Re: [PATCH 3/3] userfaultfd: apply unprivileged_userfaultfd check

[Peter Xu <peterx@redhat.com>]

On Mon, Mar 11, 2019 at 05:37:01PM +0800, Peter Xu wrote:
> Apply the unprivileged_userfaultfd check when doing userfaultfd
> syscall.  We didn't check it in other paths of userfaultfd (e.g., the
> ioctl() path) because we don't want to drag down the fast path of
> userfaultfd, as suggested by Andrea.
> 
> Suggested-by: Andrea Arcangeli <aarcange@redhat.com>
> Suggested-by: Mike Rapoport <rppt@linux.vnet.ibm.com>
> Signed-off-by: Peter Xu <peterx@redhat.com>
> ---
>  fs/userfaultfd.c | 25 +++++++++++++++++++++++++
>  1 file changed, 25 insertions(+)
> 
> diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
> index c2188464555a..effdcfc88629 100644
> --- a/fs/userfaultfd.c
> +++ b/fs/userfaultfd.c
> @@ -951,6 +951,28 @@ void userfaultfd_unmap_complete(struct mm_struct *mm, struct list_head *uf)
>  	}
>  }
>  
> +/* Whether current process allows to use userfaultfd syscalls */
> +static bool userfaultfd_allowed(void)
> +{
> +	bool allowed = false;
> +
> +	switch (unprivileged_userfaultfd) {
> +	case UFFD_UNPRIV_ENABLED:
> +		allowed = true;
> +		break;
> +	case UFFD_UNPRIV_KVM:
> +		allowed = !!test_bit(MMF_USERFAULTFD_ALLOW,
> +				     &current->mm->flags);
> +		/* Fall through */

Sorry I should squash this in otherwise compilation of !CONFIG_KVM
will break:

diff --git a/fs/userfaultfd.c b/fs/userfaultfd.c
index effdcfc88629..1b3fa5935643 100644
--- a/fs/userfaultfd.c
+++ b/fs/userfaultfd.c
@@ -960,10 +960,12 @@ static bool userfaultfd_allowed(void)
        case UFFD_UNPRIV_ENABLED:
                allowed = true;
                break;
+#if IS_ENABLED(CONFIG_KVM)
        case UFFD_UNPRIV_KVM:
                allowed = !!test_bit(MMF_USERFAULTFD_ALLOW,
                                     &current->mm->flags);
                /* Fall through */
+#endif
        case UFFD_UNPRIV_DISABLED:
                allowed = allowed || ns_capable(current_user_ns(),
                                                CAP_SYS_PTRACE);

Will wait for more comments before I repost.  Sorry for the noise.

Regards,

-- 
Peter Xu