Re: [PATCH v2 6/7] s390: vfio-ap: handle bind and unbind of AP queue device

From: Tony Krowiak
Date: Mon May 06 2019 - 16:44:39 EST

On 5/6/19 6:55 AM, Pierre Morel wrote:
On 03/05/2019 23:14, Tony Krowiak wrote:
There is nothing preventing a root user from inadvertently unbinding an
AP queue device that is in use by a guest from the vfio_ap device driver
and binding it to a zcrypt driver. This can result in a queue being
accessible from both the host and a guest.

This patch introduces safeguards that prevent sharing of an AP queue
between the host when a queue device is unbound from the vfio_ap device
driver. In addition, this patch restores guest access to AP queue devices
bound to the vfio_ap driver if the queue's APQN is assigned to an mdev
device in use by a guest.

Signed-off-by: Tony Krowiak <akrowiak@xxxxxxxxxxxxx>
---
 drivers/s390/crypto/vfio_ap_drv.c | 12 +++-
 drivers/s390/crypto/vfio_ap_ops.c | 100 +++++++++++++++++++++++++++++++++-
 drivers/s390/crypto/vfio_ap_private.h | 2 +
 3 files changed, 111 insertions(+), 3 deletions(-)

diff --git a/drivers/s390/crypto/vfio_ap_drv.c b/drivers/s390/crypto/vfio_ap_drv.c
index e9824c35c34f..c215978daf39 100644
--- a/drivers/s390/crypto/vfio_ap_drv.c
+++ b/drivers/s390/crypto/vfio_ap_drv.c
@@ -42,12 +42,22 @@ MODULE_DEVICE_TABLE(vfio_ap, ap_queue_ids);
 static int vfio_ap_queue_dev_probe(struct ap_device *apdev)
 {
+ÂÂÂ struct ap_queue *queue = to_ap_queue(&apdev->device);
+
+ÂÂÂ mutex_lock(&matrix_dev->lock);
+ÂÂÂ vfio_ap_mdev_probe_queue(queue);
+ÂÂÂ mutex_unlock(&matrix_dev->lock);
+
ÂÂÂÂÂ return 0;
 }
 static void vfio_ap_queue_dev_remove(struct ap_device *apdev)
 {
-ÂÂÂ /* Nothing to do yet */
+ÂÂÂ struct ap_queue *queue = to_ap_queue(&apdev->device);
+
+ÂÂÂ mutex_lock(&matrix_dev->lock);
+ÂÂÂ vfio_ap_mdev_remove_queue(queue);
+ÂÂÂ mutex_unlock(&matrix_dev->lock);
 }
 static void vfio_ap_matrix_dev_release(struct device *dev)
diff --git a/drivers/s390/crypto/vfio_ap_ops.c b/drivers/s390/crypto/vfio_ap_ops.c
index ede45184eb67..40324951bd37 100644
--- a/drivers/s390/crypto/vfio_ap_ops.c
+++ b/drivers/s390/crypto/vfio_ap_ops.c
@@ -226,8 +226,6 @@ static struct device *vfio_ap_get_queue_dev(unsigned long apid,
ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ &apqn, match_apqn);
 }
-
-
 static int vfio_ap_mdev_validate_masks(unsigned long *apm, unsigned long *aqm)
 {
ÂÂÂÂÂ int ret;
@@ -259,6 +257,27 @@ static bool vfio_ap_queues_on_drv(unsigned long *apm, unsigned long *aqm)
ÂÂÂÂÂ return true;
 }
+static bool vfio_ap_card_on_drv(struct ap_queue *queue, unsigned long *aqm)
+{
+ÂÂÂ unsigned long apid, apqi;
+ÂÂÂ struct device *dev;
+
+ÂÂÂ apid = AP_QID_CARD(queue->qid);
+
+ÂÂÂ for_each_set_bit_inv(apqi, aqm, AP_DOMAINS) {
+ÂÂÂÂÂÂÂ if (queue->qid == AP_MKQID(apid, apqi))
+ÂÂÂÂÂÂÂÂÂÂÂ continue;
+
+ÂÂÂÂÂÂÂ dev = vfio_ap_get_queue_dev(apid, apqi);
+ÂÂÂÂÂÂÂ if (!dev)
+ÂÂÂÂÂÂÂÂÂÂÂ return false;
+
+ÂÂÂÂÂÂÂ put_device(dev);
+ÂÂÂ }
+
+ÂÂÂ return true;
+}
+
 /**
ÂÂ * assign_adapter_store
ÂÂ *
@@ -1017,3 +1036,80 @@ void vfio_ap_mdev_unregister(void)
 {
ÂÂÂÂÂ mdev_unregister_device(&matrix_dev->device);
 }
+
+static struct ap_matrix_mdev *vfio_ap_mdev_find_matrix_mdev(unsigned long apid,
+ÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂÂ unsigned long apqi)
+{
+ÂÂÂ struct ap_matrix_mdev *matrix_mdev;
+
+ÂÂÂ list_for_each_entry(matrix_mdev, &matrix_dev->mdev_list, node) {
+ÂÂÂÂÂÂÂ if (test_bit_inv(apid, matrix_mdev->matrix.apm) &&
+ÂÂÂÂÂÂÂÂÂÂÂ test_bit_inv(apqi, matrix_mdev->matrix.aqm))
+ÂÂÂÂÂÂÂÂÂÂÂ return matrix_mdev;
