.org>,
        John Garry <john.g.garry@oracle.com>,
        Ojaswin Mujoo <ojaswin@linux.ibm.com>,
        linux-ext4@vger.kernel.org, Ritesh Harjani <ritesh.list@gmail.com>,
        linux-kernel@vger.kernel.org,
        "Darrick J . Wong" <djwong@kernel.org>,
        linux-block@vger.kernel.org, linux-xfs@vger.kernel.org,
        linux-fsdevel@vger.kernel.org, dchinner@redhat.com
Subject: Re: [RFC 0/7] ext4: Allocator changes for atomic write support with
 DIO
Message-ID: <ZXh55vLzrs9VTGHc@infradead.org>
References: <cover.1701339358.git.ojaswin@linux.ibm.com>
 <8c06c139-f994-442b-925e-e177ef2c5adb@oracle.com>
 <ZW3WZ6prrdsPc55Z@li-bb2b2a4c-3307-11b2-a85c-8fa5c3a69313.ibm.com>
 <de90e79b-83f2-428f-bac6-0754708aa4a8@oracle.com>
 <ZXbqVs0TdoDcJ352@li-bb2b2a4c-3307-11b2-a85c-8fa5c3a69313.ibm.com>
 <c4cf3924-f67d-4f04-8460-054dbad70b93@oracle.com>
 <ZXhb0tKFvAge/GWf@infradead.org>
 <20231212151613.GA142380@mit.edu>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20231212151613.GA142380@mit.edu>
Precedence: bulk

On Tue, Dec 12, 2023 at 10:16:13AM -0500, Theodore Ts'o wrote:
> On Tue, Dec 12, 2023 at 05:10:42AM -0800, Christoph Hellwig wrote:
> > On Tue, Dec 12, 2023 at 07:46:51AM +0000, John Garry wrote:
> > > It is assumed that the user will fallocate/dd the complete file before
> > > issuing atomic writes, and we will have extent alignment and length as
> > > required.
> > 
> > I don't think that's a long time maintainable usage model.
> 
> For databases that are trying to use this to significantly improve
> their performance by eliminating double writes, the allocation and
> writes are being done by a single process.  So for *that* use case, it
> is quite maintainable.

That's not the freaking point.  We need to have proper kernel interfaces
that don't rely on intimate knowledge and control of details.  We need
to build proper genral purpose interfaces and not layer hacks on top of
hacks.