KVM: x86: fix singlestepping over syscall
authorPaolo Bonzini <pbonzini@redhat.com>
Wed, 7 Jun 2017 13:13:14 +0000 (15:13 +0200)
committerRadim Krčmář <rkrcmar@redhat.com>
Thu, 22 Jun 2017 14:13:29 +0000 (16:13 +0200)
TF is handled a bit differently for syscall and sysret, compared
to the other instructions: TF is checked after the instruction completes,
so that the OS can disable #DB at a syscall by adding TF to FMASK.
When the sysret is executed the #DB is taken "as if" the syscall insn
just completed.

KVM emulates syscall so that it can trap 32-bit syscall on Intel processors.
Fix the behavior, otherwise you could get #DB on a user stack which is not
nice.  This does not affect Linux guests, as they use an IST or task gate
for #DB.

This fixes CVE-2017-7518.

Cc: stable@vger.kernel.org
Reported-by: Andy Lutomirski <luto@kernel.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
Signed-off-by: Radim Krčmář <rkrcmar@redhat.com>
arch/x86/include/asm/kvm_emulate.h
arch/x86/kvm/emulate.c
arch/x86/kvm/x86.c

index 0559626..722d0e5 100644 (file)
@@ -296,6 +296,7 @@ struct x86_emulate_ctxt {
 
        bool perm_ok; /* do not check permissions if true */
        bool ud;        /* inject an #UD if host doesn't support insn */
+       bool tf;        /* TF value before instruction (after for syscall/sysret) */
 
        bool have_exception;
        struct x86_exception exception;
index 0816ab2..80890de 100644 (file)
@@ -2742,6 +2742,7 @@ static int em_syscall(struct x86_emulate_ctxt *ctxt)
                ctxt->eflags &= ~(X86_EFLAGS_VM | X86_EFLAGS_IF);
        }
 
+       ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
        return X86EMUL_CONTINUE;
 }
 
index 87d3cb9..0e846f0 100644 (file)
@@ -5313,6 +5313,8 @@ static void init_emulate_ctxt(struct kvm_vcpu *vcpu)
        kvm_x86_ops->get_cs_db_l_bits(vcpu, &cs_db, &cs_l);
 
        ctxt->eflags = kvm_get_rflags(vcpu);
+       ctxt->tf = (ctxt->eflags & X86_EFLAGS_TF) != 0;
+
        ctxt->eip = kvm_rip_read(vcpu);
        ctxt->mode = (!is_protmode(vcpu))               ? X86EMUL_MODE_REAL :
                     (ctxt->eflags & X86_EFLAGS_VM)     ? X86EMUL_MODE_VM86 :
@@ -5528,36 +5530,25 @@ static int kvm_vcpu_check_hw_bp(unsigned long addr, u32 type, u32 dr7,
        return dr6;
 }
 
-static void kvm_vcpu_check_singlestep(struct kvm_vcpu *vcpu, unsigned long rflags, int *r)
+static void kvm_vcpu_do_singlestep(struct kvm_vcpu *vcpu, int *r)
 {
        struct kvm_run *kvm_run = vcpu->run;
 
-       /*
-        * rflags is the old, "raw" value of the flags.  The new value has
-        * not been saved yet.
-        *
-        * This is correct even for TF set by the guest, because "the
-        * processor will not generate this exception after the instruction
-        * that sets the TF flag".
-        */
-       if (unlikely(rflags & X86_EFLAGS_TF)) {
-               if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
-                       kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 |
-                                                 DR6_RTM;
-                       kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
-                       kvm_run->debug.arch.exception = DB_VECTOR;
-                       kvm_run->exit_reason = KVM_EXIT_DEBUG;
-                       *r = EMULATE_USER_EXIT;
-               } else {
-                       /*
-                        * "Certain debug exceptions may clear bit 0-3.  The
-                        * remaining contents of the DR6 register are never
-                        * cleared by the processor".
-                        */
-                       vcpu->arch.dr6 &= ~15;
-                       vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
-                       kvm_queue_exception(vcpu, DB_VECTOR);
-               }
+       if (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP) {
+               kvm_run->debug.arch.dr6 = DR6_BS | DR6_FIXED_1 | DR6_RTM;
+               kvm_run->debug.arch.pc = vcpu->arch.singlestep_rip;
+               kvm_run->debug.arch.exception = DB_VECTOR;
+               kvm_run->exit_reason = KVM_EXIT_DEBUG;
+               *r = EMULATE_USER_EXIT;
+       } else {
+               /*
+                * "Certain debug exceptions may clear bit 0-3.  The
+                * remaining contents of the DR6 register are never
+                * cleared by the processor".
+                */
+               vcpu->arch.dr6 &= ~15;
+               vcpu->arch.dr6 |= DR6_BS | DR6_RTM;
+               kvm_queue_exception(vcpu, DB_VECTOR);
        }
 }
 
@@ -5567,7 +5558,17 @@ int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
        int r = EMULATE_DONE;
 
        kvm_x86_ops->skip_emulated_instruction(vcpu);
-       kvm_vcpu_check_singlestep(vcpu, rflags, &r);
+
+       /*
+        * rflags is the old, "raw" value of the flags.  The new value has
+        * not been saved yet.
+        *
+        * This is correct even for TF set by the guest, because "the
+        * processor will not generate this exception after the instruction
+        * that sets the TF flag".
+        */
+       if (unlikely(rflags & X86_EFLAGS_TF))
+               kvm_vcpu_do_singlestep(vcpu, &r);
        return r == EMULATE_DONE;
 }
 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
@@ -5726,8 +5727,9 @@ restart:
                toggle_interruptibility(vcpu, ctxt->interruptibility);
                vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
                kvm_rip_write(vcpu, ctxt->eip);
-               if (r == EMULATE_DONE)
-                       kvm_vcpu_check_singlestep(vcpu, rflags, &r);
+               if (r == EMULATE_DONE &&
+                   (ctxt->tf || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
+                       kvm_vcpu_do_singlestep(vcpu, &r);
                if (!ctxt->have_exception ||
                    exception_type(ctxt->exception.vector) == EXCPT_TRAP)
                        __kvm_set_rflags(vcpu, ctxt->eflags);