Commit | Line | Data |
---|---|---|
c5acad84 TH |
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 <asm/cacheflush.h> | |
7 | #include <asm/tlbflush.h> | |
8 | ||
9 | /** | |
10 | * struct wp_walk - Private struct for pagetable walk callbacks | |
11 | * @range: Range for mmu notifiers | |
12 | * @tlbflush_start: Address of first modified pte | |
13 | * @tlbflush_end: Address of last modified pte + 1 | |
14 | * @total: Total number of modified ptes | |
15 | */ | |
16 | struct wp_walk { | |
17 | struct mmu_notifier_range range; | |
18 | unsigned long tlbflush_start; | |
19 | unsigned long tlbflush_end; | |
20 | unsigned long total; | |
21 | }; | |
22 | ||
23 | /** | |
24 | * wp_pte - Write-protect a pte | |
25 | * @pte: Pointer to the pte | |
26 | * @addr: The virtual page address | |
27 | * @walk: pagetable walk callback argument | |
28 | * | |
29 | * The function write-protects a pte and records the range in | |
30 | * virtual address space of touched ptes for efficient range TLB flushes. | |
31 | */ | |
32 | static int wp_pte(pte_t *pte, unsigned long addr, unsigned long end, | |
33 | struct mm_walk *walk) | |
34 | { | |
35 | struct wp_walk *wpwalk = walk->private; | |
36 | pte_t ptent = *pte; | |
37 | ||
38 | if (pte_write(ptent)) { | |
39 | pte_t old_pte = ptep_modify_prot_start(walk->vma, addr, pte); | |
40 | ||
41 | ptent = pte_wrprotect(old_pte); | |
42 | ptep_modify_prot_commit(walk->vma, addr, pte, old_pte, ptent); | |
43 | wpwalk->total++; | |
44 | wpwalk->tlbflush_start = min(wpwalk->tlbflush_start, addr); | |
45 | wpwalk->tlbflush_end = max(wpwalk->tlbflush_end, | |
46 | addr + PAGE_SIZE); | |
47 | } | |
48 | ||
49 | return 0; | |
50 | } | |
51 | ||
52 | /** | |
53 | * struct clean_walk - Private struct for the clean_record_pte function. | |
54 | * @base: struct wp_walk we derive from | |
55 | * @bitmap_pgoff: Address_space Page offset of the first bit in @bitmap | |
56 | * @bitmap: Bitmap with one bit for each page offset in the address_space range | |
57 | * covered. | |
58 | * @start: Address_space page offset of first modified pte relative | |
59 | * to @bitmap_pgoff | |
60 | * @end: Address_space page offset of last modified pte relative | |
61 | * to @bitmap_pgoff | |
62 | */ | |
63 | struct clean_walk { | |
64 | struct wp_walk base; | |
65 | pgoff_t bitmap_pgoff; | |
66 | unsigned long *bitmap; | |
67 | pgoff_t start; | |
68 | pgoff_t end; | |
69 | }; | |
70 | ||
71 | #define to_clean_walk(_wpwalk) container_of(_wpwalk, struct clean_walk, base) | |
72 | ||
73 | /** | |
74 | * clean_record_pte - Clean a pte and record its address space offset in a | |
75 | * bitmap | |
76 | * @pte: Pointer to the pte | |
77 | * @addr: The virtual page address | |
78 | * @walk: pagetable walk callback argument | |
79 | * | |
80 | * The function cleans a pte and records the range in | |
81 | * virtual address space of touched ptes for efficient TLB flushes. | |
82 | * It also records dirty ptes in a bitmap representing page offsets | |
83 | * in the address_space, as well as the first and last of the bits | |
84 | * touched. | |
85 | */ | |
86 | static int clean_record_pte(pte_t *pte, unsigned long addr, | |
87 | unsigned long end, struct mm_walk *walk) | |
88 | { | |
89 | struct wp_walk *wpwalk = walk->private; | |
90 | struct clean_walk *cwalk = to_clean_walk(wpwalk); | |
91 | pte_t ptent = *pte; | |
92 | ||
