KVM: x86: avoid calling x86 emulator without a decoded instruction

Whenever x86_decode_emulated_instruction() detects a breakpoint, it
returns the value that kvm_vcpu_check_breakpoint() writes into its
pass-by-reference second argument.  Unfortunately this is completely
bogus because the expected outcome of x86_decode_emulated_instruction
is an EMULATION_* value.

Then, if kvm_vcpu_check_breakpoint() does "*r = 0" (corresponding to
a KVM_EXIT_DEBUG userspace exit), it is misunderstood as EMULATION_OK
and x86_emulate_instruction() is called without having decoded the
instruction.  This causes various havoc from running with a stale
emulation context.

The fix is to move the call to kvm_vcpu_check_breakpoint() where it was
before commit 4aa2691dcbd3 ("KVM: x86: Factor out x86 instruction
emulation with decoding") introduced x86_decode_emulated_instruction().
The other caller of the function does not need breakpoint checks,
because it is invoked as part of a vmexit and the processor has already
checked those before executing the instruction that #GP'd.

This fixes CVE-2022-1852.

Reported-by: Qiuhao Li <qiuhao@sysec.org>
Reported-by: Gaoning Pan <pgn@zju.edu.cn>
Reported-by: Yongkang Jia <kangel@zju.edu.cn>
Fixes: 4aa2691dcbd3 ("KVM: x86: Factor out x86 instruction emulation with decoding")
Cc: stable@vger.kernel.org
Signed-off-by: Sean Christopherson <seanjc@google.com>
Message-Id: <20220311032801.3467418-2-seanjc@google.com>
[Rewrote commit message according to Qiuhao's report, since a patch
 already existed to fix the bug. - Paolo]
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>

1 files changed, 19 insertions, 12 deletions

diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
index bc507d6414f4..b81ef4f497f4 100644
--- a/arch/x86/kvm/x86.c
+++ b/arch/x86/kvm/x86.c
@@ -8296,7 +8296,7 @@ int kvm_skip_emulated_instruction(struct kvm_vcpu *vcpu)
 }
 EXPORT_SYMBOL_GPL(kvm_skip_emulated_instruction);
 
-static bool kvm_vcpu_check_breakpoint(struct kvm_vcpu *vcpu, int *r)
+static bool kvm_vcpu_check_code_breakpoint(struct kvm_vcpu *vcpu, int *r)
 {
 	if (unlikely(vcpu->guest_debug & KVM_GUESTDBG_USE_HW_BP) &&
 	    (vcpu->arch.guest_debug_dr7 & DR7_BP_EN_MASK)) {
@@ -8365,25 +8365,23 @@ static bool is_vmware_backdoor_opcode(struct x86_emulate_ctxt *ctxt)
 }
 
 /*
- * Decode to be emulated instruction. Return EMULATION_OK if success.
+ * Decode an instruction for emulation.  The caller is responsible for handling
+ * code breakpoints.  Note, manually detecting code breakpoints is unnecessary
+ * (and wrong) when emulating on an intercepted fault-like exception[*], as
+ * code breakpoints have higher priority and thus have already been done by
+ * hardware.
+ *
+ * [*] Except #MC, which is higher priority, but KVM should never emulate in
+ *     response to a machine check.
  */
 int x86_decode_emulated_instruction(struct kvm_vcpu *vcpu, int emulation_type,
 				    void *insn, int insn_len)
 {
-	int r = EMULATION_OK;
 	struct x86_emulate_ctxt *ctxt = vcpu->arch.emulate_ctxt;
+	int r;
 
 	init_emulate_ctxt(vcpu);
 
-	/*
-	 * We will reenter on the same instruction since we do not set
-	 * complete_userspace_io. This does not handle watchpoints yet,
-	 * those would be handled in the emulate_ops.
-	 */
-	if (!(emulation_type & EMULTYPE_SKIP) &&
-	    kvm_vcpu_check_breakpoint(vcpu, &r))
-		return r;
-
 	r = x86_decode_insn(ctxt, insn, insn_len, emulation_type);
 
 	trace_kvm_emulate_insn_start(vcpu);
@@ -8416,6 +8414,15 @@ int x86_emulate_instruction(struct kvm_vcpu *vcpu, gpa_t cr2_or_gpa,
 	if (!(emulation_type & EMULTYPE_NO_DECODE)) {
 		kvm_clear_exception_queue(vcpu);
 
+		/*
+		 * Return immediately if RIP hits a code breakpoint, such #DBs
+		 * are fault-like and are higher priority than any faults on
+		 * the code fetch itself.
+		 */
+		if (!(emulation_type & EMULTYPE_SKIP) &&
+		    kvm_vcpu_check_code_breakpoint(vcpu, &r))
+			return r;
+
 		r = x86_decode_emulated_instruction(vcpu, emulation_type,
 						    insn, insn_len);
 		if (r != EMULATION_OK)  {