Re: [PATCH] KVM: x86/xen: Fix eventfd error handling in kvm_xen_eventfd_assign()

[Paolo Bonzini <pbonzini@redhat.com>]

On 10/28/22 11:26, Eiichi Tsukata wrote:
> Should not call eventfd_ctx_put() in case of error.
> 
> Fixes: 2fd6df2f2b47 ("KVM: x86/xen: intercept EVTCHNOP_send from guests")
> Reported-by: syzbot+6f0c896c5a9449a10ded@syzkaller.appspotmail.com
> Signed-off-by: Eiichi Tsukata <eiichi.tsukata@nutanix.com>
> ---
>   arch/x86/kvm/xen.c | 2 +-
>   1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
> index 93c628d3e3a9..a357994982c6 100644
> --- a/arch/x86/kvm/xen.c
> +++ b/arch/x86/kvm/xen.c
> @@ -1716,7 +1716,7 @@ static int kvm_xen_eventfd_assign(struct kvm *kvm,
>   	if (ret == -ENOSPC)
>   		ret = -EEXIST;
>   out:
> -	if (eventfd)
> +	if (eventfd && !IS_ERR(eventfd))
>   		eventfd_ctx_put(eventfd);
>   	kfree(evtchnfd);
>   	return ret;

Slightly more verbose, but cleaner:

diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 6714bbdbedf3..2dae413bd62a 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -1666,18 +1666,18 @@ static int kvm_xen_eventfd_assign(struct kvm *kvm,
  	case EVTCHNSTAT_ipi:
  		/* IPI  must map back to the same port# */
  		if (data->u.evtchn.deliver.port.port != data->u.evtchn.send_port)
-			goto out; /* -EINVAL */
+			goto out_noeventfd; /* -EINVAL */
  		break;
  
  	case EVTCHNSTAT_interdomain:
  		if (data->u.evtchn.deliver.port.port) {
  			if (data->u.evtchn.deliver.port.port >= max_evtchn_port(kvm))
-				goto out; /* -EINVAL */
+				goto out_noeventfd; /* -EINVAL */
  		} else {
  			eventfd = eventfd_ctx_fdget(data->u.evtchn.deliver.eventfd.fd);
  			if (IS_ERR(eventfd)) {
  				ret = PTR_ERR(eventfd);
-				goto out;
+				goto out_noeventfd;
  			}
  		}
  		break;
@@ -1717,6 +1717,7 @@ static int kvm_xen_eventfd_assign(struct kvm *kvm,
  out:
	if (eventfd)
  		eventfd_ctx_put(eventfd);
+out_noeventfd:
  	kfree(evtchnfd);
  	return ret;
  }

Only the last goto has to be changed in order to fix the bug, the
others are only needed to respect the LIFO order of the unwinding
labels.

Paolo