1/*
  2 * mmu_audit.c:
  3 *
  4 * Audit code for KVM MMU
  5 *
  6 * Copyright (C) 2006 Qumranet, Inc.
  7 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
  8 *
  9 * Authors:
 10 *   Yaniv Kamay  <yaniv@qumranet.com>
 11 *   Avi Kivity   <avi@qumranet.com>
 12 *   Marcelo Tosatti <mtosatti@redhat.com>
 13 *   Xiao Guangrong <xiaoguangrong@cn.fujitsu.com>
 14 *
 15 * This work is licensed under the terms of the GNU GPL, version 2.  See
 16 * the COPYING file in the top-level directory.
 17 *
 18 */
 19
 20#include <linux/ratelimit.h>
 21
 22#define audit_printk(kvm, fmt, args...)		\
 23	printk(KERN_ERR "audit: (%s) error: "	\
 24		fmt, audit_point_name[kvm->arch.audit_point], ##args)
 25
 26typedef void (*inspect_spte_fn) (struct kvm_vcpu *vcpu, u64 *sptep, int level);
 27
 28static void __mmu_spte_walk(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
 29			    inspect_spte_fn fn, int level)
 30{
 31	int i;
 32
 33	for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
 34		u64 *ent = sp->spt;
 35
 36		fn(vcpu, ent + i, level);
 37
 38		if (is_shadow_present_pte(ent[i]) &&
 39		      !is_last_spte(ent[i], level)) {
 40			struct kvm_mmu_page *child;
 41
 42			child = page_header(ent[i] & PT64_BASE_ADDR_MASK);
 43			__mmu_spte_walk(vcpu, child, fn, level - 1);
 44		}
 45	}
 46}
 47
 48static void mmu_spte_walk(struct kvm_vcpu *vcpu, inspect_spte_fn fn)
 49{
 50	int i;
 51	struct kvm_mmu_page *sp;
 52
 53	if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
 54		return;
 55
 56	if (vcpu->arch.mmu.root_level == PT64_ROOT_LEVEL) {
 57		hpa_t root = vcpu->arch.mmu.root_hpa;
 58
 59		sp = page_header(root);
 60		__mmu_spte_walk(vcpu, sp, fn, PT64_ROOT_LEVEL);
 61		return;
 62	}
 63
 64	for (i = 0; i < 4; ++i) {
 65		hpa_t root = vcpu->arch.mmu.pae_root[i];
 66
 67		if (root && VALID_PAGE(root)) {
 68			root &= PT64_BASE_ADDR_MASK;
 69			sp = page_header(root);
 70			__mmu_spte_walk(vcpu, sp, fn, 2);
 71		}
 72	}
 73
 74	return;
 75}
 76
 77typedef void (*sp_handler) (struct kvm *kvm, struct kvm_mmu_page *sp);
 78
 79static void walk_all_active_sps(struct kvm *kvm, sp_handler fn)
 80{
 81	struct kvm_mmu_page *sp;
 82
 83	list_for_each_entry(sp, &kvm->arch.active_mmu_pages, link)
 84		fn(kvm, sp);
 85}
 86
 87static void audit_mappings(struct kvm_vcpu *vcpu, u64 *sptep, int level)
 88{
 89	struct kvm_mmu_page *sp;
 90	gfn_t gfn;
 91	pfn_t pfn;
 92	hpa_t hpa;
 93
 94	sp = page_header(__pa(sptep));
 95
 96	if (sp->unsync) {
 97		if (level != PT_PAGE_TABLE_LEVEL) {
 98			audit_printk(vcpu->kvm, "unsync sp: %p "
 99				     "level = %d\n", sp, level);
100			return;
101		}
102	}
103
104	if (!is_shadow_present_pte(*sptep) || !is_last_spte(*sptep, level))
105		return;
106
107	gfn = kvm_mmu_page_get_gfn(sp, sptep - sp->spt);
108	pfn = gfn_to_pfn_atomic(vcpu->kvm, gfn);
109
110	if (is_error_pfn(pfn)) {
111		kvm_release_pfn_clean(pfn);
112		return;
113	}
114
115	hpa =  pfn << PAGE_SHIFT;
116	if ((*sptep & PT64_BASE_ADDR_MASK) != hpa)
117		audit_printk(vcpu->kvm, "levels %d pfn %llx hpa %llx "
118			     "ent %llxn", vcpu->arch.mmu.root_level, pfn,
119			     hpa, *sptep);
120}
121
122static void inspect_spte_has_rmap(struct kvm *kvm, u64 *sptep)
123{
124	unsigned long *rmapp;
125	struct kvm_mmu_page *rev_sp;
126	gfn_t gfn;
127
128
129	rev_sp = page_header(__pa(sptep));
130	gfn = kvm_mmu_page_get_gfn(rev_sp, sptep - rev_sp->spt);
131
132	if (!gfn_to_memslot(kvm, gfn)) {
133		if (!printk_ratelimit())
134			return;
135		audit_printk(kvm, "no memslot for gfn %llx\n", gfn);
136		audit_printk(kvm, "index %ld of sp (gfn=%llx)\n",
137		       (long int)(sptep - rev_sp->spt), rev_sp->gfn);
138		dump_stack();
