1 // SPDX-License-Identifier: GPL-2.0
2
3 /*
4 * Copyright (c) 2021, Google LLC.
5 * Pasha Tatashin <pasha.tatashin@soleen.com>
6 */
7 #include <linux/kstrtox.h>
8 #include <linux/mm.h>
9 #include <linux/page_table_check.h>
10 #include <linux/swap.h>
11 #include <linux/leafops.h>
12
13 #undef pr_fmt
14 #define pr_fmt(fmt) "page_table_check: " fmt
15
16 struct page_table_check {
17 atomic_t anon_map_count;
18 atomic_t file_map_count;
19 };
20
21 static bool __page_table_check_enabled __initdata =
22 IS_ENABLED(CONFIG_PAGE_TABLE_CHECK_ENFORCED);
23
24 DEFINE_STATIC_KEY_TRUE(page_table_check_disabled);
25 EXPORT_SYMBOL(page_table_check_disabled);
26
early_page_table_check_param(char * buf)27 static int __init early_page_table_check_param(char *buf)
28 {
29 return kstrtobool(buf, &__page_table_check_enabled);
30 }
31
32 early_param("page_table_check", early_page_table_check_param);
33
need_page_table_check(void)34 static bool __init need_page_table_check(void)
35 {
36 return __page_table_check_enabled;
37 }
38
init_page_table_check(void)39 static void __init init_page_table_check(void)
40 {
41 if (!__page_table_check_enabled)
42 return;
43 static_branch_disable(&page_table_check_disabled);
44 }
45
46 struct page_ext_operations page_table_check_ops = {
47 .size = sizeof(struct page_table_check),
48 .need = need_page_table_check,
49 .init = init_page_table_check,
50 .need_shared_flags = false,
51 };
52
get_page_table_check(struct page_ext * page_ext)53 static struct page_table_check *get_page_table_check(struct page_ext *page_ext)
54 {
55 BUG_ON(!page_ext);
56 return page_ext_data(page_ext, &page_table_check_ops);
57 }
58
59 /*
60 * An entry is removed from the page table, decrement the counters for that page
61 * verify that it is of correct type and counters do not become negative.
62 */
page_table_check_clear(unsigned long pfn,unsigned long pgcnt)63 static void page_table_check_clear(unsigned long pfn, unsigned long pgcnt)
64 {
65 struct page_ext_iter iter;
66 struct page_ext *page_ext;
67 struct page *page;
68 bool anon;
69
70 if (!pfn_valid(pfn))
71 return;
72
73 page = pfn_to_page(pfn);
74 BUG_ON(PageSlab(page));
75 anon = PageAnon(page);
76
77 rcu_read_lock();
78 for_each_page_ext(page, pgcnt, page_ext, iter) {
79 struct page_table_check *ptc = get_page_table_check(page_ext);
80
81 if (anon) {
82 BUG_ON(atomic_read(&ptc->file_map_count));
83 BUG_ON(atomic_dec_return(&ptc->anon_map_count) < 0);
84 } else {
85 BUG_ON(atomic_read(&ptc->anon_map_count));
86 BUG_ON(atomic_dec_return(&ptc->file_map_count) < 0);
87 }
88 }
89 rcu_read_unlock();
90 }
91
92 /*
93 * A new entry is added to the page table, increment the counters for that page
94 * verify that it is of correct type and is not being mapped with a different
95 * type to a different process.
