Skip to contentSkip to conversation

s390/kvm: Reject memory region operations for ucontrol VMs

PATCH In review Christoph Schlameuss 4 participants 6 messages Jun 24, 2024 – Jun 27, 2024

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 s390/kvm: Reject memory region operations for ucontrol VMs

This change rejects the KVM_SET_USER_MEMORY_REGION and KVM_SET_USER_MEMORY_REGION2 ioctls when called on a ucontrol VM. This is neccessary since ucontrol VMs have kvm->arch.gmap set to 0 and would thus result in a null pointer dereference further in. Memory management needs to be performed in userspace and using the ioctls KVM_S390_UCAS_MAP and KVM_S390_UCAS_UNMAP.

Also improve s390 specific documentation for KVM_SET_USER_MEMORY_REGION and KVM_SET_USER_MEMORY_REGION2.

---
 Documentation/virt/kvm/api.rst | 12 ++++++++++++
 arch/s390/kvm/kvm-s390.c       |  3 +++
 2 files changed, 15 insertions(+)
Documentation/virt/kvm/api.rst +12 -0
index a71d91978d9e..eec8df1dde06 100644
@@ -1403,6 +1403,12 @@ Instead, an abort (data abort if the cause of the page-table update
 was a load or a store, instruction abort if it was an instruction
 fetch) is injected in the guest.
 
+S390:
+^^^^^
+
+Returns -EINVAL if the VM has the KVM_VM_S390_UCONTROL flag set.
+Returns -EINVAL if called on a protected VM.
+
 4.36 KVM_SET_TSS_ADDR
 ---------------------
 
@@ -6273,6 +6279,12 @@ state.  At VM creation time, all memory is shared, i.e. the PRIVATE attribute
 is '0' for all gfns.  Userspace can control whether memory is shared/private by
 toggling KVM_MEMORY_ATTRIBUTE_PRIVATE via KVM_SET_MEMORY_ATTRIBUTES as needed.
 
+S390:
+^^^^^
+
+Returns -EINVAL if the VM has the KVM_VM_S390_UCONTROL flag set.
+Returns -EINVAL if called on a protected VM.
+
 4.141 KVM_SET_MEMORY_ATTRIBUTES
 -------------------------------
 
arch/s390/kvm/kvm-s390.c +3 -0
index 82e9631cd9ef..854d0d1410be 100644
@@ -5748,6 +5748,9 @@ int kvm_arch_prepare_memory_region(struct kvm *kvm,
 {
 	gpa_t size;
 
+	if (kvm_is_ucontrol(kvm))
+		return -EINVAL;
+
 	/* When we are protected, we should not change the memory slots */
 	if (kvm_s390_pv_get_handle(kvm))
 		return -EINVAL;

base-commit: f2661062f16b2de5d7b6a5c42a9a5c96326b8454

Signature · 1 line
2.45.2
maintainer Reviewed-by on v1 1/1
#1 lore raw
Quoted 11 lines · On Mon, 24 Jun 2024 11:59:02 +0200 Christoph Schlameuss…

This change rejects the KVM_SET_USER_MEMORY_REGION and KVM_SET_USER_MEMORY_REGION2 ioctls when called on a ucontrol VM. This is neccessary since ucontrol VMs have kvm->arch.gmap set to 0 and would thus result in a null pointer dereference further in. Memory management needs to be performed in userspace and using the ioctls KVM_S390_UCAS_MAP and KVM_S390_UCAS_UNMAP.

Also improve s390 specific documentation for KVM_SET_USER_MEMORY_REGION and KVM_SET_USER_MEMORY_REGION2.

Quoted 51 lines
---
 Documentation/virt/kvm/api.rst | 12 ++++++++++++
 arch/s390/kvm/kvm-s390.c       |  3 +++
 2 files changed, 15 insertions(+)
Documentation/virt/kvm/api.rst +12 -0
index a71d91978d9e..eec8df1dde06 100644
@@ -1403,6 +1403,12 @@ Instead, an abort (data abort if the cause of the page-table update
 was a load or a store, instruction abort if it was an instruction
 fetch) is injected in the guest.
 
+S390:
+^^^^^
+
+Returns -EINVAL if the VM has the KVM_VM_S390_UCONTROL flag set.
+Returns -EINVAL if called on a protected VM.
+
 4.36 KVM_SET_TSS_ADDR
 ---------------------
 
@@ -6273,6 +6279,12 @@ state.  At VM creation time, all memory is shared, i.e. the PRIVATE attribute
 is '0' for all gfns.  Userspace can control whether memory is shared/private by
 toggling KVM_MEMORY_ATTRIBUTE_PRIVATE via KVM_SET_MEMORY_ATTRIBUTES as needed.
 
