Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Content-Type: text/plain; charset=us-ascii
Content-Transfer-Encoding: 7bit
From:   Qian Cai <cai@lca.pw>
Mime-Version: 1.0 (1.0)
Subject: Re: [PATCH] fs: fix a data race in i_size_write/i_size_read
Date:   Thu, 20 Feb 2020 23:19:39 -0500
Message-Id: <7351A815-AA56-45FC-B15F-38F81F140253@lca.pw>
References: <CANpmjNM=+y-OwKjtsjsEkwPjpHXpt7ywaE48JyiND6dKt=Vf1Q@mail.gmail.com>
Cc:     Al Viro <viro@zeniv.linux.org.uk>,
        Christoph Hellwig <hch@infradead.org>,
        "Darrick J. Wong" <darrick.wong@oracle.com>,
        linux-xfs@vger.kernel.org,
        linux-fsdevel <linux-fsdevel@vger.kernel.org>,
        LKML <linux-kernel@vger.kernel.org>
In-Reply-To: <CANpmjNM=+y-OwKjtsjsEkwPjpHXpt7ywaE48JyiND6dKt=Vf1Q@mail.gmail.com>
To:     Marco Elver <elver@google.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk


> On Feb 19, 2020, at 4:21 AM, Marco Elver <elver@google.com> wrote:
> 
> Let's assume the WRITE_ONCE can be dropped.
> 
> The load is a different story. While load tearing may not be an issue,
> it's more likely that other optimizations can break the code. For
> example load fusing can break code that expects repeated loads in a
> loop. E.g. I found these uses of i_size_read in loops:
> 
> git grep -E '(for|while) \(.*i_size_read'
> fs/ocfs2/dir.c: while (ctx->pos < i_size_read(inode)) {
> fs/ocfs2/dir.c:                 for (i = 0; i < i_size_read(inode) &&
> i < offset; ) {
> fs/ocfs2/dir.c: while (ctx->pos < i_size_read(inode)) {
> fs/ocfs2/dir.c:         while (ctx->pos < i_size_read(inode)
> fs/squashfs/dir.c:      while (length < i_size_read(inode)) {
> fs/squashfs/namei.c:    while (length < i_size_read(dir)) {
> 
> Can i_size writes happen concurrently, and if so will these break if
> the compiler decides to just do i_size_read's load once, and keep the
> result in a register?

Al, is it more acceptable to add READ_ONCE() only then?