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
 22char 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_LEVEL) {
 66		hpa_t root = vcpu->arch.mmu.root_hpa;
 67
 68		sp = page_header(root);
 69		__mmu_spte_walk(vcpu, sp, fn, PT64_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	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 = gfn_to_pfn_atomic(vcpu->kvm, gfn);
118
119	if (is_error_pfn(pfn)) {
120		kvm_release_pfn_clean(pfn);
121		return;
122	}
123
124	hpa =  pfn << PAGE_SHIFT;
125	if ((*sptep & PT64_BASE_ADDR_MASK) != hpa)
126		audit_printk(vcpu->kvm, "levels %d pfn %llx hpa %llx "
127			     "ent %llxn", vcpu->arch.mmu.root_level, pfn,
128			     hpa, *sptep);
129}
130
131static void inspect_spte_has_rmap(struct kvm *kvm, u64 *sptep)
132{
133	static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
134	unsigned long *rmapp;
135	struct kvm_mmu_page *rev_sp;
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	if (!gfn_to_memslot(kvm, gfn)) {
142		if (!__ratelimit(&ratelimit_state))
143			return;
144		audit_printk(kvm, "no memslot for gfn %llx\n", gfn);
145		audit_printk(kvm, "index %ld of sp (gfn=%llx)\n",
146		       (long int)(sptep - rev_sp->spt), rev_sp->gfn);
147		dump_stack();
148		return;
149	}
150
151	rmapp = gfn_to_rmap(kvm, gfn, rev_sp->role.level);
152	if (!*rmapp) {
153		if (!__ratelimit(&ratelimit_state))
154			return;
155		audit_printk(kvm, "no rmap for writable spte %llx\n",
156			     *sptep);
157		dump_stack();
158	}
159}
160
161static void audit_sptes_have_rmaps(struct kvm_vcpu *vcpu, u64 *sptep, int level)
162{
163	if (is_shadow_present_pte(*sptep) && is_last_spte(*sptep, level))
164		inspect_spte_has_rmap(vcpu->kvm, sptep);
165}
166
167static void audit_spte_after_sync(struct kvm_vcpu *vcpu, u64 *sptep, int level)
168{
169	struct kvm_mmu_page *sp = page_header(__pa(sptep));
170
171	if (vcpu->kvm->arch.audit_point == AUDIT_POST_SYNC && sp->unsync)
172		audit_printk(vcpu->kvm, "meet unsync sp(%p) after sync "
173			     "root.\n", sp);
174}
175
176static void check_mappings_rmap(struct kvm *kvm, struct kvm_mmu_page *sp)
177{
178	int i;
179
180	if (sp->role.level != PT_PAGE_TABLE_LEVEL)
181		return;
182
183	for (i = 0; i < PT64_ENT_PER_PAGE; ++i) {
184		if (!is_rmap_spte(sp->spt[i]))
185			continue;
186
187		inspect_spte_has_rmap(kvm, sp->spt + i);
188	}
189}
190
191static void audit_write_protection(struct kvm *kvm, struct kvm_mmu_page *sp)
192{
193	struct kvm_memory_slot *slot;
194	unsigned long *rmapp;
195	u64 *sptep;
196	struct rmap_iterator iter;
197
198	if (sp->role.direct || sp->unsync || sp->role.invalid)
199		return;
200
201	slot = gfn_to_memslot(kvm, sp->gfn);
202	rmapp = &slot->rmap[sp->gfn - slot->base_gfn];
203
204	for (sptep = rmap_get_first(*rmapp, &iter); sptep;
205	     sptep = rmap_get_next(&iter)) {
206		if (is_writable_pte(*sptep))
207			audit_printk(kvm, "shadow page has writable "
208				     "mappings: gfn %llx role %x\n",
209				     sp->gfn, sp->role.word);
210	}
211}
212
213static void audit_sp(struct kvm *kvm, struct kvm_mmu_page *sp)
214{
215	check_mappings_rmap(kvm, sp);
216	audit_write_protection(kvm, sp);
217}
218
219static void audit_all_active_sps(struct kvm *kvm)
220{
221	walk_all_active_sps(kvm, audit_sp);
222}
223
224static void audit_spte(struct kvm_vcpu *vcpu, u64 *sptep, int level)
225{
226	audit_sptes_have_rmaps(vcpu, sptep, level);
227	audit_mappings(vcpu, sptep, level);
228	audit_spte_after_sync(vcpu, sptep, level);
229}
230
231static void audit_vcpu_spte(struct kvm_vcpu *vcpu)
232{
233	mmu_spte_walk(vcpu, audit_spte);
234}
235
236static bool mmu_audit;
237static struct static_key mmu_audit_key;
238
239static void __kvm_mmu_audit(struct kvm_vcpu *vcpu, int point)
240{
241	static DEFINE_RATELIMIT_STATE(ratelimit_state, 5 * HZ, 10);
242
243	if (!__ratelimit(&ratelimit_state))
244		return;
245
246	vcpu->kvm->arch.audit_point = point;
247	audit_all_active_sps(vcpu->kvm);
248	audit_vcpu_spte(vcpu);
249}
250
251static inline void kvm_mmu_audit(struct kvm_vcpu *vcpu, int point)
252{
253	if (static_key_false((&mmu_audit_key)))
254		__kvm_mmu_audit(vcpu, point);
255}
256
257static void mmu_audit_enable(void)
258{
259	if (mmu_audit)
260		return;
261
262	static_key_slow_inc(&mmu_audit_key);
263	mmu_audit = true;
264}
265
266static void mmu_audit_disable(void)
267{
268	if (!mmu_audit)
269		return;
270
271	static_key_slow_dec(&mmu_audit_key);
272	mmu_audit = false;
273}
274
275static int mmu_audit_set(const char *val, const struct kernel_param *kp)
276{
277	int ret;
278	unsigned long enable;
279
280	ret = strict_strtoul(val, 10, &enable);
281	if (ret < 0)
282		return -EINVAL;
283
284	switch (enable) {
285	case 0:
286		mmu_audit_disable();
287		break;
288	case 1:
289		mmu_audit_enable();
290		break;
291	default:
292		return -EINVAL;
293	}
294
295	return 0;
296}
297
298static struct kernel_param_ops audit_param_ops = {
299	.set = mmu_audit_set,
300	.get = param_get_bool,
301};
302
303module_param_cb(mmu_audit, &audit_param_ops, &mmu_audit, 0644);