Received-SPF: pass (google.com: domain of linux-ext4+bounces-423-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Date: Wed, 13 Dec 2023 11:29:57 +0530
From: Ojaswin Mujoo <ojaswin@linux.ibm.com>
To: John Garry <john.g.garry@oracle.com>
Cc: linux-ext4@vger.kernel.org, "Theodore Ts'o" <tytso@mit.edu>,
        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: <ZXlIXWIqP9xipYzL@li-bb2b2a4c-3307-11b2-a85c-8fa5c3a69313.ibm.com>
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>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <c4cf3924-f67d-4f04-8460-054dbad70b93@oracle.com>

On Tue, Dec 12, 2023 at 07:46:51AM +0000, John Garry wrote:
> On 11/12/2023 10:54, Ojaswin Mujoo wrote:
> > > This seems a rather big drawback.
> > So if I'm not wrong, we force the extent alignment as well as the size
> > of the extent in xfs right.
> 
> For XFS in my v1 patchset, we only force alignment (but not size).
> 
> 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.
> 
> However - as we have seen with a trial user - it can create a problem if we
> don't do that and we write 4K and then overwrite with a 16K atomic write to
> a file, as 2x extents may be allocated for the complete 16K and it cannot be
> issued as a single BIO.

So currently, if we don't fallocate beforehand in xfs and the user
tries to do the 16k overwrite to an offset having a 4k extent, how are
we handling it?

Here ext4 will return an error indicating atomic write can't happen at
this particular offset. The way I see it is if the user passes atomic
flag to pwritev2 and we are unable to ensure atomicity for any reason we
return error, which seems like a fair approach for a generic interface. 

> 
> > 
> > We didn't want to overly restrict the users of atomic writes by
> > forcing
> > the extents to be of a certain alignment/size irrespective of the
> > size
> > of write. The design in this patchset provides this flexibility at
> > the
> > cost of some added precautions that the user should take (eg not
> > doing
> > an atomic write on a pre existing unaligned extent etc).
> 
> Doesn't bigalloc already give you what you require here?

Yes, but its an mkfs time feature and it also applies to each an every
file which might not be desirable for all use cases. 

Regards,
ojaswin