Re: [Cocci] [PATCH] net: Zeroing the structure ethtool_wolinfo in ethtool_get_wol()

[Julia Lawall]

On Sat, 26 Oct 2019, Joe Perches wrote:

> On Sat, 2019-10-26 at 15:54 +0800, zhanglin wrote:
> > memset() the structure ethtool_wolinfo that has padded bytes
> > but the padded bytes have not been zeroed out.
> []
> > diff --git a/net/core/ethtool.c b/net/core/ethtool.c
> []
> > @@ -1471,11 +1471,13 @@ static int ethtool_reset(struct net_device *dev, char __user *useraddr)
> >
> >  static int ethtool_get_wol(struct net_device *dev, char __user *useraddr)
> >  {
> > -	struct ethtool_wolinfo wol = { .cmd = ETHTOOL_GWOL };
> > +	struct ethtool_wolinfo wol;
> >
> >  	if (!dev->ethtool_ops->get_wol)
> >  		return -EOPNOTSUPP;
> >
> > +	memset(&wol, 0, sizeof(struct ethtool_wolinfo));
> > +	wol.cmd = ETHTOOL_GWOL;
> >  	dev->ethtool_ops->get_wol(dev, &wol);
> >
> >  	if (copy_to_user(useraddr, &wol, sizeof(wol)))
>
> It seems likely there are more of these.
>
> Is there any way for coccinelle to find them?
>
> There are ~4000 structs in include/uapi and
> there are ~3000 uses of copy_to_user in the tree.
>
> $ git grep -P '\bstruct\s+\w+\s*{' include/uapi/ | cut -f2 -d" "|sort|uniq|wc -l
> 3785
> $ git grep -w copy_to_user|wc -l
> 2854
>
> A trivial grep and manual search using:
>
> $ git grep -B20 -w copy_to_user | grep -A20 -P '\bstruct\s+\w+\s*=\s*{'
>
> shows at least 1 (I didn't look very hard and stopped after finding 1):
>
>    include/uapi/linux/utsname.h:struct oldold_utsname {
>    include/uapi/linux/utsname.h-   char sysname[9];
>    include/uapi/linux/utsname.h-   char nodename[9];
>    include/uapi/linux/utsname.h-   char release[9];
>    include/uapi/linux/utsname.h-   char version[9];
>    include/uapi/linux/utsname.h-   char machine[9];
>    include/uapi/linux/utsname.h-};
>
> and
>
>    kernel/sys.c-	struct oldold_utsname tmp = {};
>    kernel/sys.c-
>    kernel/sys.c-	if (!name)
>    kernel/sys.c-		return -EFAULT;
>    kernel/sys.c-
>    kernel/sys.c-	down_read(&uts_sem);
>    kernel/sys.c-	memcpy(&tmp.sysname, &utsname()->sysname, __OLD_UTS_LEN);
>    kernel/sys.c-	memcpy(&tmp.nodename, &utsname()->nodename, __OLD_UTS_LEN);
>    kernel/sys.c-	memcpy(&tmp.release, &utsname()->release, __OLD_UTS_LEN);
>    kernel/sys.c-	memcpy(&tmp.version, &utsname()->version, __OLD_UTS_LEN);
>    kernel/sys.c-	memcpy(&tmp.machine, &utsname()->machine, __OLD_UTS_LEN);
>    kernel/sys.c-	up_read(&uts_sem);
>    kernel/sys.c:	if (copy_to_user(name, &tmp, sizeof(tmp)))
>
> where there is likely 3 bytes of padding after 45 bytes of data
> in the struct.

I looked into this at one point, but didn't get as far as generating
patches.  I think that the approach was roughly to collect the types of
the fields, and then generate code that would use BUILD_BUG_ON to complain
if the sum of the sizes was not the same as the size of the structure.
The problem was that I wasn't sure what was a real problem, nor what was
the best way to solve it.

julia