xref: /linux/mm/page_table_check.c (revision f4cdf7ca9a1fdcca413157df19753f388a5a224e)
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 
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 
34 static bool __init need_page_table_check(void)
35 {
36 	return __page_table_check_enabled;
37 }
38 
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 
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  */
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  */
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  */
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 
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(mm, addr, pte) && !pte_special(pte))
155 		page_table_check_clear(pte_pfn(pte), PAGE_SIZE >> PAGE_SHIFT);
156 }
157 EXPORT_SYMBOL(__page_table_check_pte_clear);
158 
159 static inline bool page_table_check_huge_zero_pmd(pmd_t pmd)
160 {
161 	unsigned long pfn = pmd_pfn(pmd);
162 
163 	if (!pfn_valid(pfn))
164 		return false;
165 
166 	return is_huge_zero_folio(page_folio(pfn_to_page(pfn)));
167 }
168 
169 void __page_table_check_pmd_clear(struct mm_struct *mm, unsigned long addr,
170 				  pmd_t pmd)
171 {
172 	if (&init_mm == mm)
173 		return;
174 
175 	if (pmd_user_accessible_page(mm, addr, pmd) &&
176 	    !page_table_check_huge_zero_pmd(pmd))
177 		page_table_check_clear(pmd_pfn(pmd), PMD_SIZE >> PAGE_SHIFT);
178 }
179 EXPORT_SYMBOL(__page_table_check_pmd_clear);
180 
181 void __page_table_check_pud_clear(struct mm_struct *mm, unsigned long addr,
182 				  pud_t pud)
183 {
184 	if (&init_mm == mm)
185 		return;
186 
187 	if (pud_user_accessible_page(mm, addr, pud))
188 		page_table_check_clear(pud_pfn(pud), PUD_SIZE >> PAGE_SHIFT);
189 }
190 EXPORT_SYMBOL(__page_table_check_pud_clear);
191 
192 /* Whether the swap entry cached writable information */
193 static inline bool softleaf_cached_writable(softleaf_t entry)
194 {
195 	return softleaf_is_device_private_write(entry) ||
196 		softleaf_is_migration_write(entry);
197 }
198 
199 static void page_table_check_pte_flags(pte_t pte)
200 {
201 	if (pte_present(pte)) {
202 		WARN_ON_ONCE(pte_uffd_wp(pte) && pte_write(pte));
203 	} else if (pte_swp_uffd_wp(pte)) {
204 		const softleaf_t entry = softleaf_from_pte(pte);
205 
206 		WARN_ON_ONCE(softleaf_cached_writable(entry));
207 	}
208 }
209 
210 void __page_table_check_ptes_set(struct mm_struct *mm, unsigned long addr,
211 				 pte_t *ptep, pte_t pte, unsigned int nr)
212 {
213 	unsigned int i;
214 
215 	if (&init_mm == mm)
216 		return;
217 
218 	page_table_check_pte_flags(pte);
219 
220 	for (i = 0; i < nr; i++)
221 		__page_table_check_pte_clear(mm, addr + PAGE_SIZE * i, ptep_get(ptep + i));
222 	if (pte_user_accessible_page(mm, addr, pte) && !pte_special(pte))
223 		page_table_check_set(pte_pfn(pte), nr, pte_write(pte));
224 }
225 EXPORT_SYMBOL(__page_table_check_ptes_set);
226 
227 static inline void page_table_check_pmd_flags(pmd_t pmd)
228 {
229 	if (pmd_present(pmd)) {
230 		if (pmd_uffd_wp(pmd))
231 			WARN_ON_ONCE(pmd_write(pmd));
232 	} else if (pmd_swp_uffd_wp(pmd)) {
233 		const softleaf_t entry = softleaf_from_pmd(pmd);
234 
235 		WARN_ON_ONCE(softleaf_cached_writable(entry));
236 	}
237 }
238 
239 void __page_table_check_pmds_set(struct mm_struct *mm, unsigned long addr,
240 		pmd_t *pmdp, pmd_t pmd, unsigned int nr)
241 {
242 	unsigned long stride = PMD_SIZE >> PAGE_SHIFT;
243 	unsigned int i;
244 
245 	if (&init_mm == mm)
246 		return;
247 
248 	page_table_check_pmd_flags(pmd);
249 
250 	for (i = 0; i < nr; i++)
251 		__page_table_check_pmd_clear(mm, addr + PMD_SIZE * i, *(pmdp + i));
252 	if (pmd_user_accessible_page(mm, addr, pmd) &&
253 	    !page_table_check_huge_zero_pmd(pmd))
254 		page_table_check_set(pmd_pfn(pmd), stride * nr, pmd_write(pmd));
255 }
256 EXPORT_SYMBOL(__page_table_check_pmds_set);
257 
258 void __page_table_check_puds_set(struct mm_struct *mm, unsigned long addr,
259 		pud_t *pudp, pud_t pud,	unsigned int nr)
260 {
261 	unsigned long stride = PUD_SIZE >> PAGE_SHIFT;
262 	unsigned int i;
263 
264 	if (&init_mm == mm)
265 		return;
266 
267 	for (i = 0; i < nr; i++)
268 		__page_table_check_pud_clear(mm, addr + PUD_SIZE * i, *(pudp + i));
269 	if (pud_user_accessible_page(mm, addr, pud))
270 		page_table_check_set(pud_pfn(pud), stride * nr, pud_write(pud));
271 }
272 EXPORT_SYMBOL(__page_table_check_puds_set);
273 
274 void __page_table_check_pte_clear_range(struct mm_struct *mm,
275 					unsigned long addr,
276 					pmd_t pmd)
277 {
278 	if (&init_mm == mm)
279 		return;
280 
281 	if (!pmd_bad(pmd) && !pmd_leaf(pmd)) {
282 		pte_t *ptep = pte_offset_map(&pmd, addr);
283 		unsigned long i;
284 
285 		if (WARN_ON(!ptep))
286 			return;
287 		for (i = 0; i < PTRS_PER_PTE; i++) {
288 			__page_table_check_pte_clear(mm, addr, ptep_get(ptep));
289 			addr += PAGE_SIZE;
290 			ptep++;
291 		}
292 		pte_unmap(ptep - PTRS_PER_PTE);
293 	}
294 }
295