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
 22static char const *audit_point_name[] = {
 23	"pre page fault",
 24	"post page fault",
 25	"pre pte write",
 26	"post pte write",
 27	"pre sync",
 28	"post sync"
 29};
 30
 31#define audit_printk(kvm, fmt, args...)		\
 32	printk(KERN_ERR "audit: (%s) error: "	\
 33		fmt, audit_point_name[kvm->arch.audit_point], ##args)
 34
 35typedef void (*inspect_spte_fn) (struct kvm_vcpu *vcpu, u64 *sptep, int level);
 36
 37static void __mmu_spte_walk(struct kvm_vcpu *vcpu, struct kvm_mmu_page *sp,
 38			    inspect_spte_fn fn, int level)
 39{
 40	int i;
 41
 42	for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
 43		u64 *ent = sp->spt;
 44
 45		fn(vcpu, ent + i, level);
 46
 47		if (is_shadow_present_pte(ent[i]) &&
 48		      !is_last_spte(ent[i], level)) {
 49			struct kvm_mmu_page *child;
 50
 51			child = page_header(ent[i] & PT64_BASE_ADDR_MASK);
 52			__mmu_spte_walk(vcpu, child, fn, level - 1);
 53		}
 54	}
 55}
 56
 57static void mmu_spte_walk(struct kvm_vcpu *vcpu, inspect_spte_fn fn)
 58{
 59	int i;
 60	struct kvm_mmu_page *sp;
 61
 62	if (!VALID_PAGE(vcpu->arch.mmu.root_hpa))
 63		return;
 64
 65	if (vcpu->arch.mmu.root_level >= PT64_ROOT_4LEVEL) {
 66		hpa_t root = vcpu->arch.mmu.root_hpa;
 67
 68		sp = page_header(root);
 69		__mmu_spte_walk(vcpu, sp, fn, vcpu->arch.mmu.root_level);
 70		return;
 71	}
 72
 73	for (i = 0; i < 4; ++i) {
 74		hpa_t root = vcpu->arch.mmu.pae_root[i];
 75
 76		if (root && VALID_PAGE(root)) {
 77			root &= PT64_BASE_ADDR_MASK;
 78			sp = page_header(root);
 79			__mmu_spte_walk(vcpu, sp, fn, 2);
 80		}
 81	}
 82
 83	return;
 84}
 85
 86typedef void (*sp_handler) (struct kvm *kvm, struct kvm_mmu_page *sp);
 87
 88static void walk_all_active_sps(struct kvm *kvm, sp_handler fn)
 89{
 90	struct kvm_mmu_page *sp;
 91
 92	list_for_each_entry(sp, &kvm->arch.active_mmu_pages, link)
 93		fn(kvm, sp);
 94}
 95
 96static void audit_mappings(struct kvm_vcpu *vcpu, u64 *sptep, int level)
 97{
 98	struct kvm_mmu_page *sp;
 99	gfn_t gfn;
100	kvm_pfn_t pfn;
101	hpa_t hpa;
102
103	sp = page_header(__pa(sptep));
104
105	if (sp->unsync) {
106		if (level != PT_PAGE_TABLE_LEVEL) {
107			audit_printk(vcpu->kvm, "unsync sp: %p "
108				     "level = %d\n", sp, level);
109			return;
110		}
111	}
112
113	if (!is_shadow_present_pte(*sptep) || !is_last_spte(*sptep, level))
114		return;
115
116	gfn = kvm_mmu_page_get_gfn(sp, sptep - sp->spt);
117	pfn = kvm_vcpu_gfn_to_pfn_atomic(vcpu, gfn);
118
119	if (is_error_pfn(pfn))
120		return;
121
122	hpa =  pfn << PAGE_SHIFT;
123	if ((*sptep & PT64_BASE_ADDR_MASK) != hpa)
124		audit_printk(vcpu->kvm, "levels %d pfn %llx hpa %llx "
125			     "ent %llxn", vcpu->arch.mmu.root_level, pfn,
126			     hpa, *sptep);
127}
128
129static void inspect_spte_has_rmap(struct kvm *kvm, u64 *sptep)
130{
131	static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
132	struct kvm_rmap_head *rmap_head;
133	struct kvm_mmu_page *rev_sp;
134	struct kvm_memslots *slots;
135	struct kvm_memory_slot *slot;
136	gfn_t gfn;
137
138	rev_sp = page_header(__pa(sptep));
139	gfn = kvm_mmu_page_get_gfn(rev_sp, sptep - rev_sp->spt);
140
141	slots = kvm_memslots_for_spte_role(kvm, rev_sp->role);
142	slot = __gfn_to_memslot(slots, gfn);
143	if (!slot) {
144		if (!__ratelimit(&ratelimit_state))
145			return;
146		audit_printk(kvm, "no memslot for gfn %llx\n", gfn);
147		audit_printk(kvm, "index %ld of sp (gfn=%llx)\n",
148		       (long int)(sptep - rev_sp->spt), rev_sp->gfn);
149		dump_stack();
150		return;
151	}
152
153	rmap_head = __gfn_to_rmap(gfn, rev_sp->role.level, slot);
154	if (!rmap_head->val) {
155		if (!__ratelimit(&ratelimit_state))
156			return;
