From: Claudio Imbrenda <imbrenda@xxxxxxxxxxxxx>
[ Upstream commit 292a7d6fca33df70ca4b8e9b0d0e74adf87582dc ]
On machines without the Destroy Secure Configuration Fast UVC, the
topmost level of page tables is set aside and freed asynchronously
as last step of the asynchronous teardown.
Each gmap has a host_to_guest radix tree mapping host (userspace)
addresses (with 1M granularity) to gmap segment table entries (pmds).
If a guest is smaller than 2GB, the topmost level of page tables is the
segment table (i.e. there are only 2 levels). Replacing it means that
the pointers in the host_to_guest mapping would become stale and cause
all kinds of nasty issues.
This patch fixes the issue by disallowing asynchronous teardown for
guests with only 2 levels of page tables. Userspace should (and already
does) try using the normal destroy if the asynchronous one fails.
Update s390_replace_asce so it refuses to replace segment type ASCEs.
This is still needed in case the normal destroy VM fails.
Fixes: fb491d5500a7 ("KVM: s390: pv: asynchronous destroy for reboot")
Reviewed-by: Marc Hartmayer <mhartmay@xxxxxxxxxxxxx>
Reviewed-by: Janosch Frank <frankja@xxxxxxxxxxxxx>
Signed-off-by: Claudio Imbrenda <imbrenda@xxxxxxxxxxxxx>
Message-Id: <20230421085036.52511-2-imbrenda@xxxxxxxxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
---
arch/s390/kvm/pv.c | 5 +++++
arch/s390/mm/gmap.c | 7 +++++++
2 files changed, 12 insertions(+)
diff --git a/arch/s390/kvm/pv.c b/arch/s390/kvm/pv.c
index e032ebbf51b97..3ce5f4351156a 100644
--- a/arch/s390/kvm/pv.c
+++ b/arch/s390/kvm/pv.c
@@ -314,6 +314,11 @@ int kvm_s390_pv_set_aside(struct kvm *kvm, u16 *rc, u16 *rrc)
*/
if (kvm->arch.pv.set_aside)
return -EINVAL;
+
+ /* Guest with segment type ASCE, refuse to destroy asynchronously */
+ if ((kvm->arch.gmap->asce & _ASCE_TYPE_MASK) == _ASCE_TYPE_SEGMENT)
+ return -EINVAL;
+
priv = kzalloc(sizeof(*priv), GFP_KERNEL);
if (!priv)
return -ENOMEM;
diff --git a/arch/s390/mm/gmap.c b/arch/s390/mm/gmap.c
index 74e1d873dce05..784fc6cbddb1a 100644
--- a/arch/s390/mm/gmap.c
+++ b/arch/s390/mm/gmap.c
@@ -2830,6 +2830,9 @@ EXPORT_SYMBOL_GPL(s390_unlist_old_asce);
* s390_replace_asce - Try to replace the current ASCE of a gmap with a copy
* @gmap: the gmap whose ASCE needs to be replaced
*
+ * If the ASCE is a SEGMENT type then this function will return -EINVAL,
+ * otherwise the pointers in the host_to_guest radix tree will keep pointing
+ * to the wrong pages, causing use-after-free and memory corruption.
* If the allocation of the new top level page table fails, the ASCE is not
* replaced.
* In any case, the old ASCE is always removed from the gmap CRST list.
@@ -2844,6 +2847,10 @@ int s390_replace_asce(struct gmap *gmap)
s390_unlist_old_asce(gmap);
+ /* Replacing segment type ASCEs would cause serious issues */
+ if ((gmap->asce & _ASCE_TYPE_MASK) == _ASCE_TYPE_SEGMENT)
+ return -EINVAL;
+
page = alloc_pages(GFP_KERNEL_ACCOUNT, CRST_ALLOC_ORDER);
if (!page)
return -ENOMEM;
--
2.39.2
