Re: [PATCH v2] usb: gadget: core: Check for unset descriptor

From: Alan Stern
Date: Thu Jul 25 2024 - 14:23:42 EST


On Thu, Jul 25, 2024 at 06:56:19AM +0200, Greg Kroah-Hartman wrote:
> On Wed, Jul 24, 2024 at 09:04:20PM -0400, crwulff@xxxxxxxxx wrote:
> > From: Chris Wulff <crwulff@xxxxxxxxx>
> >
> > Make sure the descriptor has been set before looking at maxpacket.
> > This fixes a null pointer panic in this case.
> >
> > This may happen if the gadget doesn't properly set up the endpoint
> > for the current speed, or the gadget descriptors are malformed and
> > the descriptor for the speed/endpoint are not found.
> >
> > No current gadget driver is known to have this problem, but this
> > may cause a hard-to-find bug during development of new gadgets.
> >
> > Fixes: 54f83b8c8ea9 ("USB: gadget: Reject endpoints with 0 maxpacket value")
> > Cc: stable@xxxxxxxxxxxxxxx
> > Signed-off-by: Chris Wulff <crwulff@xxxxxxxxx>
> > ---
> > v2: Added WARN_ONCE message & clarification on causes
> > v1: https://lore.kernel.org/all/20240721192048.3530097-2-crwulff@xxxxxxxxx/
> > ---
> > drivers/usb/gadget/udc/core.c | 10 ++++------
> > 1 file changed, 4 insertions(+), 6 deletions(-)
> >
> > diff --git a/drivers/usb/gadget/udc/core.c b/drivers/usb/gadget/udc/core.c
> > index 2dfae7a17b3f..81f9140f3681 100644
> > --- a/drivers/usb/gadget/udc/core.c
> > +++ b/drivers/usb/gadget/udc/core.c
> > @@ -118,12 +118,10 @@ int usb_ep_enable(struct usb_ep *ep)
> > goto out;
> >
> > /* UDC drivers can't handle endpoints with maxpacket size 0 */
> > - if (usb_endpoint_maxp(ep->desc) == 0) {
> > - /*
> > - * We should log an error message here, but we can't call
> > - * dev_err() because there's no way to find the gadget
> > - * given only ep.
> > - */
> > + if (!ep->desc || usb_endpoint_maxp(ep->desc) == 0) {
> > + WARN_ONCE(1, "%s: ep%d (%s) has %s\n", __func__, ep->address, ep->name,
> > + (!ep->desc) ? "NULL descriptor" : "maxpacket 0");
>
> So you just rebooted a machine that hit this, that's not good at all.
> Please log the error and recover, don't crash a system (remember,
> panic-on-warn is enabled in billions of Linux systems.)

That should not be a problem. This WARN_ONCE is expected never to be
triggered except by a buggy gadget driver. It's a debugging tool; the
developer will get an indication in the kernel log of where the problem
is instead of just a panic.

However, if you feel strongly about this, Chris probably won't mind
changing it to pr_err() plus dump_stack() instead of WARN_ONCE().

Alan Stern