1 | // SPDX-License-Identifier: GPL-2.0 |
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 pagetable walk callbacks |
12 | * @range: Range for mmu notifiers |
13 | * @tlbflush_start: Address of first modified pte |
14 | * @tlbflush_end: Address of last modified pte + 1 |
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 address |
28 | * @end: The end of protecting virtual address |
29 | * @walk: pagetable walk callback argument |
30 | * |
31 | * The function write-protects a pte and records the range in |
32 | * virtual address space of touched ptes for efficient range TLB flushes. |
33 | */ |
34 | static int wp_pte(pte_t *pte, unsigned long addr, unsigned long end, |
35 | struct mm_walk *walk) |
36 | { |
37 | struct wp_walk *wpwalk = walk->private; |
38 | pte_t ptent = ptep_get(ptep: pte); |
39 | |
40 | if (pte_write(pte: ptent)) { |
41 | pte_t old_pte = ptep_modify_prot_start(vma: walk->vma, addr, ptep: pte); |
42 | |
43 | ptent = pte_wrprotect(pte: old_pte); |
44 | ptep_modify_prot_commit(vma: walk->vma, addr, ptep: pte, old_pte, pte: ptent); |
45 | wpwalk->total++; |
46 | wpwalk->tlbflush_start = min(wpwalk->tlbflush_start, addr); |
47 | wpwalk->tlbflush_end = max(wpwalk->tlbflush_end, |
48 | addr + PAGE_SIZE); |
49 | } |
50 | |
51 | return 0; |
52 | } |
53 | |
54 | /** |
55 | * struct clean_walk - Private struct for the clean_record_pte function. |
56 | * @base: struct wp_walk we derive from |
57 | * @bitmap_pgoff: Address_space Page offset of the first bit in @bitmap |
58 | * @bitmap: Bitmap with one bit for each page offset in the address_space range |
59 | * covered. |
60 | * @start: Address_space page offset of first modified pte relative |
61 | * to @bitmap_pgoff |
62 | * @end: Address_space page offset of last modified pte relative |
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(_wpwalk, struct clean_walk, base) |
74 | |
75 | /** |
76 | * clean_record_pte - Clean a pte and record its address space offset in a |
77 | * bitmap |
78 | * @pte: Pointer to the pte |
79 | * @addr: The start of virtual address to be clean |
80 | * @end: The end of virtual address to be clean |
81 | * @walk: pagetable walk callback argument |
82 | * |
83 | * The function cleans a pte and records the range in |
84 | * virtual address space of touched ptes for efficient TLB flushes. |
85 | * It also records dirty ptes in a bitmap representing page offsets |
86 | * in the address_space, as well as the first and last of the bits |
87 | * touched. |
88 | */ |
89 | static int clean_record_pte(pte_t *pte, unsigned long addr, |
90 | unsigned long end, struct mm_walk *walk) |
91 | { |
92 | struct wp_walk *wpwalk = walk->private; |
93 | struct clean_walk *cwalk = to_clean_walk(wpwalk); |
94 | pte_t ptent = ptep_get(ptep: pte); |
95 | |
96 | if (pte_dirty(pte: ptent)) { |
97 | pgoff_t pgoff = ((addr - walk->vma->vm_start) >> PAGE_SHIFT) + |
98 | walk->vma->vm_pgoff - cwalk->bitmap_pgoff; |
99 | pte_t old_pte = ptep_modify_prot_start(vma: walk->vma, addr, ptep: pte); |
100 | |
101 | ptent = pte_mkclean(pte: old_pte); |
102 | ptep_modify_prot_commit(vma: walk->vma, addr, ptep: pte, old_pte, pte: ptent); |
103 | |
104 | wpwalk->total++; |
105 | wpwalk->tlbflush_start = min(wpwalk->tlbflush_start, addr); |
106 | wpwalk->tlbflush_end = max(wpwalk->tlbflush_end, |
107 | addr + PAGE_SIZE); |
108 | |
109 | __set_bit(pgoff, cwalk->bitmap); |
110 | cwalk->start = min(cwalk->start, pgoff); |
111 | cwalk->end = max(cwalk->end, pgoff + 1); |
112 | } |
113 | |
114 | return 0; |
115 | } |
116 | |
117 | /* |
118 | * wp_clean_pmd_entry - The pagewalk pmd callback. |
119 | * |
