1 // SPDX-License-Identifier: GPL-2.0 1
2 #include <linux/pagewalk.h>
3 #include <linux/hugetlb.h>
4 #include <linux/bitops.h>
5 #include <linux/mmu_notifier.h>
6 #include <linux/mm_inline.h>
7 #include <asm/cacheflush.h>
8 #include <asm/tlbflush.h>
9
10 /**
11 * struct wp_walk - Private struct for pagetab
12 * @range: Range for mmu notifiers
13 * @tlbflush_start: Address of first modified
14 * @tlbflush_end: Address of last modified pte
15 * @total: Total number of modified ptes
16 */
17 struct wp_walk {
18 struct mmu_notifier_range range;
19 unsigned long tlbflush_start;
20 unsigned long tlbflush_end;
21 unsigned long total;
22 };
23
24 /**
25 * wp_pte - Write-protect a pte
26 * @pte: Pointer to the pte
27 * @addr: The start of protecting virtual addr
28 * @end: The end of protecting virtual address
29 * @walk: pagetable walk callback argument
30 *
31 * The function write-protects a pte and recor
32 * virtual address space of touched ptes for e
33 */
34 static int wp_pte(pte_t *pte, unsigned long ad
35 struct mm_walk *walk)
36 {
37 struct wp_walk *wpwalk = walk->private
38 pte_t ptent = ptep_get(pte);
39
40 if (pte_write(ptent)) {
41 pte_t old_pte = ptep_modify_pr
42
43 ptent = pte_wrprotect(old_pte)
44 ptep_modify_prot_commit(walk->
45 wpwalk->total++;
46 wpwalk->tlbflush_start = min(w
47 wpwalk->tlbflush_end = max(wpw
48 add
49 }
50
51 return 0;
52 }
53
54 /**
55 * struct clean_walk - Private struct for the
56 * @base: struct wp_walk we derive from
57 * @bitmap_pgoff: Address_space Page offset of
58 * @bitmap: Bitmap with one bit for each page
59 * covered.
60 * @start: Address_space page offset of first
61 * to @bitmap_pgoff
62 * @end: Address_space page offset of last mod
63 * to @bitmap_pgoff
64 */
65 struct clean_walk {
66 struct wp_walk base;
67 pgoff_t bitmap_pgoff;
68 unsigned long *bitmap;
69 pgoff_t start;
70 pgoff_t end;
71 };
72
73 #define to_clean_walk(_wpwalk) container_of(_w
74
75 /**
76 * clean_record_pte - Clean a pte and record i
77 * bitmap
78 * @pte: Pointer to the pte
79 * @addr: The start of virtual address to be c
80 * @end: The end of virtual address to be clea
81 * @walk: pagetable walk callback argument
82 *
83 * The function cleans a pte and records the r
84 * virtual address space of touched ptes for e
85 * It also records dirty ptes in a bitmap repr
86 * in the address_space, as well as the first
87 * touched.
88 */
89 static int clean_record_pte(pte_t *pte, unsign
90 unsigned long end,
91 {
92 struct wp_walk *wpwalk = walk->private
93 struct clean_walk *cwalk = to_clean_wa
94 pte_t ptent = ptep_get(pte);
95
96 if (pte_dirty(ptent)) {
97 pgoff_t pgoff = ((addr - walk-
98 walk->vma->vm_pgoff -
99 pte_t old_pte = ptep_modify_pr
100
101 ptent = pte_mkclean(old_pte);
102 ptep_modify_prot_commit(walk->
103
104 wpwalk->total++;
105 wpwalk->tlbflush_start = min(w
106 wpwalk->tlbflush_end = max(wpw
107 add
108
109 __set_bit(pgoff, cwalk->bitmap
110 cwalk->start = min(cwalk->star
111 cwalk->end = max(cwalk->end, p
112 }
113
114 return 0;
115 }
116
117 /*
118 * wp_clean_pmd_entry - The pagewalk pmd callb
119 *
120 * Dirty-tracking should take place on the PTE
121 * WARN() if encountering a dirty huge pmd.
122 * Furthermore, never split huge pmds, since t
123 * causes dirty info loss. The pagefault handl
124 * that if needed.
125 */
126 static int wp_clean_pmd_entry(pmd_t *pmd, unsi
127 struct mm_walk *
128 {
129 pmd_t pmdval = pmdp_get_lockless(pmd);
130
131 /* Do not split a huge pmd, present or
132 if (pmd_trans_huge(pmdval) || pmd_devm
133 WARN_ON(pmd_write(pmdval) || p
134 walk->action = ACTION_CONTINUE
135 }
136 return 0;
137 }
138
139 /*
140 * wp_clean_pud_entry - The pagewalk pud callb
141 *
142 * Dirty-tracking should take place on the PTE
143 * WARN() if encountering a dirty huge puds.
144 * Furthermore, never split huge puds, since t
145 * causes dirty info loss. The pagefault handl
146 * that if needed.
147 */
148 static int wp_clean_pud_entry(pud_t *pud, unsi
149 struct mm_walk *
150 {
151 #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_P
152 pud_t pudval = READ_ONCE(*pud);
153
154 /* Do not split a huge pud */
155 if (pud_trans_huge(pudval) || pud_devm
156 WARN_ON(pud_write(pudval) || p
157 walk->action = ACTION_CONTINUE
158 }
159 #endif
160 return 0;
161 }
162
163 /*
164 * wp_clean_pre_vma - The pagewalk pre_vma cal
165 *
166 * The pre_vma callback performs the cache flu
167 * and calls the necessary mmu notifiers.
