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:   Sun, 19 Jul 2020 11:23:38 +0200
From:   Greg KH <gregkh@linuxfoundation.org>
To:     Rustam Kovhaev <rkovhaev@gmail.com>
Cc:     devel@driverdev.osuosl.org,
        syzbot+c2a1fa67c02faa0de723@syzkaller.appspotmail.com,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] staging: wlan-ng: properly check endpoint types
Message-ID: <20200719092338.GC171181@kroah.com>
References: <20200718155836.86384-1-rkovhaev@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20200718155836.86384-1-rkovhaev@gmail.com>
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Sat, Jul 18, 2020 at 08:58:36AM -0700, Rustam Kovhaev wrote:
> As syzkaller detected, wlan-ng driver submits bulk urb without checking
> that the endpoint type is actually bulk, add usb_urb_ep_type_check()
> 
> Reported-and-tested-by: syzbot+c2a1fa67c02faa0de723@syzkaller.appspotmail.com
> Link: https://syzkaller.appspot.com/bug?extid=c2a1fa67c02faa0de723
> Signed-off-by: Rustam Kovhaev <rkovhaev@gmail.com>
> ---
>  drivers/staging/wlan-ng/hfa384x_usb.c | 14 ++++++++++++++
>  1 file changed, 14 insertions(+)
> 
> diff --git a/drivers/staging/wlan-ng/hfa384x_usb.c b/drivers/staging/wlan-ng/hfa384x_usb.c
> index fa1bf8b069fd..7cde60ea68a2 100644
> --- a/drivers/staging/wlan-ng/hfa384x_usb.c
> +++ b/drivers/staging/wlan-ng/hfa384x_usb.c
> @@ -339,6 +339,12 @@ static int submit_rx_urb(struct hfa384x *hw, gfp_t memflags)
>  
>  	hw->rx_urb_skb = skb;
>  
> +	result = usb_urb_ep_type_check(&hw->rx_urb);
> +	if (result) {
> +	       netdev_warn(hw->wlandev->netdev, "invalid rx endpoint");
> +	       goto cleanup;
> +	}

In looking at this again, can you just make these checks in the probe
function, and abort binding the driver to the device at that point in
time?  This feels really late in the init sequence.

The real problem here is in the hfa384x_create() function, where it
blindly takes the 1 and 2 endpoints and assumes that those are the
"correct type".  How about checking the types there, and if they are
incorrect, returning an error from that function and have the caller
return the error as well.

That should keep anything else in the driver from being initialized and
set up, and stop bad devices from being bound to the driver at a much
earlier point in time.

Note, just checking for the valid type/direction of those endpoints
should be sufficient.

thanks,

greg k-h