KVM: x86: fix escape of guest dr6 to the host
authorWanpeng Li <wanpeng.li@hotmail.com>
Wed, 13 Dec 2017 09:46:40 +0000 (10:46 +0100)
committerRadim Krčmář <rkrcmar@redhat.com>
Tue, 16 Jan 2018 15:34:13 +0000 (16:34 +0100)
syzkaller reported:

   WARNING: CPU: 0 PID: 12927 at arch/x86/kernel/traps.c:780 do_debug+0x222/0x250
   CPU: 0 PID: 12927 Comm: syz-executor Tainted: G           OE    4.15.0-rc2+ #16
   RIP: 0010:do_debug+0x222/0x250
   Call Trace:
    <#DB>
    debug+0x3e/0x70
   RIP: 0010:copy_user_enhanced_fast_string+0x10/0x20
    </#DB>
    _copy_from_user+0x5b/0x90
    SyS_timer_create+0x33/0x80
    entry_SYSCALL_64_fastpath+0x23/0x9a

The testcase sets a watchpoint (with perf_event_open) on a buffer that is
passed to timer_create() as the struct sigevent argument.  In timer_create(),
copy_from_user()'s rep movsb triggers the BP.  The testcase also sets
the debug registers for the guest.

However, KVM only restores host debug registers when the host has active
watchpoints, which triggers a race condition when running the testcase with
multiple threads.  The guest's DR6.BS bit can escape to the host before
another thread invokes timer_create(), and do_debug() complains.

The fix is to respect do_debug()'s dr6 invariant when leaving KVM.

Reported-by: Dmitry Vyukov <dvyukov@google.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Wanpeng Li <wanpeng.li@hotmail.com>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
arch/x86/kvm/x86.c

index 35d0b3de697bc5e1d6f95c40cd18e06d8c937173..a3dd44bb6f1e61b662c1a52bab7d14a3c57aa6bf 100644 (file)
@@ -2961,6 +2961,12 @@ void kvm_arch_vcpu_put(struct kvm_vcpu *vcpu)
        pagefault_enable();
        kvm_x86_ops->vcpu_put(vcpu);
        vcpu->arch.last_host_tsc = rdtsc();
+       /*
+        * If userspace has set any breakpoints or watchpoints, dr6 is restored
+        * on every vmexit, but if not, we might have a stale dr6 from the
+        * guest. do_debug expects dr6 to be cleared after it runs, do the same.
+        */
+       set_debugreg(0, 6);
 }
 
 static int kvm_vcpu_ioctl_get_lapic(struct kvm_vcpu *vcpu,