93 | if (pte_dirty(ptent)) { | |
94 | pgoff_t pgoff = ((addr - walk->vma->vm_start) >> PAGE_SHIFT) + | |
95 | walk->vma->vm_pgoff - cwalk->bitmap_pgoff; | |
96 | pte_t old_pte = ptep_modify_prot_start(walk->vma, addr, pte); | |
97 | ||
98 | ptent = pte_mkclean(old_pte); | |
99 | ptep_modify_prot_commit(walk->vma, addr, pte, old_pte, ptent); | |
100 | ||
101 | wpwalk->total++; | |
102 | wpwalk->tlbflush_start = min(wpwalk->tlbflush_start, addr); | |
103 | wpwalk->tlbflush_end = max(wpwalk->tlbflush_end, | |
104 | addr + PAGE_SIZE); | |
105 | ||
106 | __set_bit(pgoff, cwalk->bitmap); | |
107 | cwalk->start = min(cwalk->start, pgoff); | |
108 | cwalk->end = max(cwalk->end, pgoff + 1); | |
109 | } | |
110 | ||
111 | return 0; | |
112 | } | |
113 | ||
b2a403fd TH |
114 | /* |
115 | * wp_clean_pmd_entry - The pagewalk pmd callback. | |
116 | * | |
117 | * Dirty-tracking should take place on the PTE level, so | |
118 | * WARN() if encountering a dirty huge pmd. | |
119 | * Furthermore, never split huge pmds, since that currently | |
120 | * causes dirty info loss. The pagefault handler should do | |
121 | * that if needed. | |
122 | */ | |
c5acad84 TH |
123 | static int wp_clean_pmd_entry(pmd_t *pmd, unsigned long addr, unsigned long end, |
124 | struct mm_walk *walk) | |
125 | { | |
c5acad84 TH |
126 | pmd_t pmdval = pmd_read_atomic(pmd); |
127 | ||
b2a403fd TH |
128 | if (!pmd_trans_unstable(&pmdval)) |
129 | return 0; | |
130 | ||
131 | if (pmd_none(pmdval)) { | |
132 | walk->action = ACTION_AGAIN; | |
133 | return 0; | |
134 | } | |
135 | ||
136 | /* Huge pmd, present or migrated */ | |
137 | walk->action = ACTION_CONTINUE; | |
c5acad84 TH |
138 | if (pmd_trans_huge(pmdval) || pmd_devmap(pmdval)) |
139 | WARN_ON(pmd_write(pmdval) || pmd_dirty(pmdval)); | |
140 | ||
141 | return 0; | |
142 | } | |
143 | ||
b2a403fd TH |
144 | /* |
145 | * wp_clean_pud_entry - The pagewalk pud callback. | |
146 | * | |
147 | * Dirty-tracking should take place on the PTE level, so | |
148 | * WARN() if encountering a dirty huge puds. | |
149 | * Furthermore, never split huge puds, since that currently | |
150 | * causes dirty info loss. The pagefault handler should do | |
151 | * that if needed. | |
152 | */ | |
c5acad84 TH |
153 | static int wp_clean_pud_entry(pud_t *pud, unsigned long addr, unsigned long end, |
154 | struct mm_walk *walk) | |
155 | { | |
c5acad84 TH |
156 | pud_t pudval = READ_ONCE(*pud); |
157 | ||
b2a403fd TH |
158 | if (!pud_trans_unstable(&pudval)) |
159 | return 0; | |
160 | ||
161 | if (pud_none(pudval)) { | |
162 | walk->action = ACTION_AGAIN; | |
163 | return 0; | |
164 | } | |
165 | ||
166 | /* Huge pud */ | |
167 | walk->action = ACTION_CONTINUE; | |
c5acad84 TH |
168 | if (pud_trans_huge(pudval) || pud_devmap(pudval)) |
169 | WARN_ON(pud_write(pudval) || pud_dirty(pudval)); | |
170 | ||
171 | return 0; | |
172 | } | |
173 | ||
174 | /* | |
175 | * wp_clean_pre_vma - The pagewalk pre_vma callback. | |
176 | * | |
177 | * The pre_vma callback performs the cache flush, stages the tlb flush | |
178 | * and calls the necessary mmu notifiers. | |
179 | */ | |
180 | static int wp_clean_pre_vma(unsigned long start, unsigned long end, | |
181 | struct mm_walk *walk) | |
182 | { | |
183 | struct wp_walk *wpwalk = walk->private; | |
184 | ||
185 | wpwalk->tlbflush_start = end; | |
186 | wpwalk->tlbflush_end = start; | |
