KVM: ensure __gfn_to_pfn_memslot initializes *writable
authorPaolo Bonzini <pbonzini@redhat.com>
Tue, 23 Feb 2016 14:36:01 +0000 (15:36 +0100)
committerPaolo Bonzini <pbonzini@redhat.com>
Fri, 4 Mar 2016 11:35:20 +0000 (12:35 +0100)
For the kvm_is_error_hva, ubsan complains if the uninitialized writable
is passed to __direct_map, even though the value itself is not used
(__direct_map goes to mmu_set_spte->set_spte->set_mmio_spte but never
looks at that argument).

Ensuring that __gfn_to_pfn_memslot initializes *writable is cheap and
avoids this kind of issue.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
virt/kvm/kvm_main.c

index 39c36d4..1eae052 100644 (file)
@@ -1434,11 +1434,17 @@ kvm_pfn_t __gfn_to_pfn_memslot(struct kvm_memory_slot *slot, gfn_t gfn,
 {
        unsigned long addr = __gfn_to_hva_many(slot, gfn, NULL, write_fault);
 
-       if (addr == KVM_HVA_ERR_RO_BAD)
+       if (addr == KVM_HVA_ERR_RO_BAD) {
+               if (writable)
+                       *writable = false;
                return KVM_PFN_ERR_RO_FAULT;
+       }
 
-       if (kvm_is_error_hva(addr))
+       if (kvm_is_error_hva(addr)) {
+               if (writable)
+                       *writable = false;
                return KVM_PFN_NOSLOT;
+       }
 
        /* Do not map writable pfn in the readonly memslot. */
        if (writable && memslot_is_readonly(slot)) {