KVM: x86: use kvm_set_cr3/cr4 in ioctl_set_sregs

Matt T. Yourst notes that kvm_arch_vcpu_ioctl_set_sregs lacks validity checking for the new cr3 value:

"Userspace callers of KVM_SET_SREGS can pass a bogus value of cr3 to the kernel. This will trigger a NULL pointer access in gfn_to_rmap() when userspace next tries to call KVM_RUN on the affected VCPU and kvm attempts to activate the new non-existent page table root.

This happens since kvm only validates that cr3 points to a valid guest physical memory page when code *inside* the guest sets cr3. However, kvm currently trusts the userspace caller (e.g. QEMU) on the host machine to always supply a valid page table root, rather than properly validating it along with the rest of the reloaded guest state."

http://sourceforge.net/tracker/?func=detail&atid=893831&aid=2687641&group_id=180599

Follow Avi's suggestion to use kvm_set_cr3, and do the same for assigment of cr4. Note kvm_set_cr4 unconditionally resets the mmu context, as long as cr4 is valid.

Signed-off-by Marcelo Tosatti <[email protected]>

arch/x86/kvm/x86.c +1 -4

index 148cde2..89fb3c7 100644
@@ -3985,25 +3985,19 @@ int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 	kvm_x86_ops->set_gdt(vcpu, &dt);
 
 	vcpu->arch.cr2 = sregs->cr2;
-	mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
-	vcpu->arch.cr3 = sregs->cr3;
 
+	kvm_set_cr3(vcpu, sregs->cr3);
 	kvm_set_cr8(vcpu, sregs->cr8);
 
 	mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
 	kvm_set_apic_base(vcpu, sregs->apic_base);
 
-	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
-
 	mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;
 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);

Quoted 40 lines · Marcelo Tosatti

Matt T. Yourst notes that kvm_arch_vcpu_ioctl_set_sregs lacks validity checking for the new cr3 value:

"Userspace callers of KVM_SET_SREGS can pass a bogus value of cr3 to the kernel. This will trigger a NULL pointer access in gfn_to_rmap() when userspace next tries to call KVM_RUN on the affected VCPU and kvm attempts to activate the new non-existent page table root.

This happens since kvm only validates that cr3 points to a valid guest physical memory page when code *inside* the guest sets cr3. However, kvm currently trusts the userspace caller (e.g. QEMU) on the host machine to always supply a valid page table root, rather than properly validating it along with the rest of the reloaded guest state."

http://sourceforge.net/tracker/?func=detail&atid=893831&aid=2687641&group_id=180599

Follow Avi's suggestion to use kvm_set_cr3, and do the same for assigment of cr4. Note kvm_set_cr4 unconditionally resets the mmu context, as long as cr4 is valid.

Signed-off-by Marcelo Tosatti <[email protected]>

arch/x86/kvm/x86.c +0 -3

index 148cde2..89fb3c7 100644
@@ -3985,25 +3985,19 @@ int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 	kvm_x86_ops->set_gdt(vcpu, &dt);
  	vcpu->arch.cr2 = sregs->cr2;
-	mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
-	vcpu->arch.cr3 = sregs->cr3;
 +	kvm_set_cr3(vcpu, sregs->cr3);
 	kvm_set_cr8(vcpu, sregs->cr8);
  	mmu_reset_needed |= vcpu->arch.shadow_efer != sregs->efer;
 	kvm_x86_ops->set_efer(vcpu, sregs->efer);
 	kvm_set_apic_base(vcpu, sregs->apic_base);
 -	kvm_x86_ops->decache_cr4_guest_bits(vcpu);
-
 	mmu_reset_needed |= vcpu->arch.cr0 != sregs->cr0;

arch/x86/kvm/x86.c int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu, · line 4000

Quoted 10 lines

arch/x86/kvm/x86.c +1 -3

 	kvm_x86_ops->set_cr0(vcpu, sregs->cr0);
 	vcpu->arch.cr0 = sregs->cr0;
 -	mmu_reset_needed |= vcpu->arch.cr4 != sregs->cr4;
-	kvm_x86_ops->set_cr4(vcpu, sregs->cr4);
-	if (!is_long_mode(vcpu) && is_pae(vcpu))
-		load_pdptrs(vcpu, vcpu->arch.cr3);
+	kvm_set_cr4(vcpu, sregs->cr4);
  	if (mmu_reset_needed)
 		kvm_mmu_reset_context(vcpu);
 

Consider the following:

current state:
cr3 = 0
cr4.pae = 0

new state:
cr3 = 0x800
cr4.pae = 1

When you call kvm_set_cr3(), it will inject a #GP into the guest because we are setting bit 11 when cr4.pae=0, which is illegal. However the new cr4.pae=1, so the new state was in fact legal!