187 | ||
188 | mmu_notifier_range_init(&wpwalk->range, MMU_NOTIFY_PROTECTION_PAGE, 0, | |
189 | walk->vma, walk->mm, start, end); | |
190 | mmu_notifier_invalidate_range_start(&wpwalk->range); | |
191 | flush_cache_range(walk->vma, start, end); | |
192 | ||
193 | /* | |
194 | * We're not using tlb_gather_mmu() since typically | |
195 | * only a small subrange of PTEs are affected, whereas | |
196 | * tlb_gather_mmu() records the full range. | |
197 | */ | |
198 | inc_tlb_flush_pending(walk->mm); | |
199 | ||
200 | return 0; | |
201 | } | |
202 | ||
203 | /* | |
204 | * wp_clean_post_vma - The pagewalk post_vma callback. | |
205 | * | |
206 | * The post_vma callback performs the tlb flush and calls necessary mmu | |
207 | * notifiers. | |
208 | */ | |
209 | static void wp_clean_post_vma(struct mm_walk *walk) | |
210 | { | |
211 | struct wp_walk *wpwalk = walk->private; | |
212 | ||
213 | if (mm_tlb_flush_nested(walk->mm)) | |
214 | flush_tlb_range(walk->vma, wpwalk->range.start, | |
215 | wpwalk->range.end); | |
216 | else if (wpwalk->tlbflush_end > wpwalk->tlbflush_start) | |
217 | flush_tlb_range(walk->vma, wpwalk->tlbflush_start, | |
218 | wpwalk->tlbflush_end); | |
219 | ||
220 | mmu_notifier_invalidate_range_end(&wpwalk->range); | |
221 | dec_tlb_flush_pending(walk->mm); | |
222 | } | |
223 | ||
224 | /* | |
225 | * wp_clean_test_walk - The pagewalk test_walk callback. | |
226 | * | |
227 | * Won't perform dirty-tracking on COW, read-only or HUGETLB vmas. | |
228 | */ | |
229 | static int wp_clean_test_walk(unsigned long start, unsigned long end, | |
230 | struct mm_walk *walk) | |
231 | { | |
232 | unsigned long vm_flags = READ_ONCE(walk->vma->vm_flags); | |
233 | ||
234 | /* Skip non-applicable VMAs */ | |
235 | if ((vm_flags & (VM_SHARED | VM_MAYWRITE | VM_HUGETLB)) != | |
236 | (VM_SHARED | VM_MAYWRITE)) | |
237 | return 1; | |
238 | ||
239 | return 0; | |
240 | } | |
241 | ||
242 | static const struct mm_walk_ops clean_walk_ops = { | |
243 | .pte_entry = clean_record_pte, | |
244 | .pmd_entry = wp_clean_pmd_entry, | |
245 | .pud_entry = wp_clean_pud_entry, | |
246 | .test_walk = wp_clean_test_walk, | |
247 | .pre_vma = wp_clean_pre_vma, | |
248 | .post_vma = wp_clean_post_vma | |
249 | }; | |
250 | ||
251 | static const struct mm_walk_ops wp_walk_ops = { | |
252 | .pte_entry = wp_pte, | |
253 | .pmd_entry = wp_clean_pmd_entry, | |
254 | .pud_entry = wp_clean_pud_entry, | |
255 | .test_walk = wp_clean_test_walk, | |
256 | .pre_vma = wp_clean_pre_vma, | |
257 | .post_vma = wp_clean_post_vma | |
258 | }; | |
259 | ||
260 | /** | |
261 | * wp_shared_mapping_range - Write-protect all ptes in an address space range | |
262 | * @mapping: The address_space we want to write protect | |
263 | * @first_index: The first page offset in the range | |
264 | * @nr: Number of incremental page offsets to cover | |
265 | * | |
266 | * Note: This function currently skips transhuge page-table entries, since | |
267 | * it's intended for dirty-tracking on the PTE level. It will warn on | |
268 | * encountering transhuge write-enabled entries, though, and can easily be | |
269 | * extended to handle them as well. | |
270 | * | |
271 | * Return: The number of ptes actually write-protected. Note that | |
272 | * already write-protected ptes are not counted. | |
273 | */ | |
274 | unsigned long wp_shared_mapping_range(struct address_space *mapping, | |