96 */
page_table_check_set(unsigned long pfn,unsigned long pgcnt,bool rw)97 static void page_table_check_set(unsigned long pfn, unsigned long pgcnt,
98 bool rw)
99 {
100 struct page_ext_iter iter;
101 struct page_ext *page_ext;
102 struct page *page;
103 bool anon;
104
105 if (!pfn_valid(pfn))
106 return;
107
108 page = pfn_to_page(pfn);
109 BUG_ON(PageSlab(page));
110 anon = PageAnon(page);
111
112 rcu_read_lock();
113 for_each_page_ext(page, pgcnt, page_ext, iter) {
114 struct page_table_check *ptc = get_page_table_check(page_ext);
115
116 if (anon) {
117 BUG_ON(atomic_read(&ptc->file_map_count));
118 BUG_ON(atomic_inc_return(&ptc->anon_map_count) > 1 && rw);
119 } else {
120 BUG_ON(atomic_read(&ptc->anon_map_count));
121 BUG_ON(atomic_inc_return(&ptc->file_map_count) < 0);
122 }
123 }
124 rcu_read_unlock();
125 }
126
127 /*
128 * page is on free list, or is being allocated, verify that counters are zeroes
129 * crash if they are not.
130 */
__page_table_check_zero(struct page * page,unsigned int order)131 void __page_table_check_zero(struct page *page, unsigned int order)
132 {
133 struct page_ext_iter iter;
134 struct page_ext *page_ext;
135
136 BUG_ON(PageSlab(page));
137
138 rcu_read_lock();
139 for_each_page_ext(page, 1 << order, page_ext, iter) {
140 struct page_table_check *ptc = get_page_table_check(page_ext);
141
142 BUG_ON(atomic_read(&ptc->anon_map_count));
143 BUG_ON(atomic_read(&ptc->file_map_count));
144 }
145 rcu_read_unlock();
146 }
147
__page_table_check_pte_clear(struct mm_struct * mm,unsigned long addr,pte_t pte)148 void __page_table_check_pte_clear(struct mm_struct *mm, unsigned long addr,
149 pte_t pte)
150 {
151 if (&init_mm == mm)
152 return;
153
154 if (pte_user_accessible_page(pte, addr)) {
155 page_table_check_clear(pte_pfn(pte), PAGE_SIZE >> PAGE_SHIFT);
156 }
157 }
158 EXPORT_SYMBOL(__page_table_check_pte_clear);
159
__page_table_check_pmd_clear(struct mm_struct * mm,unsigned long addr,pmd_t pmd)160 void __page_table_check_pmd_clear(struct mm_struct *mm, unsigned long addr,
161 pmd_t pmd)
162 {
163 if (&init_mm == mm)
164 return;
165
166 if (pmd_user_accessible_page(pmd, addr)) {
167 page_table_check_clear(pmd_pfn(pmd), PMD_SIZE >> PAGE_SHIFT);
168 }
169 }
170 EXPORT_SYMBOL(__page_table_check_pmd_clear);
171
__page_table_check_pud_clear(struct mm_struct * mm,unsigned long addr,pud_t pud)172 void __page_table_check_pud_clear(struct mm_struct *mm, unsigned long addr,
173 pud_t pud)
174 {
175 if (&init_mm == mm)
176 return;
177
178 if (pud_user_accessible_page(pud, addr)) {
179 page_table_check_clear(pud_pfn(pud), PUD_SIZE >> PAGE_SHIFT);
180 }
181 }
182 EXPORT_SYMBOL(__page_table_check_pud_clear);
183
184 /* Whether the swap entry cached writable information */
softleaf_cached_writable(softleaf_t entry)185 static inline bool softleaf_cached_writable(softleaf_t entry)
186 {
187 return softleaf_is_device_private_write(entry) ||
188 softleaf_is_migration_write(entry);
189 }
190
page_table_check_pte_flags(pte_t pte)191 static void page_table_check_pte_flags(pte_t pte)
192 {
193 if (pte_present(pte)) {
