Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Mon, 22 Feb 2021 16:59:37 +0300
From:   Dan Carpenter <dan.carpenter@oracle.com>
To:     karthek <mail@karthek.com>
Cc:     linux-kernel@vger.kernel.org, devel@driverdev.osuosl.org
Subject: Re: [PATCH] staging: wlan-ng/p80211 : check userspacebuf size for
 sanity
Message-ID: <20210222135937.GV2222@kadam>
References: <YDOnoLJzHYXMZBA/@karthik-strix-linux.karthek.com>
 <20210222132132.GU2222@kadam>
 <YDO0vtJyyGSSi44n@karthik-strix-linux.karthek.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <YDO0vtJyyGSSi44n@karthik-strix-linux.karthek.com>
User-Agent: Mutt/1.9.4 (2018-02-28)
Precedence: bulk

I have added the Driver Devel list to the CC list.  Adding linux-kernel
is sort of useless.  The correct people who are interested in this patch
are all on the Driver Devel list.

On Mon, Feb 22, 2021 at 07:12:22PM +0530, karthek wrote:
> On Mon, Feb 22, 2021 at 04:21:33PM +0300, Dan Carpenter wrote:
> > On Mon, Feb 22, 2021 at 06:16:24PM +0530, karthek wrote:
> > > currently p80211knetdev_do_ioctl() is testing user passed
> > > struct ifreq for sanity by checking for presence of a magic number,
> > > in addition to that also check size field, preventing buffer overflow
> > > before passing data to p80211req_dorequest() which casts it
> > > to *struct p80211msg
> > > 
> > > Signed-off-by: karthek <mail@karthek.com>
> > > ---
> > > is this correct?
> > > is it necessary to check for size in addition to magicnum?
> > > did i even understand the problem correctly?
> > > 
> > >  drivers/staging/wlan-ng/p80211netdev.c | 5 ++++-
> > >  1 file changed, 4 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/staging/wlan-ng/p80211netdev.c b/drivers/staging/wlan-ng/p80211netdev.c
> > > index 70570e8a5..c7b78d870 100644
> > > --- a/drivers/staging/wlan-ng/p80211netdev.c
> > > +++ b/drivers/staging/wlan-ng/p80211netdev.c
> > > @@ -568,7 +568,10 @@ static int p80211knetdev_do_ioctl(struct net_device *dev,
> > >  		result = -EINVAL;
> > >  		goto bail;
> > >  	}
> > > -
> > > +	if (req->len < sizeof(struct p80211msg)) {
> > > +		result = -EINVAL;
> > > +		goto bail;
> > > +	}
> > 
> > Please don't send private emails.  Always CC the list.
> sorry
> > 
> > That's only a partial solution.  You need to check in p80211req_handlemsg()
> > as well and probably other places.
> currently p80211req_handlemsg() is only referenced in p80211req_dorequest()
> can we check that there instead?

If I have to do all the work in finding the buffer overflows, then I
should write my own patch.  :/

Anyway the p80211knetdev_do_ioctl() function calls p80211req_dorequest()
which calls p80211req_handlemsg(wlandev, msg); and
wlandev->mlmerequest(wlandev, msg);.

We have already discussed the p80211req_handlemsg() function.  The
wlandev->mlmerequest() function is always just prism2sta_mlmerequest().
The prism2sta_mlmerequest() calls a bunch of functions and each of those
functions need to have a different limit check added to prevent memory
corruption.

Homework #1: Should we get rid of the wlandev->mlmerequest() pointer
and just call prism2sta_mlmerequest() directly?

Homework #2: Another solution is to just delete all these custom IOCTLs.
I don't know what they do so I don't know if they are necessary or not.

regards,
dan carpenter