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:   Wed, 23 May 2018 15:57:08 +0200
From:   Guillaume Nault <g.nault@alphalink.fr>
To:     Eric Biggers <ebiggers3@gmail.com>
Cc:     linux-ppp@vger.kernel.org, Paul Mackerras <paulus@samba.org>,
        netdev@vger.kernel.org, linux-fsdevel@vger.kernel.org,
        linux-kernel@vger.kernel.org, syzkaller-bugs@googlegroups.com,
        Eric Biggers <ebiggers@google.com>
Subject: Re: [PATCH] ppp: remove the PPPIOCDETACH ioctl
Message-ID: <20180523135708.GB1569@alphalink.fr>
References: <20180523032958.GE658@sol.localdomain>
 <20180523035952.25768-1-ebiggers3@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20180523035952.25768-1-ebiggers3@gmail.com>
User-Agent: Mutt/1.9.5 (2018-04-13)
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Tue, May 22, 2018 at 08:59:52PM -0700, Eric Biggers wrote:
> From: Eric Biggers <ebiggers@google.com>
> 
> The PPPIOCDETACH ioctl effectively tries to "close" the given ppp file
> before f_count has reached 0, which is fundamentally a bad idea.  It
> does check 'f_count < 2', which excludes concurrent operations on the
> file since they would only be possible with a shared fd table, in which
> case each fdget() would take a file reference.  However, it fails to
> account for the fact that even with 'f_count == 1' the file can still be
> linked into epoll instances.  As reported by syzbot, this can trivially
> be used to cause a use-after-free.
> 
> Yet, the only known user of PPPIOCDETACH is pppd versions older than
> ppp-2.4.2, which was released almost 15 years ago (November 2003).
> Also, PPPIOCDETACH apparently stopped working reliably at around the
> same time, when the f_count check was added to the kernel, e.g. see
> https://lkml.org/lkml/2002/12/31/83.  Also, the current 'f_count < 2'
> check makes PPPIOCDETACH only work in single-threaded applications; it
> always fails if called from a multithreaded application.
> 
> All pppd versions released in the last 15 years just close() the file
> descriptor instead.
> 
> Therefore, instead of hacking around this bug by exporting epoll
> internals to modules, and probably missing other related bugs, just
> remove the PPPIOCDETACH ioctl and see if anyone actually notices.
> 
> Reported-by: syzbot+16363c99d4134717c05b@syzkaller.appspotmail.com
> Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
> Cc: stable@vger.kernel.org
> Signed-off-by: Eric Biggers <ebiggers@google.com>
> ---
>  Documentation/networking/ppp_generic.txt |  6 -----
>  drivers/net/ppp/ppp_generic.c            | 29 ------------------------
>  fs/compat_ioctl.c                        |  1 -
>  include/uapi/linux/ppp-ioctl.h           |  1 -
>  4 files changed, 37 deletions(-)
> 
> diff --git a/Documentation/networking/ppp_generic.txt b/Documentation/networking/ppp_generic.txt
> index 091d20273dcb..61daf4b39600 100644
> --- a/Documentation/networking/ppp_generic.txt
> +++ b/Documentation/networking/ppp_generic.txt
> @@ -300,12 +300,6 @@ unattached instance are:
>  The ioctl calls available on an instance of /dev/ppp attached to a
>  channel are:
>  
> -* PPPIOCDETACH detaches the instance from the channel.  This ioctl is
> -  deprecated since the same effect can be achieved by closing the
> -  instance.  In order to prevent possible races this ioctl will fail
> -  with an EINVAL error if more than one file descriptor refers to this
> -  instance (i.e. as a result of dup(), dup2() or fork()).
> -
>  * PPPIOCCONNECT connects this channel to a PPP interface.  The
>    argument should point to an int containing the interface unit
>    number.  It will return an EINVAL error if the channel is already
> diff --git a/drivers/net/ppp/ppp_generic.c b/drivers/net/ppp/ppp_generic.c
> index dc7c7ec43202..dce8812fe802 100644
> --- a/drivers/net/ppp/ppp_generic.c
> +++ b/drivers/net/ppp/ppp_generic.c
> @@ -603,35 +603,6 @@ static long ppp_ioctl(struct file *file, unsigned int cmd, unsigned long arg)
>  		goto out;
>  	}
>  
> -	if (cmd == PPPIOCDETACH) {
> -		/*
> -		 * We have to be careful here... if the file descriptor
> -		 * has been dup'd, we could have another process in the
> -		 * middle of a poll using the same file *, so we had
> -		 * better not free the interface data structures -
> -		 * instead we fail the ioctl.  Even in this case, we
> -		 * shut down the interface if we are the owner of it.
> -		 * Actually, we should get rid of PPPIOCDETACH, userland
> -		 * (i.e. pppd) could achieve the same effect by closing
> -		 * this fd and reopening /dev/ppp.
> -		 */
> -		err = -EINVAL;
> -		if (pf->kind == INTERFACE) {
> -			ppp = PF_TO_PPP(pf);
> -			rtnl_lock();
> -			if (file == ppp->owner)
> -				unregister_netdevice(ppp->dev);
> -			rtnl_unlock();
> -		}
> -		if (atomic_long_read(&file->f_count) < 2) {
> -			ppp_release(NULL, file);
> -			err = 0;
> -		} else
> -			pr_warn("PPPIOCDETACH file->f_count=%ld\n",
> -				atomic_long_read(&file->f_count));
> -		goto out;
> -	}
> -

I'd rather add
+	if (cmd == PPPIOCDETACH) {
+		err = -EINVAL;
+		goto out;
+	}

Making PPPIOCDETACH unknown to ppp_generic means that the ioctl would
be handled by the underlying channel when pf->kind == CHANNEL (see the
chan->ops->ioctl() call further down). That shouldn't be a problem per
se, but even though PPPIOCDETACH is unsupported, I feel that it should
remain a ppp_generic thing. I don't really want its value to be reused
for other purposes in the future or have different behaviour depending
on the underlying channel.

Also PPPIOCDETACH can already fail with -EINVAL. Therefore, if ever
there really were programs out there using this call, they'd already
have to handle this case. Unconditionally returning -EINVAL would
further minimise possibilities for breakage.