Re: [PATCH] wifi: ath6kl: Check that the read operation returns a data length of 0

[Greg KH <gregkh@linuxfoundation.org>]

On Sun, Aug 25, 2024 at 04:14:17PM +0800, Edward Adam Davis wrote:
> On Sun, 25 Aug 2024 09:25:37 +0200, Greg KH wrote:
> > > If the data length returned by the device is 0, the read operation
> > > should be considered a failure.
> > >
> > > Reported-and-tested-by: syzbot+92c6dd14aaa230be6855@syzkaller.appspotmail.com
> > > Signed-off-by: Edward Adam Davis <eadavis@qq.com>
> > > ---
> > >  drivers/net/wireless/ath/ath6kl/usb.c | 3 +++
> > >  1 file changed, 3 insertions(+)
> > >
> > > diff --git a/drivers/net/wireless/ath/ath6kl/usb.c b/drivers/net/wireless/ath/ath6kl/usb.c
> > > index 5220809841a6..2a89bab81b24 100644
> > > --- a/drivers/net/wireless/ath/ath6kl/usb.c
> > > +++ b/drivers/net/wireless/ath/ath6kl/usb.c
> > > @@ -1034,6 +1034,9 @@ static int ath6kl_usb_bmi_read(struct ath6kl *ar, u8 *buf, u32 len)
> > >  		ath6kl_err("Unable to read the bmi data from the device: %d\n",
> > >  			   ret);
> > >  		return ret;
> > > +	} else {
> > > +		ath6kl_err("Actual read the bmi data length is 0 from the device\n");
> > > +		return -EIO;
> > 
> > Close, but not quite there.  ath6kl_usb_submit_ctrl_in() needs to verify
> > that the actual amount of data was read that was asked for.  If a short
> > read happens (or a long one), then an error needs to propagate out, not
> > just 0.  See the "note:" line in that function for what needs to be
> > properly checked.
> > 
> > hope this helps,
> Thanks for your analysis.
> I have carefully read your analysis and I am not sure if the following
> understanding is appropriate:
> diff --git a/drivers/net/wireless/ath/ath6kl/usb.c b/drivers/net/wireless/ath/ath6kl/usb.c
> index 2a89bab81b24..35884316a8c8 100644
> --- a/drivers/net/wireless/ath/ath6kl/usb.c
> +++ b/drivers/net/wireless/ath/ath6kl/usb.c
> @@ -932,6 +932,15 @@ static int ath6kl_usb_submit_ctrl_in(struct ath6kl_usb *ar_usb,
> 
>         kfree(buf);

First off, this should be using usb_control_msg_send() instead of having
to roll their own buffer handling, right?

> +       /* There are two types of read failure situations that need to be captured:
> +        * 1. short read: ret < size && ret >= 0
> +        * 2. long read: ret > size
> +        * */
> +       if (req == ATH6KL_USB_CONTROL_REQ_RECV_BMI_RESP && ret != size) {
> +               ath6kl_warn("Actual read the data length is: %d, but input size is %d\n", ret, size);
> +               return -EIO;
> +       }

If you switch to usb_control_msg_send() this logic gets a lot simpler.
Perhaps do that instead?

If not, then you need to check for "short writes" or zero writes, see
the documentation for usb_control_msg() for what it returns.  Your
comment is not correct here, there are 3 different return "states" that
you need to handle.

And why are you caring about what the req type is?

thanks,

greg k-h