157		audit_printk(kvm, "no rmap for writable spte %llx\n",
158			     *sptep);
159		dump_stack();
160	}
161}
162
163static void audit_sptes_have_rmaps(struct kvm_vcpu *vcpu, u64 *sptep, int level)
164{
165	if (is_shadow_present_pte(*sptep) && is_last_spte(*sptep, level))
166		inspect_spte_has_rmap(vcpu->kvm, sptep);
167}
168
169static void audit_spte_after_sync(struct kvm_vcpu *vcpu, u64 *sptep, int level)
170{
171	struct kvm_mmu_page *sp = page_header(__pa(sptep));
172
173	if (vcpu->kvm->arch.audit_point == AUDIT_POST_SYNC && sp->unsync)
174		audit_printk(vcpu->kvm, "meet unsync sp(%p) after sync "
175			     "root.\n", sp);
176}
177
178static void check_mappings_rmap(struct kvm *kvm, struct kvm_mmu_page *sp)
179{
180	int i;
181
182	if (sp->role.level != PT_PAGE_TABLE_LEVEL)
183		return;
184
185	for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
186		if (!is_shadow_present_pte(sp->spt[i]))
187			continue;
188
189		inspect_spte_has_rmap(kvm, sp->spt + i);
190	}
191}
192
193static void audit_write_protection(struct kvm *kvm, struct kvm_mmu_page *sp)
194{
195	struct kvm_rmap_head *rmap_head;
196	u64 *sptep;
197	struct rmap_iterator iter;
198	struct kvm_memslots *slots;
199	struct kvm_memory_slot *slot;
200
201	if (sp->role.direct || sp->unsync || sp->role.invalid)
202		return;
203
204	slots = kvm_memslots_for_spte_role(kvm, sp->role);
205	slot = __gfn_to_memslot(slots, sp->gfn);
206	rmap_head = __gfn_to_rmap(sp->gfn, PT_PAGE_TABLE_LEVEL, slot);
207
208	for_each_rmap_spte(rmap_head, &iter, sptep) {
209		if (is_writable_pte(*sptep))
210			audit_printk(kvm, "shadow page has writable "
211				     "mappings: gfn %llx role %x\n",
212				     sp->gfn, sp->role.word);
213	}
214}
215
216static void audit_sp(struct kvm *kvm, struct kvm_mmu_page *sp)
217{
218	check_mappings_rmap(kvm, sp);
219	audit_write_protection(kvm, sp);
220}
221
222static void audit_all_active_sps(struct kvm *kvm)
223{
224	walk_all_active_sps(kvm, audit_sp);
225}
226
227static void audit_spte(struct kvm_vcpu *vcpu, u64 *sptep, int level)
228{
229	audit_sptes_have_rmaps(vcpu, sptep, level);
230	audit_mappings(vcpu, sptep, level);
231	audit_spte_after_sync(vcpu, sptep, level);
232}
233
234static void audit_vcpu_spte(struct kvm_vcpu *vcpu)
235{
236	mmu_spte_walk(vcpu, audit_spte);
237}
238
239static bool mmu_audit;
240static struct static_key mmu_audit_key;
241
242static void __kvm_mmu_audit(struct kvm_vcpu *vcpu, int point)
243{
244	static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
245
246	if (!__ratelimit(&ratelimit_state))
247		return;
248
249	vcpu->kvm->arch.audit_point = point;
250	audit_all_active_sps(vcpu->kvm);
251	audit_vcpu_spte(vcpu);
252}
253
254static inline void kvm_mmu_audit(struct kvm_vcpu *vcpu, int point)
255{
256	if (static_key_false((&mmu_audit_key)))
257		__kvm_mmu_audit(vcpu, point);
258}
259
260static void mmu_audit_enable(void)
261{
262	if (mmu_audit)
263		return;
264
265	static_key_slow_inc(&mmu_audit_key);
266	mmu_audit = true;
267}
268
269static void mmu_audit_disable(void)
270{
271	if (!mmu_audit)
272		return;
273
274	static_key_slow_dec(&mmu_audit_key);
275	mmu_audit = false;
276}
277
278static int mmu_audit_set(const char *val, const struct kernel_param *kp)
279{
280	int ret;
281	unsigned long enable;
282
283	ret = kstrtoul(val, 10, &enable);
284	if (ret < 0)
285		return -EINVAL;
286
287	switch (enable) {
288	case 0:
289		mmu_audit_disable();
290		break;
291	case 1:
292		mmu_audit_enable();
293		break;
294	default:
295		return -EINVAL;
296	}
297
298	return 0;
299}
300
301static const struct kernel_param_ops audit_param_ops = {
302	.set = mmu_audit_set,
303	.get = param_get_bool,
304};
305
306arch_param_cb(mmu_audit, &audit_param_ops, &mmu_audit, 0644);