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