[bug report] vxlan: keep flags and vni in network byte order

[bug report] vxlan: keep flags and vni in network byte order

[Dan Carpenter]

Hello Jiri Benc,

The patch 54bfd872bf16: "vxlan: keep flags and vni in network byte
order" from Feb 16, 2016, leads to the following static checker
warning:

	./include/net/vxlan.h:340 vxlan_vni()
	warn: potential shift truncation.  '0xffffff00 << 8'

include/net/vxlan.h
   335  static inline __be32 vxlan_vni(__be32 vni_field)
   336  {
   337  #if defined(__BIG_ENDIAN)
   338          return (__force __be32)((__force u32)vni_field >> 8);
   339  #else
   340          return (__force __be32)((__force u32)(vni_field & VXLAN_VNI_MASK) << 8);

Are you sure this is correct?  VXLAN_VNI_MASK already has an << 8 shift
so this feels like maybe it is a double shift.

   341  #endif
   342  }

regards,
dan carpenter

Re: [bug report] vxlan: keep flags and vni in network byte order

[Jiri Benc]

On Fri, 11 Nov 2016 15:49:41 +0300, Dan Carpenter wrote:
> Hello Jiri Benc,
> 
> The patch 54bfd872bf16: "vxlan: keep flags and vni in network byte
> order" from Feb 16, 2016, leads to the following static checker
> warning:
> 
> 	./include/net/vxlan.h:340 vxlan_vni()
> 	warn: potential shift truncation.  '0xffffff00 << 8'
> 
> include/net/vxlan.h
>    335  static inline __be32 vxlan_vni(__be32 vni_field)
>    336  {
>    337  #if defined(__BIG_ENDIAN)
>    338          return (__force __be32)((__force u32)vni_field >> 8);
>    339  #else
>    340          return (__force __be32)((__force u32)(vni_field & VXLAN_VNI_MASK) << 8);
> 
> Are you sure this is correct?  VXLAN_VNI_MASK already has an << 8 shift
> so this feels like maybe it is a double shift.

The static checker clearly can't understand endian conversion macros.
VXLAN_VNI_MASK is:

cpu_to_be32(VXLAN_VID_MASK << 8)

Thus on little endian, it becomes 0x00ffffff.

 Jiri

Re: [bug report] vxlan: keep flags and vni in network byte order

[Jiri Benc]

On Fri, 11 Nov 2016 14:05:09 +0100, Jiri Benc wrote:
> The static checker clearly can't understand endian conversion macros.
> VXLAN_VNI_MASK is:
> 
> cpu_to_be32(VXLAN_VID_MASK << 8)
> 
> Thus on little endian, it becomes 0x00ffffff.

Btw, until the unspecified static checker you used is not fixed to
resolve such macros correctly, it can't be really used to check the
kernel source, yet alone to generate bug reports. It will trip over
false positive in gazillion places.

 Jiri

Re: [bug report] vxlan: keep flags and vni in network byte order

[Dan Carpenter]

On Fri, Nov 11, 2016 at 02:12:03PM +0100, Jiri Benc wrote:
> On Fri, 11 Nov 2016 14:05:09 +0100, Jiri Benc wrote:
> > The static checker clearly can't understand endian conversion macros.
> > VXLAN_VNI_MASK is:
> > 
> > cpu_to_be32(VXLAN_VID_MASK << 8)
> > 
> > Thus on little endian, it becomes 0x00ffffff.

Ah right.

> 
> Btw, until the unspecified static checker you used is not fixed to
> resolve such macros correctly, it can't be really used to check the
> kernel source, yet alone to generate bug reports. It will trip over
> false positive in gazillion places.

Nope.  Just this one.

regards,
dan carpenter