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;
MIME-Version: 1.0
In-Reply-To: <20180409131021.5132-1-stefanha@redhat.com>
References: <20180409131021.5132-1-stefanha@redhat.com>
From:   Linus Torvalds <torvalds@linux-foundation.org>
Date:   Mon, 9 Apr 2018 09:52:13 -0700
Message-ID: <CA+55aFzJ_JPtU5SjAHHv--p3ZcAw4gVidxvrLLZF7w+kbHCNdg@mail.gmail.com>
Subject: Re: [PATCH] vhost: fix vhost_vq_access_ok() log check
To:     Stefan Hajnoczi <stefanha@redhat.com>
Cc:     virtualization <virtualization@lists.linux-foundation.org>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        Jason Wang <jasowang@redhat.com>,
        KVM list <kvm@vger.kernel.org>,
        "Michael S. Tsirkin" <mst@redhat.com>,
        syzkaller-bugs@googlegroups.com
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, Apr 9, 2018 at 6:10 AM, Stefan Hajnoczi <stefanha@redhat.com> wrote:
> @@ -1246,7 +1246,7 @@ int vhost_vq_access_ok(struct vhost_virtqueue *vq)
>  {
>         int ret = vq_log_access_ok(vq, vq->log_base);
>
> -       if (ret || vq->iotlb)
> +       if (!ret || vq->iotlb)
>                 return ret;

That logic is still very non-obvious.

This code already had one bug because of an odd illegible test
sequence. Let's not keep the crazy code.

Why not just do the *obvious* thing, and get rid of "ret" entirely,
and make the damn thing return a boolean, and then just write it all
as

    /* Caller should have vq mutex and device mutex */
    bool vhost_vq_access_ok(struct vhost_virtqueue *vq)
    {
            if (!vq_log_access_ok(vq, vq->log_base))
                    return false;

            if (vq->iotlb || vq_access_ok(vq, vq->num, vq->desc,
vq->avail, vq->used);
    }

which makes the logic obvious: if vq_log_access_ok() fails, then then
vhost_vq_access_ok() fails unconditionally.

Otherwise, we need to have an iotlb, or a successful vq_access_ok() check.

Doesn't that all make more sense, and avoid the insane "ret" value use
that is really quite subtle?

                    Linus