+S390:
+^^^^^
+
+Returns -EINVAL if the VM has the KVM_VM_S390_UCONTROL flag set.
+Returns -EINVAL if called on a protected VM.
+
 4.141 KVM_SET_MEMORY_ATTRIBUTES
 -------------------------------
 
arch/s390/kvm/kvm-s390.c +3 -0
index 82e9631cd9ef..854d0d1410be 100644
@@ -5748,6 +5748,9 @@ int kvm_arch_prepare_memory_region(struct kvm *kvm,
 {
 	gpa_t size;
 
+	if (kvm_is_ucontrol(kvm))
+		return -EINVAL;
+
 	/* When we are protected, we should not change the memory slots */
 	if (kvm_s390_pv_get_handle(kvm))
 		return -EINVAL;

base-commit: f2661062f16b2de5d7b6a5c42a9a5c96326b8454

maintainer
#2 lore raw
In reply to Christoph Schlameuss
Quoted 10 lines · Christoph Schlameuss

This change rejects the KVM_SET_USER_MEMORY_REGION and KVM_SET_USER_MEMORY_REGION2 ioctls when called on a ucontrol VM. This is neccessary since ucontrol VMs have kvm->arch.gmap set to 0 and would thus result in a null pointer dereference further in. Memory management needs to be performed in userspace and using the ioctls KVM_S390_UCAS_MAP and KVM_S390_UCAS_UNMAP.

Also improve s390 specific documentation for KVM_SET_USER_MEMORY_REGION and KVM_SET_USER_MEMORY_REGION2.

I'll add this tag when picking since it's a fix:

Fixes 27e0393f15fc ("KVM: s390: ucontrol: per vcpu address spaces")

It's a bit hard to track down the commit that should have contained your code to begin with and I think this is the closest we'll get.

maintainer
#3 lore raw
In reply to Christoph Schlameuss
Quoted 10 lines · Christoph Schlameuss

This change rejects the KVM_SET_USER_MEMORY_REGION and KVM_SET_USER_MEMORY_REGION2 ioctls when called on a ucontrol VM. This is neccessary since ucontrol VMs have kvm->arch.gmap set to 0 and would thus result in a null pointer dereference further in. Memory management needs to be performed in userspace and using the ioctls KVM_S390_UCAS_MAP and KVM_S390_UCAS_UNMAP.

Also improve s390 specific documentation for KVM_SET_USER_MEMORY_REGION and KVM_SET_USER_MEMORY_REGION2.

Would be nice to have a selftest for ucontrol VMs, too... just saying :)

Paolo

maintainer
#4 lore raw
Quoted 10 lines · Paolo Bonzini
Quoted 8 lines · Christoph Schlameuss

This change rejects the KVM_SET_USER_MEMORY_REGION and KVM_SET_USER_MEMORY_REGION2 ioctls when called on a ucontrol VM. This is neccessary since ucontrol VMs have kvm->arch.gmap set to 0 and would thus result in a null pointer dereference further in. Memory management needs to be performed in userspace and using the ioctls KVM_S390_UCAS_MAP and KVM_S390_UCAS_UNMAP.

Quoted 7 lines
Quoted 2 lines

Also improve s390 specific documentation for KVM_SET_USER_MEMORY_REGION and KVM_SET_USER_MEMORY_REGION2.

Would be nice to have a selftest for ucontrol VMs, too... just saying :)

Paolo

Already in the works, he just hasn't posted it yet :) We did do a couple rounds of internal feedback on the tests first.

#5 lore raw
Quoted 14 lines · On Thu, 27 Jun 2024 14:32:51 +0200 Janosch Frank <[email protected]> wrote:
Quoted 13 lines · Paolo Bonzini
Quoted 10 lines · Christoph Schlameuss

This change rejects the KVM_SET_USER_MEMORY_REGION and KVM_SET_USER_MEMORY_REGION2 ioctls when called on a ucontrol VM. This is neccessary since ucontrol VMs have kvm->arch.gmap set to 0 and would thus result in a null pointer dereference further in. Memory management needs to be performed in userspace and using the ioctls KVM_S390_UCAS_MAP and KVM_S390_UCAS_UNMAP.

Also improve s390 specific documentation for KVM_SET_USER_MEMORY_REGION and KVM_SET_USER_MEMORY_REGION2.

Quoted 7 lines
Quoted 4 lines

Would be nice to have a selftest for ucontrol VMs, too... just saying :)

Paolo

Already in the works, he just hasn't posted it yet :) We did do a couple rounds of internal feedback on the tests first.

I do also have a test case for this specifically, but it depends on the base fixture. So I would send it together with that soon.

Christoph