[Linux-kernel-mentees] [PATCH v3] staging: comedi: check validity of wMaxPacketSize of usb endpoints found
gregkh at linuxfoundation.org
Sat Oct 10 09:35:19 UTC 2020
On Sat, Oct 10, 2020 at 01:59:32PM +0530, Anant Thazhemadam wrote:
> While finding usb endpoints in vmk80xx_find_usb_endpoints(), check if
> wMaxPacketSize = 0 for the endpoints found.
> Some devices have isochronous endpoints that have wMaxPacketSize = 0
> (as required by the USB-2 spec).
> However, since this doesn't apply here, wMaxPacketSize = 0 can be
> considered to be invalid.
> Reported-by: syzbot+009f546aa1370056b1c2 at syzkaller.appspotmail.com
> Tested-by: syzbot+009f546aa1370056b1c2 at syzkaller.appspotmail.com
> Signed-off-by: Anant Thazhemadam <anant.thazhemadam at gmail.com>
> Changes in v3:
> * Correctly list version information
> Changes in v2:
> * Fix coding style issue
> The error (as detected by syzbot) is generated in
> vmk80xx_write_packet() (which is called in vmk80xx_reset_device()) when
> it tries to assign devpriv->usb_tx_buf = cmd.
> This NULL pointer dereference issue arises because
> size = usb_endpoint_maxp(devpriv->ep_tx) = 0.
> This can be traced back to vmk80xx_find_usb_endpoints(), where the usb
> endpoints are found, and assigned accordingly.
> (For some more insight, in vmk80xx_find_usb_endpoints(),
> if one of intf->cur_altsetting->iface_desc->endpoints' desc value = 0,
> and consequently this endpoint is assigned to devpriv->ep_tx,
> this issue gets triggered.)
> Checking if the wMaxPacketSize of an endpoint is invalid and returning
> an error value accordingly, seems to fix the error.
> We could also alternatively perform this checking (if the size is 0 or not)
> in vmk80xx_reset_device() itself, but it only seemed like covering up the issue
> at that place, rather than fixing it, so I wasn't sure that was any better.
> However, if I'm not wrong, this might end up causing the probe to fail, and I'm
> not sure if that's the right thing to do in cases like this, and if it isn't I'd
> like some input on what exactly is the required course of action in cases like this.
> drivers/staging/comedi/drivers/vmk80xx.c | 3 +++
> 1 file changed, 3 insertions(+)
> diff --git a/drivers/staging/comedi/drivers/vmk80xx.c b/drivers/staging/comedi/drivers/vmk80xx.c
> index 65dc6c51037e..cb0a965d3c37 100644
> --- a/drivers/staging/comedi/drivers/vmk80xx.c
> +++ b/drivers/staging/comedi/drivers/vmk80xx.c
> @@ -667,6 +667,9 @@ static int vmk80xx_find_usb_endpoints(struct comedi_device *dev)
> if (!devpriv->ep_rx || !devpriv->ep_tx)
> return -ENODEV;
> + if (!usb_endpoint_maxp(devpriv->ep_rx) || !usb_endpoint_maxp(devpriv->ep_tx))
> + return -EINVAL;
> return 0;
Why not just rewrite vmk80xx_find_usb_endpoints() to use the
usb_find_common_endpoints() or other helper functions like
usb_find_bulk_in_endpoint() or others, so that this type of thing is
Ah, wait, no, the packet size is not checked there, sorry, maybe that
will not help out here. Is a bulk urb allowed to have a 0 size? If
not, maybe we should just forbid that in the core? Time to go read the
More information about the Linux-kernel-mentees