120 | * Dirty-tracking should take place on the PTE level, so |
121 | * WARN() if encountering a dirty huge pmd. |
122 | * Furthermore, never split huge pmds, since that currently |
123 | * causes dirty info loss. The pagefault handler should do |
124 | * that if needed. |
125 | */ |
126 | static int wp_clean_pmd_entry(pmd_t *pmd, unsigned long addr, unsigned long end, |
127 | struct mm_walk *walk) |
128 | { |
129 | pmd_t pmdval = pmdp_get_lockless(pmdp: pmd); |
130 | |
131 | /* Do not split a huge pmd, present or migrated */ |
132 | if (pmd_trans_huge(pmd: pmdval) || pmd_devmap(pmd: pmdval)) { |
133 | WARN_ON(pmd_write(pmdval) || pmd_dirty(pmdval)); |
134 | walk->action = ACTION_CONTINUE; |
135 | } |
136 | return 0; |
137 | } |
138 | |
139 | /* |
140 | * wp_clean_pud_entry - The pagewalk pud callback. |
141 | * |
142 | * Dirty-tracking should take place on the PTE level, so |
143 | * WARN() if encountering a dirty huge puds. |
144 | * Furthermore, never split huge puds, since that currently |
145 | * causes dirty info loss. The pagefault handler should do |
146 | * that if needed. |
147 | */ |
148 | static int wp_clean_pud_entry(pud_t *pud, unsigned long addr, unsigned long end, |
149 | struct mm_walk *walk) |
150 | { |
151 | #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD |
152 | pud_t pudval = READ_ONCE(*pud); |
153 | |
154 | /* Do not split a huge pud */ |
155 | if (pud_trans_huge(pud: pudval) || pud_devmap(pud: pudval)) { |
156 | WARN_ON(pud_write(pudval) || pud_dirty(pudval)); |
157 | walk->action = ACTION_CONTINUE; |
158 | } |
159 | #endif |
160 | return 0; |
161 | } |
162 | |
163 | /* |
164 | * wp_clean_pre_vma - The pagewalk pre_vma callback. |
165 | * |
166 | * The pre_vma callback performs the cache flush, stages the tlb flush |
167 | * and calls the necessary mmu notifiers. |
168 | */ |
169 | static int wp_clean_pre_vma(unsigned long start, unsigned long end, |
170 | struct mm_walk *walk) |
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(range: &wpwalk->range, event: MMU_NOTIFY_PROTECTION_PAGE, flags: 0, |
178 | mm: walk->mm, start, end); |
179 | mmu_notifier_invalidate_range_start(range: &wpwalk->range); |
180 | flush_cache_range(vma: walk->vma, start, end); |
181 | |
182 | /* |
183 | * We're not using tlb_gather_mmu() since typically |
184 | * only a small subrange of PTEs are affected, whereas |
185 | * tlb_gather_mmu() records the full range. |
186 | */ |
187 | inc_tlb_flush_pending(mm: walk->mm); |
188 | |
189 | return 0; |
190 | } |
191 | |
192 | /* |
193 | * wp_clean_post_vma - The pagewalk post_vma callback. |
194 | * |
195 | * The post_vma callback performs the tlb flush and calls necessary mmu |
196 | * notifiers. |
197 | */ |
198 | static void wp_clean_post_vma(struct mm_walk *walk) |
199 | { |
200 | struct wp_walk *wpwalk = walk->private; |
201 | |
202 | if (mm_tlb_flush_nested(mm: walk->mm)) |
203 | flush_tlb_range(walk->vma, wpwalk->range.start, |
204 | wpwalk->range.end); |
205 | else if (wpwalk->tlbflush_end > wpwalk->tlbflush_start) |
206 | flush_tlb_range(walk->vma, wpwalk->tlbflush_start, |
207 | wpwalk->tlbflush_end); |
208 | |
209 | mmu_notifier_invalidate_range_end(range: &wpwalk->range); |
210 | dec_tlb_flush_pending(mm: walk->mm); |
211 | } |
212 | |
213 | /* |
214 | * wp_clean_test_walk - The pagewalk test_walk callback. |
215 | * |
216 | * Won't perform dirty-tracking on COW, read-only or HUGETLB vmas. |
217 | */ |
218 | static int wp_clean_test_walk(unsigned long start, unsigned long end, |
219 | struct mm_walk *walk) |
220 | { |
221 | unsigned long vm_flags = READ_ONCE(walk->vma->vm_flags); |
