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/swapops.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,pte_t pte)148 void __page_table_check_pte_clear(struct mm_struct *mm, pte_t pte)
149 {
150 	if (&init_mm == mm)
151 		return;
152 
153 	if (pte_user_accessible_page(pte)) {
154 		page_table_check_clear(pte_pfn(pte), PAGE_SIZE >> PAGE_SHIFT);
155 	}
156 }
157 EXPORT_SYMBOL(__page_table_check_pte_clear);
158 
__page_table_check_pmd_clear(struct mm_struct * mm,pmd_t pmd)159 void __page_table_check_pmd_clear(struct mm_struct *mm, pmd_t pmd)
160 {
161 	if (&init_mm == mm)
162 		return;
163 
164 	if (pmd_user_accessible_page(pmd)) {
165 		page_table_check_clear(pmd_pfn(pmd), PMD_SIZE >> PAGE_SHIFT);
166 	}
167 }
168 EXPORT_SYMBOL(__page_table_check_pmd_clear);
169 
__page_table_check_pud_clear(struct mm_struct * mm,pud_t pud)170 void __page_table_check_pud_clear(struct mm_struct *mm, pud_t pud)
171 {
172 	if (&init_mm == mm)
173 		return;
174 
175 	if (pud_user_accessible_page(pud)) {
176 		page_table_check_clear(pud_pfn(pud), PUD_SIZE >> PAGE_SHIFT);
177 	}
178 }
179 EXPORT_SYMBOL(__page_table_check_pud_clear);
180 
181 /* Whether the swap entry cached writable information */
swap_cached_writable(swp_entry_t entry)182 static inline bool swap_cached_writable(swp_entry_t entry)
183 {
184 	return is_writable_device_private_entry(entry) ||
185 	       is_writable_migration_entry(entry);
186 }
187 
page_table_check_pte_flags(pte_t pte)188 static inline void page_table_check_pte_flags(pte_t pte)
189 {
190 	if (pte_present(pte) && pte_uffd_wp(pte))
191 		WARN_ON_ONCE(pte_write(pte));
192 	else if (is_swap_pte(pte) && pte_swp_uffd_wp(pte))
193 		WARN_ON_ONCE(swap_cached_writable(pte_to_swp_entry(pte)));
194 }
195 
__page_table_check_ptes_set(struct mm_struct * mm,pte_t * ptep,pte_t pte,unsigned int nr)196 void __page_table_check_ptes_set(struct mm_struct *mm, pte_t *ptep, pte_t pte,
197 		unsigned int nr)
198 {
199 	unsigned int i;
200 
201 	if (&init_mm == mm)
202 		return;
203 
204 	page_table_check_pte_flags(pte);
205 
206 	for (i = 0; i < nr; i++)
207 		__page_table_check_pte_clear(mm, ptep_get(ptep + i));
208 	if (pte_user_accessible_page(pte))
209 		page_table_check_set(pte_pfn(pte), nr, pte_write(pte));
210 }
211 EXPORT_SYMBOL(__page_table_check_ptes_set);
212 
page_table_check_pmd_flags(pmd_t pmd)213 static inline void page_table_check_pmd_flags(pmd_t pmd)
214 {
215 	if (pmd_present(pmd) && pmd_uffd_wp(pmd))
216 		WARN_ON_ONCE(pmd_write(pmd));
217 	else if (is_swap_pmd(pmd) && pmd_swp_uffd_wp(pmd))
218 		WARN_ON_ONCE(swap_cached_writable(pmd_to_swp_entry(pmd)));
219 }
220 
__page_table_check_pmd_set(struct mm_struct * mm,pmd_t * pmdp,pmd_t pmd)221 void __page_table_check_pmd_set(struct mm_struct *mm, pmd_t *pmdp, pmd_t pmd)
222 {
223 	if (&init_mm == mm)
224 		return;
225 
226 	page_table_check_pmd_flags(pmd);
227 
228 	__page_table_check_pmd_clear(mm, *pmdp);
229 	if (pmd_user_accessible_page(pmd)) {
230 		page_table_check_set(pmd_pfn(pmd), PMD_SIZE >> PAGE_SHIFT,
231 				     pmd_write(pmd));
232 	}
233 }
234 EXPORT_SYMBOL(__page_table_check_pmd_set);
235 
__page_table_check_pud_set(struct mm_struct * mm,pud_t * pudp,pud_t pud)236 void __page_table_check_pud_set(struct mm_struct *mm, pud_t *pudp, pud_t pud)
237 {
238 	if (&init_mm == mm)
239 		return;
240 
241 	__page_table_check_pud_clear(mm, *pudp);
242 	if (pud_user_accessible_page(pud)) {
243 		page_table_check_set(pud_pfn(pud), PUD_SIZE >> PAGE_SHIFT,
244 				     pud_write(pud));
245 	}
246 }
247 EXPORT_SYMBOL(__page_table_check_pud_set);
248 
__page_table_check_pte_clear_range(struct mm_struct * mm,unsigned long addr,pmd_t pmd)249 void __page_table_check_pte_clear_range(struct mm_struct *mm,
250 					unsigned long addr,
251 					pmd_t pmd)
252 {
253 	if (&init_mm == mm)
254 		return;
255 
256 	if (!pmd_bad(pmd) && !pmd_leaf(pmd)) {
257 		pte_t *ptep = pte_offset_map(&pmd, addr);
258 		unsigned long i;
259 
260 		if (WARN_ON(!ptep))
261 			return;
262 		for (i = 0; i < PTRS_PER_PTE; i++) {
263 			__page_table_check_pte_clear(mm, ptep_get(ptep));
264 			addr += PAGE_SIZE;
265 			ptep++;
266 		}
267 		pte_unmap(ptep - PTRS_PER_PTE);
268 	}
269 }
270