Re: [PATCH] net: xfrm: Fix potential oops in xfrm_user_rcv_msg and array out of bounds

[Steffen Klassert <steffen.klassert@secunet.com>]

On Tue, Mar 05, 2019 at 03:08:49PM +0800, Su Yanjun <suyj.fnst@cn.fujitsu.com> wrote:
> On 2019/3/5 14:49, Herbert Xu wrote:
> 
> > On Sun, Mar 03, 2019 at 10:47:39PM -0500, Su Yanjun wrote:
> > > When i review xfrm_user.c code, i found some potentical bug in it.
> > > 
> > > In xfrm_user_rcvmsg if type parameter from user space is set to
> > > XFRM_MSG_MAX or  XFRM_MSG_NEWSADINFO or XFRM_MSG_NEWSPDINFO. It will cause
> > > xfrm_user_rcv_msg  referring to null entry in xfrm_dispatch array.
> > > 
> > > Signed-off-by: Su Yanjun <suyj.fnst@cn.fujitsu.com>
> > > ---
> > >   net/xfrm/xfrm_user.c | 4 +++-
> > >   1 file changed, 3 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/net/xfrm/xfrm_user.c b/net/xfrm/xfrm_user.c
> > > index a131f9f..d832783 100644
> > > --- a/net/xfrm/xfrm_user.c
> > > +++ b/net/xfrm/xfrm_user.c
> > > @@ -2630,11 +2630,13 @@ static int xfrm_user_rcv_msg(struct sk_buff *skb, struct nlmsghdr *nlh,
> > >   		return -EOPNOTSUPP;
> > >   	type = nlh->nlmsg_type;
> > > -	if (type > XFRM_MSG_MAX)
> > > +	if (type >= XFRM_MSG_MAX)
> > >   		return -EINVAL;
> > Your patch is wrong.  Please check the definition of XFRM_MSG_MAX.
> 
> I see, thanks for your reply.
> 
>  	type -= XFRM_MSG_BASE;
>  	link = &xfrm_dispatch[type];
> +	if (!link)
> +		return -EOPNOTSUPP;
> 
> Here **link** may refer to null entry for special types such as
> XFRM_MSG_MAX or  XFRM_MSG_NEWSADINFO or XFRM_MSG_NEWSPDINFO
> Am i miss something?

'link' is always a valid pointer into that array.