222 | |
223 | /* Skip non-applicable VMAs */ |
224 | if ((vm_flags & (VM_SHARED | VM_MAYWRITE | VM_HUGETLB)) != |
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 ptes in an address space range |
251 | * @mapping: The address_space we want to write protect |
252 | * @first_index: The first page offset in the range |
253 | * @nr: Number of incremental page offsets to cover |
254 | * |
255 | * Note: This function currently skips transhuge page-table entries, since |
256 | * it's intended for dirty-tracking on the PTE level. It will warn on |
257 | * encountering transhuge write-enabled entries, though, and can easily be |
258 | * extended to handle them as well. |
259 | * |
260 | * Return: The number of ptes actually write-protected. Note that |
261 | * already write-protected ptes are not counted. |
262 | */ |
263 | unsigned long wp_shared_mapping_range(struct address_space *mapping, |
264 | pgoff_t first_index, pgoff_t nr) |
265 | { |
266 | struct wp_walk wpwalk = { .total = 0 }; |
267 | |
268 | i_mmap_lock_read(mapping); |
269 | WARN_ON(walk_page_mapping(mapping, first_index, nr, &wp_walk_ops, |
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 and record all ptes in an |
279 | * address space range |
280 | * @mapping: The address_space we want to clean |
281 | * @first_index: The first page offset in the range |
282 | * @nr: Number of incremental page offsets to cover |
283 | * @bitmap_pgoff: The page offset of the first bit in @bitmap |
284 | * @bitmap: Pointer to a bitmap of at least @nr bits. The bitmap needs to |
285 | * cover the whole range @first_index..@first_index + @nr. |
286 | * @start: Pointer to number of the first set bit in @bitmap. |
287 | * is modified as new bits are set by the function. |
288 | * @end: Pointer to the number of the last set bit in @bitmap. |
289 | * none set. The value is modified as new bits are set by the function. |
290 | * |
291 | * When this function returns there is no guarantee that a CPU has |
292 | * not already dirtied new ptes. However it will not clean any ptes not |
293 | * reported in the bitmap. The guarantees are as follows: |
294 | * |
295 | * * All ptes dirty when the function starts executing will end up recorded |
296 | * in the bitmap. |
297 | * * All ptes dirtied after that will either remain dirty, be recorded in the |
298 | * bitmap or both. |
299 | * |
300 | * If a caller needs to make sure all dirty ptes are picked up and none |
301 | * additional are added, it first needs to write-protect the address-space |
302 | * range and make sure new writers are blocked in page_mkwrite() or |
303 | * pfn_mkwrite(). And then after a TLB flush following the write-protection |
304 | * pick up all dirty bits. |
305 | * |
306 | * This function currently skips transhuge page-table entries, since |
307 | * it's intended for dirty-tracking on the PTE level. It will warn on |
308 | * encountering transhuge dirty entries, though, and can easily be extended |
309 | * to handle them as well. |
310 | * |
311 | * Return: The number of dirty ptes actually cleaned. |
312 | */ |
313 | unsigned long clean_record_shared_mapping_range(struct address_space *mapping, |
314 | pgoff_t first_index, pgoff_t nr, |
315 | pgoff_t bitmap_pgoff, |
316 | unsigned long *bitmap, |
317 | pgoff_t *start, |
318 | pgoff_t *end) |
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 : *start, |
326 | .end = none_set ? 0 : *end, |
327 | }; |
328 | |
329 | i_mmap_lock_read(mapping); |
330 | WARN_ON(walk_page_mapping(mapping, first_index, nr, &clean_walk_ops, |
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_range); |
340 | |