Re: [GIT PULL] USB fixes for 5.9-rc3

[Greg KH <gregkh@linuxfoundation.org>]

On Wed, Aug 26, 2020 at 06:53:02PM +0300, Andy Shevchenko wrote:
> On Wed, Aug 26, 2020 at 6:36 PM Greg KH <gregkh@linuxfoundation.org> wrote:
> > On Wed, Aug 26, 2020 at 05:02:38PM +0200, Marek Szyprowski wrote:
> > > On 26.08.2020 15:43, Greg KH wrote:
> 
> > > > Brooke Basile (2):
> > > >        USB: gadget: u_f: add overflow checks to VLA macros
> > >
> > > Sorry, but the above patch breaks USB Ethernet Gadget operation. It also
> > > didn't get the proper testing in linux-next (next-20200826 is the first
> > > one with this patch).
> > >
> > > This is how it explodes on Samsung Exynos (ARM 32bit) based board with
> > > g_ether module loaded:
> 
> > Odd, for a "normal" descriptor, the logic should have remained the same
> > as without this patch.
> 
> > Brooke, any ideas?
> 
> I have an idea.
> 
> Does below fix this?
> 
> 
> -- 
> With Best Regards,
> Andy Shevchenko

> diff --git a/drivers/usb/gadget/u_f.h b/drivers/usb/gadget/u_f.h
> index df4e1dcb357d..0b5c5f0dd073 100644
> --- a/drivers/usb/gadget/u_f.h
> +++ b/drivers/usb/gadget/u_f.h
> @@ -25,8 +25,8 @@
>  		size_t offset = 0;					       \
>  		if (groupname##__next != SIZE_MAX) {			       \
>  			size_t align_mask = __alignof__(type) - 1;	       \
> -			size_t offset = (groupname##__next + align_mask)       \
> -					 & ~align_mask;			       \
> +			offset = (groupname##__next + align_mask) &	       \
> +				  ~align_mask;				       \
>  			size_t size = array_size(n, sizeof(type));	       \
>  			if (check_add_overflow(offset, size,		       \
>  					       &groupname##__next)) {          \
> @@ -43,8 +43,8 @@
>  		size_t offset = 0;						\
>  		if (groupname##__next != SIZE_MAX) {				\
>  			size_t align_mask = __alignof__(type) - 1;		\
> -			size_t offset = (groupname##__next + align_mask)	\
> -					 & ~align_mask;				\
> +			offset = (groupname##__next + align_mask) &		\
> +				  ~align_mask;					\
>  			if (check_add_overflow(offset, groupname##_##name##__sz,\
>  							&groupname##__next)) {	\
>  				groupname##__next = SIZE_MAX;			\

Oh wow, yeah, that should fix the problem, but would be good to get
confirmation...

thanks,

greg k-h