diff options
Diffstat (limited to 'arch/x86/kvm/mmu_audit.c')
-rw-r--r-- | arch/x86/kvm/mmu_audit.c | 304 |
1 files changed, 304 insertions, 0 deletions
diff --git a/arch/x86/kvm/mmu_audit.c b/arch/x86/kvm/mmu_audit.c new file mode 100644 index 000000000000..5f6223b8bcf7 --- /dev/null +++ b/arch/x86/kvm/mmu_audit.c | |||
@@ -0,0 +1,304 @@ | |||
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 | |||
26 | typedef void (*inspect_spte_fn) (struct kvm_vcpu *vcpu, u64 *sptep, int level); | ||
27 | |||
28 | static 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 | |||
48 | static 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 | |||
77 | typedef void (*sp_handler) (struct kvm *kvm, struct kvm_mmu_page *sp); | ||
78 | |||
79 | static 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 | |||
87 | static 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 | if (*sptep == shadow_notrap_nonpresent_pte) { | ||
104 | audit_printk(vcpu->kvm, "notrap spte in unsync " | ||
105 | "sp: %p\n", sp); | ||
106 | return; | ||
107 | } | ||
108 | } | ||
109 | |||
110 | if (sp->role.direct && *sptep == shadow_notrap_nonpresent_pte) { | ||
111 | audit_printk(vcpu->kvm, "notrap spte in direct sp: %p\n", | ||
112 | sp); | ||
113 | return; | ||
114 | } | ||
115 | |||
116 | if (!is_shadow_present_pte(*sptep) || !is_last_spte(*sptep, level)) | ||
117 | return; | ||
118 | |||
119 | gfn = kvm_mmu_page_get_gfn(sp, sptep - sp->spt); | ||
120 | pfn = gfn_to_pfn_atomic(vcpu->kvm, gfn); | ||
121 | |||
122 | if (is_error_pfn(pfn)) { | ||
123 | kvm_release_pfn_clean(pfn); | ||
124 | return; | ||
125 | } | ||
126 | |||
127 | hpa = pfn << PAGE_SHIFT; | ||
128 | if ((*sptep & PT64_BASE_ADDR_MASK) != hpa) | ||
129 | audit_printk(vcpu->kvm, "levels %d pfn %llx hpa %llx " | ||
130 | "ent %llxn", vcpu->arch.mmu.root_level, pfn, | ||
131 | hpa, *sptep); | ||
132 | } | ||
133 | |||
134 | static void inspect_spte_has_rmap(struct kvm *kvm, u64 *sptep) | ||
135 | { | ||
136 | unsigned long *rmapp; | ||
137 | struct kvm_mmu_page *rev_sp; | ||
138 | gfn_t gfn; | ||
139 | |||
140 | |||
141 | rev_sp = page_header(__pa(sptep)); | ||
142 | gfn = kvm_mmu_page_get_gfn(rev_sp, sptep - rev_sp->spt); | ||
143 | |||
144 | if (!gfn_to_memslot(kvm, gfn)) { | ||
145 | if (!printk_ratelimit()) | ||
146 | return; | ||
147 | audit_printk(kvm, "no memslot for gfn %llx\n", gfn); | ||
148 | audit_printk(kvm, "index %ld of sp (gfn=%llx)\n", | ||
149 | (long int)(sptep - rev_sp->spt), rev_sp->gfn); | ||
150 | dump_stack(); | ||
151 | return; | ||
152 | } | ||
153 | |||
154 | rmapp = gfn_to_rmap(kvm, gfn, rev_sp->role.level); | ||
155 | if (!*rmapp) { | ||
156 | if (!printk_ratelimit()) | ||
157 | return; | ||
158 | audit_printk(kvm, "no rmap for writable spte %llx\n", | ||
159 | *sptep); | ||
160 | dump_stack(); | ||
161 | } | ||
162 | } | ||
163 | |||
164 | static void audit_sptes_have_rmaps(struct kvm_vcpu *vcpu, u64 *sptep, int level) | ||
165 | { | ||
166 | if (is_shadow_present_pte(*sptep) && is_last_spte(*sptep, level)) | ||
167 | inspect_spte_has_rmap(vcpu->kvm, sptep); | ||
168 | } | ||
169 | |||
170 | static void audit_spte_after_sync(struct kvm_vcpu *vcpu, u64 *sptep, int level) | ||
171 | { | ||
172 | struct kvm_mmu_page *sp = page_header(__pa(sptep)); | ||
173 | |||
