Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <9ef909de-1854-b4be-d272-2b4cda52329f@oppo.com>
 <20210922072326.3538-1-huangjianan@oppo.com> <e42a183f-274c-425f-2012-3ff0003e1fcb@139.com>
 <919e929d-6af7-b729-9fd2-954cd1e52999@oppo.com> <314324e7-02d7-dc43-b270-fb8117953549@139.com>
In-Reply-To: <314324e7-02d7-dc43-b270-fb8117953549@139.com>
From:   Miklos Szeredi <miklos@szeredi.hu>
Date:   Wed, 22 Sep 2021 16:00:47 +0200
Message-ID: <CAJfpegs_T5BQ+e79T=1fqTScjfaOyAftykmzK6=hdS=WhVvWsg@mail.gmail.com>
Subject: Re: [PATCH v3] ovl: fix null pointer when filesystemdoesn'tsupportdirect
 IO
To:     Chengguang Xu <cgxu519@139.com>
Cc:     Huang Jianan <huangjianan@oppo.com>,
        overlayfs <linux-unionfs@vger.kernel.org>,
        linux-erofs@lists.ozlabs.org, xiang@kernel.org, chao@kernel.org,
        guoweichao@oppo.com, yh@oppo.com, zhangshiming@oppo.com,
        guanyuwei@oppo.com, jnhuang95@gmail.com,
        linux-kernel@vger.kernel.org, linux-fsdevel@vger.kernel.org,
        Chengguang Xu <cgxu519@mykernel.net>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
Precedence: bulk

On Wed, 22 Sept 2021 at 15:21, Chengguang Xu <cgxu519@139.com> wrote:
>
> =E5=9C=A8 2021/9/22 16:24, Huang Jianan =E5=86=99=E9=81=93:
> >
> >
> > =E5=9C=A8 2021/9/22 16:06, Chengguang Xu =E5=86=99=E9=81=93:
> >> =E5=9C=A8 2021/9/22 15:23, Huang Jianan =E5=86=99=E9=81=93:
> >>> From: Huang Jianan <huangjianan@oppo.com>
> >>>
> >>> At present, overlayfs provides overlayfs inode to users. Overlayfs
> >>> inode provides ovl_aops with noop_direct_IO to avoid open failure
> >>> with O_DIRECT. But some compressed filesystems, such as erofs and
> >>> squashfs, don't support direct_IO.
> >>>
> >>> Users who use f_mapping->a_ops->direct_IO to check O_DIRECT support,
> >>> will read file through this way. This will cause overlayfs to access
> >>> a non-existent direct_IO function and cause panic due to null pointer=
:
> >>
> >> I just looked around the code more closely, in open_with_fake_path(),
> >>
> >> do_dentry_open() has already checked O_DIRECT open flag and
> >> a_ops->direct_IO of underlying real address_space.
> >>
> >> Am I missing something?
> >>
> >>
> >
> > It seems that loop_update_dio will set lo->use_dio after open file
> > without set O_DIRECT.
> > loop_update_dio will check f_mapping->a_ops->direct_IO but it deal
> > with ovl_aops with
> > noop _direct_IO.
> >
> > So I think we still need a new aops?
>
>
> It means we should only set ->direct_IO for overlayfs inodes whose
> underlying fs has DIRECT IO ability.

First let's fix the oops: ovl_read_iter()/ovl_write_iter() must check
real file's ->direct_IO if IOCB_DIRECT is set in iocb->ki_flags and
return -EINVAL if not.

To fix the loop -> overlay -> squashfs case your suggestion of having
separate aops depending on the real inode's ->direct_IO sounds good.

Thanks,
Miklos