Add KVM test for CPU lockup through malicous SVM guest

Signed-off-by: Martin Doucha <[email protected]>
Acked-by: Petr Vorel <[email protected]>
Acked-by: Cyril Hrubis <[email protected]>

Author: mdoucha

runtest/kvm

@@ -1,3 +1,4 @@
 kvm_pagefault01 kvm_pagefault01
 kvm_svm01 kvm_svm01
 kvm_svm02 kvm_svm02
+kvm_svm03 kvm_svm03

testcases/kernel/kvm/.gitignore

@@ -1,3 +1,4 @@
 /kvm_pagefault01
 /kvm_svm01
 /kvm_svm02
+/kvm_svm03

testcases/kernel/kvm/Makefile

@@ -48,6 +48,9 @@ endif
 lib_guest.o $(ARCH_OBJ): CPPFLAGS	:= $(GUEST_CPPFLAGS)
 lib_guest.o $(ARCH_OBJ): CFLAGS		:= $(GUEST_CFLAGS)
 
+kvm_svm03: CFLAGS += -pthread
+kvm_svm03: LDLIBS += -pthread
+
 include $(top_srcdir)/include/mk/generic_leaf_target.mk
 
 %-payload.o: %.c lib_guest.o $(ARCH_OBJ)

testcases/kernel/kvm/kvm_svm03.c

@@ -0,0 +1,169 @@
+// SPDX-License-Identifier: GPL-2.0-or-later
+/*
+ * Copyright (C) 2023 SUSE LLC
+ * Author: Nicolai Stange <[email protected]>
+ * LTP port: Martin Doucha <[email protected]>
+ */
+
+/*\
+ * Check that KVM correctly intercepts the CLGI instruction in a nested
+ * virtual machine even when the parent guest disables intercept.
+ * If KVM does not override the disabled intercept, it'll allow the nested VM
+ * to hold the physical CPU indefinitely and potentially perform a denial
+ * of service attack against the host kernel. CPU lockup fixed in:
+ *
+ *  commit 91b7130cb6606d8c6b3b77e54426b3f3a83f48b1
+ *  Author: Paolo Bonzini <[email protected]>
+ *  Date:   Fri May 22 12:28:52 2020 -0400
+ *
+ *  KVM: SVM: preserve VGIF across VMCB switch
+ */
+
+#include "kvm_test.h"
+
+#ifdef COMPILE_PAYLOAD
+#if defined(__i386__) || defined(__x86_64__)
+
+#include "kvm_x86_svm.h"
+
+/* Disable global interrupts */
+static int guest_clgi(void)
+{
+	int ret, *result = (int *)KVM_RESULT_BASEADDR;
+
+	/*
+	 * Make sure that result page is present in memory. CLGI may disable
+	 * page fault handling on the current CPU. The actual value
+	 * at that address is irrelevant.
+	 */
+	ret = *result;
+
+	/* Disable global interrupts */
+	asm ("clgi");
+
+	/* Signal host to kill the VM and wait */
+	tst_wait_host(NULL);
+	return ret;
+}
+
+void main(void)
+{
+	struct kvm_svm_vcpu *vcpu;
+
+	kvm_init_svm();
+	vcpu = kvm_create_svm_vcpu(guest_clgi, 1);
+	kvm_vmcb_set_intercept(vcpu->vmcb, SVM_INTERCEPT_CLGI, 0);
+	kvm_svm_vmrun(vcpu);
+
+	if (vcpu->vmcb->exitcode != SVM_EXIT_HLT)
+		tst_brk(TBROK, "Nested VM exited unexpectedly");
+}
+
+#else /* defined(__i386__) || defined(__x86_64__) */
+TST_TEST_TCONF("Test supported only on x86");
+#endif /* defined(__i386__) || defined(__x86_64__) */
+
+#else /* COMPILE_PAYLOAD */
+
+#include <pthread.h>
+#include "tst_safe_pthread.h"
+#include "tst_safe_clocks.h"
+
+static struct tst_kvm_instance test_vm = { .vm_fd = -1 };
+static pthread_mutex_t mutex;
+static int mutex_init;
+
+static void sighandler(int sig LTP_ATTRIBUTE_UNUSED)
+{
+
+}
+
+static void *vm_thread(void *arg)
+{
+	SAFE_PTHREAD_MUTEX_LOCK(&mutex);
+	tst_kvm_run_instance(&test_vm, EINTR);
+	SAFE_PTHREAD_MUTEX_UNLOCK(&mutex);
+	return arg;
+}
+
+static void setup(void)
+{
+	struct sigaction sa = { .sa_handler = sighandler };
+	pthread_mutexattr_t attr;
+
+	SAFE_PTHREAD_MUTEXATTR_INIT(&attr);
+	SAFE_PTHREAD_MUTEXATTR_SETTYPE(&attr, PTHREAD_MUTEX_NORMAL);
+	SAFE_PTHREAD_MUTEX_INIT(&mutex, &attr);
+	mutex_init = 1;
+	SAFE_PTHREAD_MUTEXATTR_DESTROY(&attr);
+	SAFE_SIGACTION(SIGUSR1, &sa, NULL);
+}
+
+static void run(void)
+{
+	struct timespec timeout;
+	pthread_t tid;
+	int ret;
+
+	tst_kvm_create_instance(&test_vm, DEFAULT_RAM_SIZE);
+
+	SAFE_PTHREAD_CREATE(&tid, NULL, vm_thread, NULL);
+	ret = tst_kvm_wait_guest(&test_vm, 2000);
+
+	if (ret == KVM_TEXIT) {
+		SAFE_PTHREAD_JOIN(tid, NULL);
+		tst_brk(TCONF, "Guest exited early");
+	}
+
+	if (ret)
+		tst_brk(TBROK, "Wait for guest initialization timed out");
+
+	SAFE_PTHREAD_KILL(tid, SIGUSR1);
+	SAFE_CLOCK_GETTIME(CLOCK_REALTIME, &timeout);
+	timeout.tv_sec += 2;
+
+	if (SAFE_PTHREAD_MUTEX_TIMEDLOCK(&mutex, &timeout)) {
+		tst_kvm_clear_guest_signal(&test_vm);
+		tst_res(TFAIL, "VM thread does not respond to signals");
+	} else {
+		SAFE_PTHREAD_MUTEX_UNLOCK(&mutex);
+		tst_res(TPASS, "VM thread was interrupted by signal");
+	}
+
+	SAFE_PTHREAD_JOIN(tid, NULL);
+	tst_kvm_destroy_instance(&test_vm);
+	tst_free_all();
+}
+
+static void cleanup(void)
+{
+	/*
+	 * If the mutex is locked, the VM is likely still running, cannot
+	 * clean up anything
+	 */
+	if (!mutex_init || SAFE_PTHREAD_MUTEX_TRYLOCK(&mutex))
+		return;
+
+	if (!SAFE_PTHREAD_MUTEX_UNLOCK(&mutex))
+		SAFE_PTHREAD_MUTEX_DESTROY(&mutex);
+
+	tst_kvm_destroy_instance(&test_vm);
+}
+
+static struct tst_test test = {
+	.test_all = run,
+	.setup = setup,
+	.cleanup = cleanup,
+	.min_cpus = 2,
+	.supported_archs = (const char *const []) {
+		"x86_64",
+		"x86",
+		NULL
+	},
+	.tags = (struct tst_tag[]){
+		{"linux-git", "91b7130cb660"},
+		{}
+	}
+};
+
+#endif /* COMPILE_PAYLOAD */