174 | if (vcpu->kvm->arch.audit_point == AUDIT_POST_SYNC && sp->unsync) | ||
175 | audit_printk(vcpu->kvm, "meet unsync sp(%p) after sync " | ||
176 | "root.\n", sp); | ||
177 | } | ||
178 | |||
179 | static void check_mappings_rmap(struct kvm *kvm, struct kvm_mmu_page *sp) | ||
180 | { | ||
181 | int i; | ||
182 | |||
183 | if (sp->role.level != PT_PAGE_TABLE_LEVEL) | ||
184 | return; | ||
185 | |||
186 | for (i = 0; i < PT64_ENT_PER_PAGE; ++i) { | ||
187 | if (!is_rmap_spte(sp->spt[i])) | ||
188 | continue; | ||
189 | |||
190 | inspect_spte_has_rmap(kvm, sp->spt + i); | ||
191 | } | ||
192 | } | ||
193 | |||
194 | static void audit_write_protection(struct kvm *kvm, struct kvm_mmu_page *sp) | ||
195 | { | ||
196 | struct kvm_memory_slot *slot; | ||
197 | unsigned long *rmapp; | ||
198 | u64 *spte; | ||
199 | |||
200 | if (sp->role.direct || sp->unsync || sp->role.invalid) | ||
201 | return; | ||
202 | |||
203 | slot = gfn_to_memslot(kvm, sp->gfn); | ||
204 | rmapp = &slot->rmap[sp->gfn - slot->base_gfn]; | ||
205 | |||
206 | spte = rmap_next(kvm, rmapp, NULL); | ||
207 | while (spte) { | ||
208 | if (is_writable_pte(*spte)) | ||
209 | audit_printk(kvm, "shadow page has writable " | ||
210 | "mappings: gfn %llx role %x\n", | ||
211 | sp->gfn, sp->role.word); | ||
212 | spte = rmap_next(kvm, rmapp, spte); | ||
213 | } | ||
214 | } | ||
215 | |||
216 | static 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 | |||
222 | static void audit_all_active_sps(struct kvm *kvm) | ||
223 | { | ||
224 | walk_all_active_sps(kvm, audit_sp); | ||
225 | } | ||
226 | |||
227 | static 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 | |||
234 | static void audit_vcpu_spte(struct kvm_vcpu *vcpu) | ||
235 | { | ||
236 | mmu_spte_walk(vcpu, audit_spte); | ||
237 | } | ||
238 | |||
239 | static void kvm_mmu_audit(void *ignore, 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 | |||
251 | static bool mmu_audit; | ||
252 | |||
253 | static void mmu_audit_enable(void) | ||
254 | { | ||
255 | int ret; | ||
256 | |||
257 | if (mmu_audit) | ||
258 | return; | ||
259 | |||
260 | ret = register_trace_kvm_mmu_audit(kvm_mmu_audit, NULL); | ||
261 | WARN_ON(ret); | ||
262 | |||
263 | mmu_audit = true; | ||
264 | } | ||
265 | |||
266 | static void mmu_audit_disable(void) | ||
267 | { | ||
268 | if (!mmu_audit) | ||
269 | return; | ||
270 | |||
271 | unregister_trace_kvm_mmu_audit(kvm_mmu_audit, NULL); | ||
272 | tracepoint_synchronize_unregister(); | ||
273 | mmu_audit = false; | ||
274 | } | ||
275 | |||
276 | static int mmu_audit_set(const char *val, const struct kernel_param *kp) | ||
277 | { | ||
278 | int ret; | ||
279 | unsigned long enable; | ||
280 | |||
281 | ret = strict_strtoul(val, 10, &enable); | ||
282 | if (ret < 0) | ||
283 | return -EINVAL; | ||
284 | |||
285 | switch (enable) { | ||
286 | case 0: | ||
287 | mmu_audit_disable(); | ||
288 | break; | ||
289 | case 1: | ||
290 | mmu_audit_enable(); | ||
291 | break; | ||
292 | default: | ||
293 | return -EINVAL; | ||
294 | } | ||
295 | |||
296 | return 0; | ||
297 | } | ||
298 | |||
299 | static struct kernel_param_ops audit_param_ops = { | ||
300 | .set = mmu_audit_set, | ||
301 | .get = param_get_bool, | ||
302 | }; | ||
303 | |||
304 | module_param_cb(mmu_audit, &audit_param_ops, &mmu_audit, 0644); | ||