168 */
169 static int wp_clean_pre_vma(unsigned long star
170 struct mm_walk *wa
171 {
172 struct wp_walk *wpwalk = walk->private
173
174 wpwalk->tlbflush_start = end;
175 wpwalk->tlbflush_end = start;
176
177 mmu_notifier_range_init(&wpwalk->range
178 walk->mm, star
179 mmu_notifier_invalidate_range_start(&w
180 flush_cache_range(walk->vma, start, en
181
182 /*
183 * We're not using tlb_gather_mmu() si
184 * only a small subrange of PTEs are a
185 * tlb_gather_mmu() records the full r
186 */
187 inc_tlb_flush_pending(walk->mm);
188
189 return 0;
190 }
191
192 /*
193 * wp_clean_post_vma - The pagewalk post_vma c
194 *
195 * The post_vma callback performs the tlb flus
196 * notifiers.
197 */
198 static void wp_clean_post_vma(struct mm_walk *
199 {
200 struct wp_walk *wpwalk = walk->private
201
202 if (mm_tlb_flush_nested(walk->mm))
203 flush_tlb_range(walk->vma, wpw
204 wpwalk->range.
205 else if (wpwalk->tlbflush_end > wpwalk
206 flush_tlb_range(walk->vma, wpw
207 wpwalk->tlbflu
208
209 mmu_notifier_invalidate_range_end(&wpw
210 dec_tlb_flush_pending(walk->mm);
211 }
212
213 /*
214 * wp_clean_test_walk - The pagewalk test_walk
215 *
216 * Won't perform dirty-tracking on COW, read-o
217 */
218 static int wp_clean_test_walk(unsigned long st
219 struct mm_walk *
220 {
221 unsigned long vm_flags = READ_ONCE(wal
222
223 /* Skip non-applicable VMAs */
224 if ((vm_flags & (VM_SHARED | VM_MAYWRI
225 (VM_SHARED | VM_MAYWRITE))
226 return 1;
227
228 return 0;
229 }
230
231 static const struct mm_walk_ops clean_walk_ops
232 .pte_entry = clean_record_pte,
233 .pmd_entry = wp_clean_pmd_entry,
234 .pud_entry = wp_clean_pud_entry,
235 .test_walk = wp_clean_test_walk,
236 .pre_vma = wp_clean_pre_vma,
237 .post_vma = wp_clean_post_vma
238 };
239
240 static const struct mm_walk_ops wp_walk_ops =
241 .pte_entry = wp_pte,
242 .pmd_entry = wp_clean_pmd_entry,
243 .pud_entry = wp_clean_pud_entry,
244 .test_walk = wp_clean_test_walk,
245 .pre_vma = wp_clean_pre_vma,
246 .post_vma = wp_clean_post_vma
247 };
248
249 /**
250 * wp_shared_mapping_range - Write-protect all
251 * @mapping: The address_space we want to writ
252 * @first_index: The first page offset in the
253 * @nr: Number of incremental page offsets to
254 *
255 * Note: This function currently skips transhu
256 * it's intended for dirty-tracking on the PTE
257 * encountering transhuge write-enabled entrie
258 * extended to handle them as well.
259 *
260 * Return: The number of ptes actually write-p
261 * already write-protected ptes are not counte
262 */
263 unsigned long wp_shared_mapping_range(struct a
264 pgoff_t
265 {
266 struct wp_walk wpwalk = { .total = 0 }
267
268 i_mmap_lock_read(mapping);
269 WARN_ON(walk_page_mapping(mapping, fir
270 &wpwalk));
271 i_mmap_unlock_read(mapping);
272
273 return wpwalk.total;
274 }
275 EXPORT_SYMBOL_GPL(wp_shared_mapping_range);
276
277 /**
278 * clean_record_shared_mapping_range - Clean a
279 * address space range
280 * @mapping: The address_space we want to clea
281 * @first_index: The first page offset in the
282 * @nr: Number of incremental page offsets to
283 * @bitmap_pgoff: The page offset of the first
284 * @bitmap: Pointer to a bitmap of at least @n
285 * cover the whole range @first_index..@first_
286 * @start: Pointer to number of the first set
287 * is modified as new bits are set by the func
288 * @end: Pointer to the number of the last set
289 * none set. The value is modified as new bits
290 *
291 * When this function returns there is no guar
292 * not already dirtied new ptes. However it wi
293 * reported in the bitmap. The guarantees are
294 *
295 * * All ptes dirty when the function starts e
296 * in the bitmap.
297 * * All ptes dirtied after that will either r
298 * bitmap or both.
299 *
300 * If a caller needs to make sure all dirty pt
301 * additional are added, it first needs to wri
302 * range and make sure new writers are blocked
303 * pfn_mkwrite(). And then after a TLB flush f
304 * pick up all dirty bits.
305 *
306 * This function currently skips transhuge pag
307 * it's intended for dirty-tracking on the PTE
308 * encountering transhuge dirty entries, thoug
309 * to handle them as well.
310 *
311 * Return: The number of dirty ptes actually c
312 */
313 unsigned long clean_record_shared_mapping_rang
314
315
316
317
318
319 {
320 bool none_set = (*start >= *end);
321 struct clean_walk cwalk = {
322 .base = { .total = 0 },
323 .bitmap_pgoff = bitmap_pgoff,
324 .bitmap = bitmap,
325 .start = none_set ? nr : *star
326 .end = none_set ? 0 : *end,
327 };
328
329 i_mmap_lock_read(mapping);
330 WARN_ON(walk_page_mapping(mapping, fir
331 &cwalk.base)
332 i_mmap_unlock_read(mapping);
333
334 *start = cwalk.start;
335 *end = cwalk.end;
336
337 return cwalk.base.total;
338 }
339 EXPORT_SYMBOL_GPL(clean_record_shared_mapping_
340
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.