194 WARN_ON_ONCE(pte_uffd_wp(pte) && pte_write(pte));
195 } else if (pte_swp_uffd_wp(pte)) {
196 const softleaf_t entry = softleaf_from_pte(pte);
197
198 WARN_ON_ONCE(softleaf_cached_writable(entry));
199 }
200 }
201
__page_table_check_ptes_set(struct mm_struct * mm,unsigned long addr,pte_t * ptep,pte_t pte,unsigned int nr)202 void __page_table_check_ptes_set(struct mm_struct *mm, unsigned long addr,
203 pte_t *ptep, pte_t pte, unsigned int nr)
204 {
205 unsigned int i;
206
207 if (&init_mm == mm)
208 return;
209
210 page_table_check_pte_flags(pte);
211
212 for (i = 0; i < nr; i++)
213 __page_table_check_pte_clear(mm, addr + PAGE_SIZE * i, ptep_get(ptep + i));
214 if (pte_user_accessible_page(pte, addr))
215 page_table_check_set(pte_pfn(pte), nr, pte_write(pte));
216 }
217 EXPORT_SYMBOL(__page_table_check_ptes_set);
218
page_table_check_pmd_flags(pmd_t pmd)219 static inline void page_table_check_pmd_flags(pmd_t pmd)
220 {
221 if (pmd_present(pmd)) {
222 if (pmd_uffd_wp(pmd))
223 WARN_ON_ONCE(pmd_write(pmd));
224 } else if (pmd_swp_uffd_wp(pmd)) {
225 const softleaf_t entry = softleaf_from_pmd(pmd);
226
227 WARN_ON_ONCE(softleaf_cached_writable(entry));
228 }
229 }
230
__page_table_check_pmds_set(struct mm_struct * mm,unsigned long addr,pmd_t * pmdp,pmd_t pmd,unsigned int nr)231 void __page_table_check_pmds_set(struct mm_struct *mm, unsigned long addr,
232 pmd_t *pmdp, pmd_t pmd, unsigned int nr)
233 {
234 unsigned long stride = PMD_SIZE >> PAGE_SHIFT;
235 unsigned int i;
236
237 if (&init_mm == mm)
238 return;
239
240 page_table_check_pmd_flags(pmd);
241
242 for (i = 0; i < nr; i++)
243 __page_table_check_pmd_clear(mm, addr + PMD_SIZE * i, *(pmdp + i));
244 if (pmd_user_accessible_page(pmd, addr))
245 page_table_check_set(pmd_pfn(pmd), stride * nr, pmd_write(pmd));
246 }
247 EXPORT_SYMBOL(__page_table_check_pmds_set);
248
__page_table_check_puds_set(struct mm_struct * mm,unsigned long addr,pud_t * pudp,pud_t pud,unsigned int nr)249 void __page_table_check_puds_set(struct mm_struct *mm, unsigned long addr,
250 pud_t *pudp, pud_t pud, unsigned int nr)
251 {
252 unsigned long stride = PUD_SIZE >> PAGE_SHIFT;
253 unsigned int i;
254
255 if (&init_mm == mm)
256 return;
257
258 for (i = 0; i < nr; i++)
259 __page_table_check_pud_clear(mm, addr + PUD_SIZE * i, *(pudp + i));
260 if (pud_user_accessible_page(pud, addr))
261 page_table_check_set(pud_pfn(pud), stride * nr, pud_write(pud));
262 }
263 EXPORT_SYMBOL(__page_table_check_puds_set);
264
__page_table_check_pte_clear_range(struct mm_struct * mm,unsigned long addr,pmd_t pmd)265 void __page_table_check_pte_clear_range(struct mm_struct *mm,
266 unsigned long addr,
267 pmd_t pmd)
268 {
269 if (&init_mm == mm)
270 return;
271
272 if (!pmd_bad(pmd) && !pmd_leaf(pmd)) {
273 pte_t *ptep = pte_offset_map(&pmd, addr);
274 unsigned long i;
275
276 if (WARN_ON(!ptep))
277 return;
278 for (i = 0; i < PTRS_PER_PTE; i++) {
279 __page_table_check_pte_clear(mm, addr, ptep_get(ptep));
280 addr += PAGE_SIZE;
281 ptep++;
282 }
283 pte_unmap(ptep - PTRS_PER_PTE);
284 }
285 }
286