139		return;
140	}
141
142	rmapp = gfn_to_rmap(kvm, gfn, rev_sp->role.level);
143	if (!*rmapp) {
144		if (!printk_ratelimit())
145			return;
146		audit_printk(kvm, "no rmap for writable spte %llx\n",
147			     *sptep);
148		dump_stack();
149	}
150}
151
152static void audit_sptes_have_rmaps(struct kvm_vcpu *vcpu, u64 *sptep, int level)
153{
154	if (is_shadow_present_pte(*sptep) && is_last_spte(*sptep, level))
155		inspect_spte_has_rmap(vcpu->kvm, sptep);
156}
157
158static void audit_spte_after_sync(struct kvm_vcpu *vcpu, u64 *sptep, int level)
159{
160	struct kvm_mmu_page *sp = page_header(__pa(sptep));
161
162	if (vcpu->kvm->arch.audit_point == AUDIT_POST_SYNC && sp->unsync)
163		audit_printk(vcpu->kvm, "meet unsync sp(%p) after sync "
164			     "root.\n", sp);
165}
166
167static void check_mappings_rmap(struct kvm *kvm, struct kvm_mmu_page *sp)
168{
169	int i;
170
171	if (sp->role.level != PT_PAGE_TABLE_LEVEL)
172		return;
173
174	for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
175		if (!is_rmap_spte(sp->spt[i]))
176			continue;
177
178		inspect_spte_has_rmap(kvm, sp->spt + i);
179	}
180}
181
182static void audit_write_protection(struct kvm *kvm, struct kvm_mmu_page *sp)
183{
184	struct kvm_memory_slot *slot;
185	unsigned long *rmapp;
186	u64 *spte;
187
188	if (sp->role.direct || sp->unsync || sp->role.invalid)
189		return;
190
191	slot = gfn_to_memslot(kvm, sp->gfn);
192	rmapp = &slot->rmap[sp->gfn - slot->base_gfn];
193
194	spte = rmap_next(kvm, rmapp, NULL);
195	while (spte) {
196		if (is_writable_pte(*spte))
197			audit_printk(kvm, "shadow page has writable "
198				     "mappings: gfn %llx role %x\n",
199				     sp->gfn, sp->role.word);
200		spte = rmap_next(kvm, rmapp, spte);
201	}
202}
203
204static void audit_sp(struct kvm *kvm, struct kvm_mmu_page *sp)
205{
206	check_mappings_rmap(kvm, sp);
207	audit_write_protection(kvm, sp);
208}
209
210static void audit_all_active_sps(struct kvm *kvm)
211{
212	walk_all_active_sps(kvm, audit_sp);
213}
214
215static void audit_spte(struct kvm_vcpu *vcpu, u64 *sptep, int level)
216{
217	audit_sptes_have_rmaps(vcpu, sptep, level);
218	audit_mappings(vcpu, sptep, level);
219	audit_spte_after_sync(vcpu, sptep, level);
220}
221
222static void audit_vcpu_spte(struct kvm_vcpu *vcpu)
223{
224	mmu_spte_walk(vcpu, audit_spte);
225}
226
227static void kvm_mmu_audit(void *ignore, struct kvm_vcpu *vcpu, int point)
228{
229	static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
230
231	if (!__ratelimit(&ratelimit_state))
232		return;
233
234	vcpu->kvm->arch.audit_point = point;
235	audit_all_active_sps(vcpu->kvm);
236	audit_vcpu_spte(vcpu);
237}
238
239static bool mmu_audit;
240
241static void mmu_audit_enable(void)
242{
243	int ret;
244
245	if (mmu_audit)
246		return;
247
248	ret = register_trace_kvm_mmu_audit(kvm_mmu_audit, NULL);
249	WARN_ON(ret);
250
251	mmu_audit = true;
252}
253
254static void mmu_audit_disable(void)
255{
256	if (!mmu_audit)
257		return;
258
259	unregister_trace_kvm_mmu_audit(kvm_mmu_audit, NULL);
260	tracepoint_synchronize_unregister();
261	mmu_audit = false;
262}
263
264static int mmu_audit_set(const char *val, const struct kernel_param *kp)
265{
266	int ret;
267	unsigned long enable;
268
269	ret = strict_strtoul(val, 10, &enable);
270	if (ret < 0)
271		return -EINVAL;
272
273	switch (enable) {
274	case 0:
275		mmu_audit_disable();
276		break;
277	case 1:
278		mmu_audit_enable();
279		break;
280	default:
281		return -EINVAL;
282	}
283
284	return 0;
285}
286
287static struct kernel_param_ops audit_param_ops = {
288	.set = mmu_audit_set,
289	.get = param_get_bool,
290};
291
292module_param_cb(mmu_audit, &audit_param_ops, &mmu_audit, 0644);