There are a few ways out, one is to first go back to real mode and set eveything up carefully in the right order (including EFER.LMA and EFER.LME, and CS.L). The other is to refactor kvm_set_* so that we have internal setters which won't trigger these faults (but do need to check at the end that the state is legal).

This first method is probably better since that's what the guest does when booting anyway.

Humpf. And something like this? Or GP# instead of triple fault?

arch/x86/kvm/x86.c +0 -0

index 148cde2..3e63bac 100644
@@ -3986,7 +3986,10 @@ int kvm_arch_vcpu_ioctl_set_sregs(struct kvm_vcpu *vcpu,
 

Matt T. Yourst notes that kvm_arch_vcpu_ioctl_set_sregs lacks validity checking for the new cr3 value:

"Userspace callers of KVM_SET_SREGS can pass a bogus value of cr3 to the kernel. This will trigger a NULL pointer access in gfn_to_rmap() when userspace next tries to call KVM_RUN on the affected VCPU and kvm attempts to activate the new non-existent page table root.

This happens since kvm only validates that cr3 points to a valid guest physical memory page when code *inside* the guest sets cr3. However, kvm currently trusts the userspace caller (e.g. QEMU) on the host machine to always supply a valid page table root, rather than properly validating it along with the rest of the reloaded guest state."

http://sourceforge.net/tracker/?func=detail&atid=893831&aid=2687641&group_id=180599

Check for a valid cr3 address in kvm_arch_vcpu_ioctl_set_sregs, triple fault in case of failure.

Signed-off-by Marcelo Tosatti <[email protected]>

Index: kvm/arch/x86/kvm/x86.c
===================================================================

arch/x86/kvm/x86.c +0 -1

@@ -3986,7 +3986,13 @@ int kvm_arch_vcpu_ioctl_set_sregs(struct
 
 	vcpu->arch.cr2 = sregs->cr2;
 	mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
-	vcpu->arch.cr3 = sregs->cr3;

Quoted 39 lines · Marcelo Tosatti

Matt T. Yourst notes that kvm_arch_vcpu_ioctl_set_sregs lacks validity checking for the new cr3 value:
"Userspace callers of KVM_SET_SREGS can pass a bogus value of
cr3 to
the kernel. This will trigger a NULL pointer access in gfn_to_rmap() when userspace next tries to call KVM_RUN on the affected VCPU and kvm attempts to activate the new non-existent page table root.

This happens since kvm only validates that cr3 points to a valid guest physical memory page when code *inside* the guest sets cr3. However, kvm currently trusts the userspace caller (e.g. QEMU) on the host machine to always supply a valid page table root, rather than properly validating it along with the rest of the reloaded guest state."

http://sourceforge.net/tracker/?func=detail&atid=893831&aid=2687641&group_id=180599

Check for a valid cr3 address in kvm_arch_vcpu_ioctl_set_sregs, triple fault in case of failure.

Signed-off-by Marcelo Tosatti <[email protected]>

Index: kvm/arch/x86/kvm/x86.c
===================================================================

arch/x86/kvm/x86.c +7 -1

@@ -3986,7 +3986,13 @@ int kvm_arch_vcpu_ioctl_set_sregs(struct
 
     vcpu->arch.cr2 = sregs->cr2;
     mmu_reset_needed |= vcpu->arch.cr3 != sregs->cr3;
-    vcpu->arch.cr3 = sregs->cr3;
+
+    down_read(&vcpu->kvm->slots_lock);
+    if (gfn_to_memslot(vcpu->kvm, sregs->cr3 >> PAGE_SHIFT))
+        vcpu->arch.cr3 = sregs->cr3;
+    else
+        set_bit(KVM_REQ_TRIPLE_FAULT, &vcpu->requests);
+    up_read(&vcpu->kvm->slots_lock);
 

arch/x86/kvm/x86.c int kvm_arch_vcpu_ioctl_set_sregs(struct · line 3998

Quoted 3 lines

arch/x86/kvm/x86.c +0 -0

     kvm_set_cr8(vcpu, sregs->cr8);
 

Isn't this self-defeating? If you drop slots_lock, cr3 may be invalid again by the time you set cvpu->arch.cr3?