Re: [PATCH v2 3/3] vfio: ap: AP Queue Interrupt Control VFIO ioctl calls

[Pierre Morel]

On 03/12/2018 11:04, Cornelia Huck wrote:
> On Thu, 22 Nov 2018 18:11:15 +0100
> Pierre Morel <pmorel@xxxxxxxxxxxxx> wrote:
>
> > This is the implementation of the VFIO ioctl calls to handle
> > the AQIC interception and use GISA to handle interrupts.
> >
> > Signed-off-by: Pierre Morel <pmorel@xxxxxxxxxxxxx>
> > ---
> >   drivers/s390/crypto/vfio_ap_ops.c | 110 +++++++++++++++++++++++++++++++++++++-
> >   1 file changed, 109 insertions(+), 1 deletion(-)
>
> > +static int vfio_ap_ioctl_setirq(struct mdev_device *mdev, unsigned long arg)
> > +{
> > +	struct ap_matrix_mdev *matrix_mdev = mdev_get_drvdata(mdev);
> > +	struct vfio_ap_aqic parm;
> > +	struct ap_qirq_ctrl aqic_gisa = {};
> > +	struct kvm *kvm = matrix_mdev->kvm;
> > +	struct kvm_s390_gisa *gisa = kvm->arch.gisa;
> > +	struct ap_queue_status ap_status;
> > +	unsigned long nib;
> > +
> > +	if (copy_from_user(&parm, (void __user *)arg, sizeof(parm)))
> > +		return -EFAULT;
> > +
> > +	if (parm.isc > MAX_ISC)
> > +		return -EINVAL;
> > +
> > +	if (kvm->vcpus[0]->arch.sie_block->gd & 0x01)
> > +		aqic_gisa.gf = 1;
> > +
> > +	nib = vfio_ap_get_nib(kvm, &parm);
> > +	if (!nib)
> > +		return -ENODEV;
> > +
> > +	aqic_gisa.gisc = parm.isc;
> > +	aqic_gisa.isc = kvm_s390_gisc_register(kvm, parm.isc);
>
> Oh, and as I just looked at the callers of these functions: You'll want
> to check the return code here, even though the function should not fail
> with the checking you did beforehand.

I should check.

> [I assume you'll have similar code even when using a different
> interface.]

Yes.


> > +	aqic_gisa.ir = 1;
> > +	aqic_gisa.gisa = gisa->next_alert >> 4;
> > +
> > +	ap_status = ap_aqic(parm.apqn, aqic_gisa, (void *)nib);
> > +	parm.status = *(uint32_t *)(&ap_status);
> > +
> > +	if (copy_to_user((void __user *)arg, &parm, sizeof(parm))) {
> > +		kvm_s390_gisc_unregister(kvm, parm.isc);
> > +		return -EFAULT;
> > +	}
> > +
> > +	if (ap_status.response_code) {
> > +		kvm_s390_gisc_unregister(kvm, parm.isc);
> > +		return -EIO;
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> > +static int vfio_ap_ioctl_clrirq(struct mdev_device *mdev, unsigned long arg)
> > +{
> > +	struct ap_matrix_mdev *matrix_mdev = mdev_get_drvdata(mdev);
> > +	struct vfio_ap_aqic parm;
> > +	struct ap_qirq_ctrl aqic_gisa = {};
> > +	struct ap_queue_status ap_status;
> > +	struct kvm *kvm = matrix_mdev->kvm;
> > +
> > +	if (copy_from_user(&parm, (void __user *)arg, sizeof(parm)))
> > +		return -EFAULT;
> > +
> > +	if (kvm->vcpus[0]->arch.sie_block->gd & 0x01)
> > +		aqic_gisa.gf = 1;
> > +	aqic_gisa.ir = 0;
> > +
> > +	ap_status = ap_aqic(parm.apqn, aqic_gisa, NULL);
> > +	parm.status = *(uint32_t *)(&ap_status);
> > +
> > +	kvm_s390_gisc_unregister(kvm, parm.isc);
>
> Here, you don't seem to verify the sanity of parm.isc beforehand...
> luckily the function can deal with that :)

You are right.
Anyway I will change this, because this relies on user's code which is 
not right.


> > +
> > +	if (copy_to_user((void __user *)arg, &parm, sizeof(parm)))
> > +		return -EFAULT;
> > +
> > +	return (ap_status.response_code) ? -EIO : 0;
> > +}
> > +


--
Pierre Morel
Linux/KVM/QEMU in BÃblingen - Germany