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;
Date:   Thu, 21 Nov 2019 08:48:42 -0600
From:   Pete Zaitcev <zaitcev@redhat.com>
To:     Alan Stern <stern@rowland.harvard.edu>
Cc:     syzbot <syzbot+56f9673bb4cdcbeb0e92@syzkaller.appspotmail.com>,
        <arnd@arndb.de>, <gregkh@linuxfoundation.org>,
        <jrdr.linux@gmail.com>, <keescook@chromium.org>,
        <kstewart@linuxfoundation.org>,
        Kernel development list <linux-kernel@vger.kernel.org>,
        USB list <linux-usb@vger.kernel.org>,
        <syzkaller-bugs@googlegroups.com>, <tglx@linutronix.de>,
        <viro@zeniv.linux.org.uk>, zaitcev@redhat.com
Subject: Re: possible deadlock in mon_bin_vma_fault
Message-ID: <20191121084842.095edf87@suzdal.zaitcev.lan>
In-Reply-To: <Pine.LNX.4.44L0.1911201343580.1498-100000@iolanthe.rowland.org>
References: <20191120111235.7d306f23@suzdal.zaitcev.lan>
        <Pine.LNX.4.44L0.1911201343580.1498-100000@iolanthe.rowland.org>
Organization: Red Hat, Inc.
MIME-Version: 1.0
Content-Type: text/plain; charset=WINDOWS-1252
Content-Transfer-Encoding: quoted-printable
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, 20 Nov 2019 13:47:00 -0500 (EST)
Alan Stern <stern@rowland.harvard.edu> wrote:

> > +               if (rp->mmap_active)
> > +                       return -EBUSY;

> Like that, yes, but the test has to be made while fetch_lock is held. =20

Certainly, thanks. I was rushing just to add a postscriptum.

> Incidentally, the comment for fetch_lock says that it protects b_read=20
> and b_out, but mon_bin_vma_fault doesn't use either of those fields.

I probably should change that comment to "protect the integrity of the
circular buffer, such as b_out".

Anyway... If you are looking at it too, what do you think about not using
any locks in mon_bin_vma_fault() at all? Isn't it valid? I think I tried
to be "safe", but it only uses things that are constants unless we're
opening and closing; a process cannot make page faults unless it has
some thing mapped; and that is only possible if device is open and stays
open. Can you find a hole in this reasoning?

-- Pete