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:   Mon, 9 Mar 2020 11:11:54 -0700
From:   Eric Biggers <ebiggers@kernel.org>
To:     Alexander Potapenko <glider@google.com>
Cc:     syzbot <syzbot+af962bf9e7e27bccd025@syzkaller.appspotmail.com>,
        len.brown@intel.com, LKML <linux-kernel@vger.kernel.org>,
        Linux PM <linux-pm@vger.kernel.org>,
        Pavel Machek <pavel@ucw.cz>,
        "Rafael J. Wysocki" <rjw@rjwysocki.net>,
        syzkaller-bugs <syzkaller-bugs@googlegroups.com>
Subject: Re: KMSAN: uninit-value in snapshot_compat_ioctl
Message-ID: <20200309181154.GB1073@sol.localdomain>
References: <000000000000938a57059f7cafe4@google.com>
 <20200307235437.GW15444@sol.localdomain>
 <20200308032434.GX15444@sol.localdomain>
 <CAG_fn=X8UkYx5=3ARUtW3+asc+3tEdeBg=1NKS9VzChSCp33Yg@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CAG_fn=X8UkYx5=3ARUtW3+asc+3tEdeBg=1NKS9VzChSCp33Yg@mail.gmail.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Mon, Mar 09, 2020 at 12:53:28PM +0100, 'Alexander Potapenko' via syzkaller-bugs wrote:
> > > Looks like a KMSAN false positive?  As far as I can tell, the memory is being
> > > initialized by put_user() called under set_fs(KERNEL_DS).
> 
> Why? put_user() doesn't write to kernel memory, instead it copies a
> value to the userspace.
> That's why KMSAN performs kmsan_check_memory() on it.
> It would actually be better if KMSAN printed an kernel-infoleak warning instead.

When under set_fs(KERNEL_DS), the userspace access functions like put_user() and
copy_to_user() can write to kernel memory.  It's discouraged and people have
been trying to get rid of uses of set_fs(), but a lot still remain, since
sometimes it's useful to allow code to operate on both user and kernel memory.
A common example is kernel_read().

> 
> > Although, it also looks like the problematic code can just be removed, since
> > always sizeof(compat_loff_t) == sizeof(loff_t).  I'll send a patch to do that...
> 
> Thanks!
> 

- Eric