+ÂÂÂ }
+
+ÂÂÂ return NULL;
+}
+
+void vfio_ap_mdev_probe_queue(struct ap_queue *queue)
+{
+ÂÂÂ struct ap_matrix_mdev *matrix_mdev;
+ÂÂÂ unsigned long *shadow_apm, *shadow_aqm;
+ÂÂÂ unsigned long apid = AP_QID_CARD(queue->qid);
+ÂÂÂ unsigned long apqi = AP_QID_QUEUE(queue->qid);
+
+ÂÂÂ /*
+ÂÂÂÂ * Find the mdev device to which the APQN of the queue device being
+ÂÂÂÂ * probed is assigned
+ÂÂÂÂ */
+ÂÂÂ matrix_mdev = vfio_ap_mdev_find_matrix_mdev(apid, apqi);
+
+ÂÂÂ /* Check whether we found an mdev device and it is in use by a guest */
+ÂÂÂ if (matrix_mdev && matrix_mdev->kvm) {
+ÂÂÂÂÂÂÂ shadow_apm = matrix_mdev->shadow_crycb->apm;
+ÂÂÂÂÂÂÂ shadow_aqm = matrix_mdev->shadow_crycb->aqm;
+ÂÂÂÂÂÂÂ /*
+ÂÂÂÂÂÂÂÂ * If the guest already has access to the adapter card
+ÂÂÂÂÂÂÂÂ * referenced by APID or does not have access to the queues
+ÂÂÂÂÂÂÂÂ * referenced by APQI, there is nothing to do here.
+ÂÂÂÂÂÂÂÂ */
+ÂÂÂÂÂÂÂ if (test_bit_inv(apid, shadow_apm) ||
+ÂÂÂÂÂÂÂÂÂÂÂ !test_bit_inv(apqi, shadow_aqm))
+ÂÂÂÂÂÂÂÂÂÂÂ return;
+
+ÂÂÂÂÂÂÂ /*
+ÂÂÂÂÂÂÂÂ * If each APQN with the APID of the queue being probed and an
+ÂÂÂÂÂÂÂÂ * APQI in the shadow CRYCB references a queue device that is
+ÂÂÂÂÂÂÂÂ * bound to the vfio_ap driver, then plug the adapter into the
+ÂÂÂÂÂÂÂÂ * guest.
+ÂÂÂÂÂÂÂÂ */
+ÂÂÂÂÂÂÂ if (vfio_ap_card_on_drv(queue, shadow_aqm)) {
+ÂÂÂÂÂÂÂÂÂÂÂ set_bit_inv(apid, shadow_apm);
+ÂÂÂÂÂÂÂÂÂÂÂ vfio_ap_mdev_update_crycb(matrix_mdev);
+ÂÂÂÂÂÂÂ }
+ÂÂÂ }
+}
+
+void vfio_ap_mdev_remove_queue(struct ap_queue *queue)
+{
+ÂÂÂ struct ap_matrix_mdev *matrix_mdev;
+ÂÂÂ unsigned long apid = AP_QID_CARD(queue->qid);
+ÂÂÂ unsigned long apqi = AP_QID_QUEUE(queue->qid);
+
+ÂÂÂ matrix_mdev = vfio_ap_mdev_find_matrix_mdev(apid, apqi);
+
+ÂÂÂ /*
+ÂÂÂÂ * If the queue is assigned to the mdev device and the mdev device
+ÂÂÂÂ * is in use by a guest, unplug the adapter referred to by the APID
+ÂÂÂÂ * of the APQN of the queue being removed.
+ÂÂÂÂ */
+ÂÂÂ if (matrix_mdev && matrix_mdev->kvm) {
+ÂÂÂÂÂÂÂ if (!test_bit_inv(apid, matrix_mdev->shadow_crycb->apm))
+ÂÂÂÂÂÂÂÂÂÂÂ return;
+
+ÂÂÂÂÂÂÂ clear_bit_inv(apid, matrix_mdev->shadow_crycb->apm);
+ÂÂÂÂÂÂÂ vfio_ap_mdev_update_crycb(matrix_mdev);
+ÂÂÂ }
+
+ÂÂÂ vfio_ap_mdev_reset_queue(apid, apqi);
+}
diff --git a/drivers/s390/crypto/vfio_ap_private.h b/drivers/s390/crypto/vfio_ap_private.h
index e8457aa61976..6b1f7df5b979 100644
--- a/drivers/s390/crypto/vfio_ap_private.h
+++ b/drivers/s390/crypto/vfio_ap_private.h
@@ -87,5 +87,7 @@ struct ap_matrix_mdev {
 extern int vfio_ap_mdev_register(void);
 extern void vfio_ap_mdev_unregister(void);
+void vfio_ap_mdev_remove_queue(struct ap_queue *queue);
+void vfio_ap_mdev_probe_queue(struct ap_queue *queue);
 #endif /* _VFIO_AP_PRIVATE_H_ */



AFAIU the apmask/aqmask of the AP_BUS are replacing bind/unbind for the admin. Don't they?

Yes, these interfaces are used to bind/unbind.

Then why not suppress bind/unbind for ap_queues?

I did suppress them in a previous version, but I believe Harald
objected. I don't recall the reason. If any other maintainers
agree with this, I can reinstate that change. I personally would
prefer that. I think leaving the bind/unbind interfaces confuses
the issue.


Otherwise, it seems to me to handle correctly the disappearance of a card, which is the only thing that can happen from out of the firmware queue change requires configuration change and re-IPL.

You are correct.


Even still need testing, LGTM

I would welcome and appreciate additional testing, thanks in advance.