275 | pgoff_t first_index, pgoff_t nr) | |
276 | { | |
277 | struct wp_walk wpwalk = { .total = 0 }; | |
278 | ||
279 | i_mmap_lock_read(mapping); | |
280 | WARN_ON(walk_page_mapping(mapping, first_index, nr, &wp_walk_ops, | |
281 | &wpwalk)); | |
282 | i_mmap_unlock_read(mapping); | |
283 | ||
284 | return wpwalk.total; | |
285 | } | |
286 | EXPORT_SYMBOL_GPL(wp_shared_mapping_range); | |
287 | ||
288 | /** | |
289 | * clean_record_shared_mapping_range - Clean and record all ptes in an | |
290 | * address space range | |
291 | * @mapping: The address_space we want to clean | |
292 | * @first_index: The first page offset in the range | |
293 | * @nr: Number of incremental page offsets to cover | |
294 | * @bitmap_pgoff: The page offset of the first bit in @bitmap | |
295 | * @bitmap: Pointer to a bitmap of at least @nr bits. The bitmap needs to | |
296 | * cover the whole range @first_index..@first_index + @nr. | |
297 | * @start: Pointer to number of the first set bit in @bitmap. | |
298 | * is modified as new bits are set by the function. | |
299 | * @end: Pointer to the number of the last set bit in @bitmap. | |
300 | * none set. The value is modified as new bits are set by the function. | |
301 | * | |
302 | * Note: When this function returns there is no guarantee that a CPU has | |
303 | * not already dirtied new ptes. However it will not clean any ptes not | |
304 | * reported in the bitmap. The guarantees are as follows: | |
305 | * a) All ptes dirty when the function starts executing will end up recorded | |
306 | * in the bitmap. | |
307 | * b) All ptes dirtied after that will either remain dirty, be recorded in the | |
308 | * bitmap or both. | |
309 | * | |
310 | * If a caller needs to make sure all dirty ptes are picked up and none | |
311 | * additional are added, it first needs to write-protect the address-space | |
312 | * range and make sure new writers are blocked in page_mkwrite() or | |
313 | * pfn_mkwrite(). And then after a TLB flush following the write-protection | |
314 | * pick up all dirty bits. | |
315 | * | |
316 | * Note: This function currently skips transhuge page-table entries, since | |
317 | * it's intended for dirty-tracking on the PTE level. It will warn on | |
318 | * encountering transhuge dirty entries, though, and can easily be extended | |
319 | * to handle them as well. | |
320 | * | |
321 | * Return: The number of dirty ptes actually cleaned. | |
322 | */ | |
323 | unsigned long clean_record_shared_mapping_range(struct address_space *mapping, | |
324 | pgoff_t first_index, pgoff_t nr, | |
325 | pgoff_t bitmap_pgoff, | |
326 | unsigned long *bitmap, | |
327 | pgoff_t *start, | |
328 | pgoff_t *end) | |
329 | { | |
330 | bool none_set = (*start >= *end); | |
331 | struct clean_walk cwalk = { | |
332 | .base = { .total = 0 }, | |
333 | .bitmap_pgoff = bitmap_pgoff, | |
334 | .bitmap = bitmap, | |
335 | .start = none_set ? nr : *start, | |
336 | .end = none_set ? 0 : *end, | |
337 | }; | |
338 | ||
339 | i_mmap_lock_read(mapping); | |
340 | WARN_ON(walk_page_mapping(mapping, first_index, nr, &clean_walk_ops, | |
341 | &cwalk.base)); | |
342 | i_mmap_unlock_read(mapping); | |
343 | ||
344 | *start = cwalk.start; | |
345 | *end = cwalk.end; | |
346 | ||
347 | return cwalk.base.total; | |
348 | } | |
349 | EXPORT_SYMBOL_GPL(clean_record_shared_mapping_range); |