Re: [PATCH] net: ipv4: ipmr: Prevent information leak in ipmr_sk_ioctl()

[Willem de Bruijn <willemdebruijn.kernel@gmail.com>]

Alper Ak wrote:
> struct sioc_vif_req has a padding hole after the vifi field due to
> alignment requirements. These padding bytes were uninitialized,
> potentially leaking kernel stack memory to userspace when the
> struct is copied via sock_ioctl_inout().
> 
> Reported by Smatch:
>     net/ipv4/ipmr.c:1575 ipmr_sk_ioctl() warn: check that 'buffer'
>     doesn't leak information (struct has a hole after 'vifi')
> 
> Fixes: e1d001fa5b47 ("net: ioctl: Use kernel memory on protocol ioctl callbacks")

The commit mentions other similar cases. If this is a concern for
sioc_vif_req, then it likely would alos be for sioc_mif_req6, which
similarly has a hole.

> Signed-off-by: Alper Ak <alperyasinak1@gmail.com>
> ---
>  net/ipv4/ipmr.c | 1 +
>  1 file changed, 1 insertion(+)
> 
> diff --git a/net/ipv4/ipmr.c b/net/ipv4/ipmr.c
> index ca9eaee4c2ef..18441fbe7ed7 100644
> --- a/net/ipv4/ipmr.c
> +++ b/net/ipv4/ipmr.c
> @@ -1571,6 +1571,7 @@ int ipmr_sk_ioctl(struct sock *sk, unsigned int cmd, void __user *arg)
>  	/* These userspace buffers will be consumed by ipmr_ioctl() */
>  	case SIOCGETVIFCNT: {
>  		struct sioc_vif_req buffer;
> +		memset(&buffer, 0, sizeof(buffer));
>  
>  		return sock_ioctl_inout(sk, cmd, arg, &buffer,
>  				      sizeof(buffer));

sock_ioctl_inout copies the whole struct from userspace, calls a
domain specific callback and then copies the whole struct back:

       if (copy_from_user(karg, arg, size))
               return -EFAULT;

       ret = READ_ONCE(sk->sk_prot)->ioctl(sk, cmd, karg);
       if (ret)
               return ret;

       if (copy_to_user(arg, karg, size))
               return -EFAULT;

As a result every byte of the memset will be overwritten with the
copy_from_user.


> -- 
> 2.43.0
>