linux-kernel - [PATCH v1 1/1] s390/vfio-ap: remove gmap_convert_to_secure from vfio_ap

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite for Android: free password hash cracker in your pocket

[<prev] [next>] [thread-next>] [day] [month] [year] [list]

Message-ID: <20241115135611.87836-1-imbrenda@linux.ibm.com>
Date: Fri, 15 Nov 2024 14:56:11 +0100
From: Claudio Imbrenda <imbrenda@...ux.ibm.com>
To: kvm@...r.kernel.org
Cc: linux-kernel@...r.kernel.org, linux-s390@...r.kernel.org,
        akrowiak@...ux.ibm.com, pasic@...ux.ibm.com, jjherne@...ux.ibm.com,
        freude@...ux.ibm.com, hca@...ux.ibm.com, gor@...ux.ibm.com,
        agordeev@...ux.ibm.com, borntraeger@...ibm.com, svens@...ux.ibm.com,
        frankja@...ux.ibm.com, nrb@...ux.ibm.com, nsg@...ux.ibm.com,
        seiden@...ux.ibm.com
Subject: [PATCH v1 1/1] s390/vfio-ap: remove gmap_convert_to_secure from vfio_ap_ops

If the page has been exported, do not re-import it. Imports should
only be triggered by the guest. The guest will import the page
automatically when it will need it again, there is no advantage in
importing it manually.

Moreover, vfio_pin_pages() will take an extra reference on the page and
thus will cause the import to always fail. The extra reference would be
dropped only after pointlessly trying to import the page.

Fixes: f88fb1335733 ("s390/vfio-ap: make sure nib is shared")
Signed-off-by: Claudio Imbrenda <imbrenda@...ux.ibm.com>
---
 drivers/s390/crypto/vfio_ap_ops.c | 32 +++++++++++++++++++++----------
 1 file changed, 22 insertions(+), 10 deletions(-)

diff --git a/drivers/s390/crypto/vfio_ap_ops.c b/drivers/s390/crypto/vfio_ap_ops.c
index 9f76f2d7b66e..ace1b332c930 100644
--- a/drivers/s390/crypto/vfio_ap_ops.c
+++ b/drivers/s390/crypto/vfio_ap_ops.c
@@ -360,10 +360,26 @@ static int vfio_ap_validate_nib(struct kvm_vcpu *vcpu, dma_addr_t *nib)
 	return 0;
 }
 
-static int ensure_nib_shared(unsigned long addr, struct gmap *gmap)
+/**
+ * ensure_nib_shared() - Ensure the address of the NIB is secure and shared
+ * @addr: the physical (absolute) address of the NIB
+ *
+ * This function checks whether the NIB page, which has been pinned with
+ * vfio_pin_pages(), is a shared page belonging to a secure guest.
+ *
+ * It will call uv_pin_shared() on it; if the page was already pinned shared
+ * (i.e. if the NIB belongs to a secure guest and is shared), then 0
+ * (success) is returned. If the NIB was not shared, vfio_pin_pages() had
+ * exported it and now it does not belong to the secure guest anymore. In
+ * that case, an error is returned.
+ *
+ * Context: the NIB (at physical address @addr) has to be pinned with
+ *          vfio_pin_pages() before calling this function.
+ *
+ * Return: 0 in case of success, otherwise an error < 0.
+ */
+static int ensure_nib_shared(unsigned long addr)
 {
-	int ret;
-
 	/*
 	 * The nib has to be located in shared storage since guest and
 	 * host access it. vfio_pin_pages() will do a pin shared and
@@ -374,12 +390,7 @@ static int ensure_nib_shared(unsigned long addr, struct gmap *gmap)
 	 *
 	 * If the page is already pinned shared the UV will return a success.
 	 */
-	ret = uv_pin_shared(addr);
-	if (ret) {
-		/* vfio_pin_pages() likely exported the page so let's re-import */
-		gmap_convert_to_secure(gmap, addr);
-	}
-	return ret;
+	return uv_pin_shared(addr);
 }
 
 /**
@@ -425,6 +436,7 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
 		return status;
 	}
 
+	/* The pin will probably be successful even if the NIB was not shared */
 	ret = vfio_pin_pages(&q->matrix_mdev->vdev, nib, 1,
 			     IOMMU_READ | IOMMU_WRITE, &h_page);
 	switch (ret) {
@@ -447,7 +459,7 @@ static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
 
 	/* NIB in non-shared storage is a rc 6 for PV guests */
 	if (kvm_s390_pv_cpu_is_protected(vcpu) &&
-	    ensure_nib_shared(h_nib & PAGE_MASK, kvm->arch.gmap)) {
+	    ensure_nib_shared(h_nib & PAGE_MASK)) {
 		vfio_unpin_pages(&q->matrix_mdev->vdev, nib, 1);
 		status.response_code = AP_RESPONSE_INVALID_ADDRESS;
 		return status;
-- 
2.47.0