KVM: LAPIC: Fix pv ipis out-of-bounds access
authorWanpeng Li <wanpengli@tencent.com>
Thu, 30 Aug 2018 02:03:30 +0000 (10:03 +0800)
committerRadim Krčmář <rkrcmar@redhat.com>
Fri, 7 Sep 2018 16:38:43 +0000 (18:38 +0200)
Dan Carpenter reported that the untrusted data returns from kvm_register_read()
results in the following static checker warning:
  arch/x86/kvm/lapic.c:576 kvm_pv_send_ipi()
  error: buffer underflow 'map->phys_map' 's32min-s32max'

KVM guest can easily trigger this by executing the following assembly sequence
in Ring0:

mov $10, %rax
mov $0xFFFFFFFF, %rbx
mov $0xFFFFFFFF, %rdx
mov $0, %rsi
vmcall

As this will cause KVM to execute the following code-path:
vmx_handle_exit() -> handle_vmcall() -> kvm_emulate_hypercall() -> kvm_pv_send_ipi()
which will reach out-of-bounds access.

This patch fixes it by adding a check to kvm_pv_send_ipi() against map->max_apic_id,
ignoring destinations that are not present and delivering the rest. We also check
whether or not map->phys_map[min + i] is NULL since the max_apic_id is set to the
max apic id, some phys_map maybe NULL when apic id is sparse, especially kvm
unconditionally set max_apic_id to 255 to reserve enough space for any xAPIC ID.

Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Liran Alon <liran.alon@oracle.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Liran Alon <liran.alon@oracle.com>
Cc: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Wanpeng Li <wanpengli@tencent.com>
[Add second "if (min > map->max_apic_id)" to complete the fix. -Radim]
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
arch/x86/include/asm/kvm_host.h
arch/x86/kvm/lapic.c

index 3ad10f634d4c3fe2ce71996355ddf38ca4e465c0..8e90488c3d56895f62080666e2141df1731092f3 100644 (file)
@@ -1455,7 +1455,7 @@ void kvm_vcpu_reset(struct kvm_vcpu *vcpu, bool init_event);
 void kvm_vcpu_reload_apic_access_page(struct kvm_vcpu *vcpu);
 
 int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
-                   unsigned long ipi_bitmap_high, int min,
+                   unsigned long ipi_bitmap_high, u32 min,
                    unsigned long icr, int op_64_bit);
 
 u64 kvm_get_arch_capabilities(void);
index 0cefba28c864a3a0925378ed9b438a05b554cff2..17c0472c5b344faaaac3153ff53c4600a3fcd81e 100644 (file)
@@ -548,7 +548,7 @@ int kvm_apic_set_irq(struct kvm_vcpu *vcpu, struct kvm_lapic_irq *irq,
 }
 
 int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
-                   unsigned long ipi_bitmap_high, int min,
+                   unsigned long ipi_bitmap_high, u32 min,
                    unsigned long icr, int op_64_bit)
 {
        int i;
@@ -571,18 +571,31 @@ int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
        rcu_read_lock();
        map = rcu_dereference(kvm->arch.apic_map);
 
+       if (min > map->max_apic_id)
+               goto out;
        /* Bits above cluster_size are masked in the caller.  */
-       for_each_set_bit(i, &ipi_bitmap_low, BITS_PER_LONG) {
-               vcpu = map->phys_map[min + i]->vcpu;
-               count += kvm_apic_set_irq(vcpu, &irq, NULL);
+       for_each_set_bit(i, &ipi_bitmap_low,
+               min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
+               if (map->phys_map[min + i]) {
+                       vcpu = map->phys_map[min + i]->vcpu;
+                       count += kvm_apic_set_irq(vcpu, &irq, NULL);
+               }
        }
 
        min += cluster_size;
-       for_each_set_bit(i, &ipi_bitmap_high, BITS_PER_LONG) {
-               vcpu = map->phys_map[min + i]->vcpu;
-               count += kvm_apic_set_irq(vcpu, &irq, NULL);
+
+       if (min > map->max_apic_id)
+               goto out;
+
+       for_each_set_bit(i, &ipi_bitmap_high,
+               min((u32)BITS_PER_LONG, (map->max_apic_id - min + 1))) {
+               if (map->phys_map[min + i]) {
+                       vcpu = map->phys_map[min + i]->vcpu;
+                       count += kvm_apic_set_irq(vcpu, &irq, NULL);
+               }
        }
 
+out:
        rcu_read_unlock();
        return count;
 }