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> | |
36090def | 6 | #include <linux/mm_inline.h> |
c5acad84 TH |
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 | |
f5b7e739 AS |
27 | * @addr: The start of protecting virtual address |
28 | * @end: The end of protecting virtual address | |
c5acad84 TH |
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 = *pte; | |
39 | ||
40 | if (pte_write(ptent)) { | |
41 | pte_t old_pte = ptep_modify_prot_start(walk->vma, addr, pte); | |
42 | ||
43 | ptent = pte_wrprotect(old_pte); | |
44 | ptep_modify_prot_commit(walk->vma, addr, pte, old_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 | |
f5b7e739 AS |
79 | * @addr: The start of virtual address to be clean |
80 | * @end: The end of virtual address to be clean | |
c5acad84 TH |
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 = *pte; | |
95 | ||
96 | if (pte_dirty(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(walk->vma, addr, pte); | |
100 | ||
101 | ptent = pte_mkclean(old_pte); | |
102 | ptep_modify_prot_commit(walk->vma, addr, pte, old_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 | ||
b2a403fd TH |
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 | */ | |
c5acad84 TH |
126 | static int wp_clean_pmd_entry(pmd_t *pmd, unsigned long addr, unsigned long end, |
127 | struct mm_walk *walk) | |
128 | { | |
dab6e717 | 129 | pmd_t pmdval = pmdp_get_lockless(pmd); |
c5acad84 | 130 | |
b2a403fd TH |
131 | if (!pmd_trans_unstable(&pmdval)) |
132 | return 0; | |
133 | ||
134 | if (pmd_none(pmdval)) { | |
135 | walk->action = ACTION_AGAIN; | |
136 | return 0; | |
137 | } | |
138 | ||
139 | /* Huge pmd, present or migrated */ | |
140 | walk->action = ACTION_CONTINUE; | |
c5acad84 TH |
141 | if (pmd_trans_huge(pmdval) || pmd_devmap(pmdval)) |
142 | WARN_ON(pmd_write(pmdval) || pmd_dirty(pmdval)); | |
143 | ||
144 | return 0; | |
145 | } | |
146 | ||
b2a403fd TH |
147 | /* |
148 | * wp_clean_pud_entry - The pagewalk pud callback. | |
149 | * | |
150 | * Dirty-tracking should take place on the PTE level, so | |
151 | * WARN() if encountering a dirty huge puds. | |
152 | * Furthermore, never split huge puds, since that currently | |
153 | * causes dirty info loss. The pagefault handler should do | |
154 | * that if needed. | |
155 | */ | |
c5acad84 TH |
156 | static int wp_clean_pud_entry(pud_t *pud, unsigned long addr, unsigned long end, |
157 | struct mm_walk *walk) | |
158 | { | |
c5acad84 TH |
159 | pud_t pudval = READ_ONCE(*pud); |
160 | ||
b2a403fd TH |
161 | if (!pud_trans_unstable(&pudval)) |
162 | return 0; | |
163 | ||
164 | if (pud_none(pudval)) { | |
165 | walk->action = ACTION_AGAIN; | |
166 | return 0; | |
167 | } | |
168 | ||
94036f4c | 169 | #ifdef CONFIG_HAVE_ARCH_TRANSPARENT_HUGEPAGE_PUD |
b2a403fd TH |
170 | /* Huge pud */ |
171 | walk->action = ACTION_CONTINUE; | |
c5acad84 TH |
172 | if (pud_trans_huge(pudval) || pud_devmap(pudval)) |
173 | WARN_ON(pud_write(pudval) || pud_dirty(pudval)); | |
94036f4c | 174 | #endif |
c5acad84 TH |
175 | |
176 | return 0; | |
177 | } | |
178 | ||
179 | /* | |
180 | * wp_clean_pre_vma - The pagewalk pre_vma callback. | |
181 | * | |
182 | * The pre_vma callback performs the cache flush, stages the tlb flush | |
183 | * and calls the necessary mmu notifiers. | |
184 | */ | |
185 | static int wp_clean_pre_vma(unsigned long start, unsigned long end, | |
186 | struct mm_walk *walk) | |
187 | { | |
188 | struct wp_walk *wpwalk = walk->private; | |
189 | ||
190 | wpwalk->tlbflush_start = end; | |
191 | wpwalk->tlbflush_end = start; | |
192 | ||
193 | mmu_notifier_range_init(&wpwalk->range, MMU_NOTIFY_PROTECTION_PAGE, 0, | |
7d4a8be0 | 194 | walk->mm, start, end); |
c5acad84 TH |
195 | mmu_notifier_invalidate_range_start(&wpwalk->range); |
196 | flush_cache_range(walk->vma, start, end); | |
197 | ||
198 | /* | |
199 | * We're not using tlb_gather_mmu() since typically | |
200 | * only a small subrange of PTEs are affected, whereas | |
201 | * tlb_gather_mmu() records the full range. | |
202 | */ | |
203 | inc_tlb_flush_pending(walk->mm); | |
204 | ||
205 | return 0; | |
206 | } | |
207 | ||
208 | /* | |
209 | * wp_clean_post_vma - The pagewalk post_vma callback. | |
210 | * | |
211 | * The post_vma callback performs the tlb flush and calls necessary mmu | |
212 | * notifiers. | |
213 | */ | |
214 | static void wp_clean_post_vma(struct mm_walk *walk) | |
215 | { | |
216 | struct wp_walk *wpwalk = walk->private; | |
217 | ||
218 | if (mm_tlb_flush_nested(walk->mm)) | |
219 | flush_tlb_range(walk->vma, wpwalk->range.start, | |
220 | wpwalk->range.end); | |
221 | else if (wpwalk->tlbflush_end > wpwalk->tlbflush_start) | |
222 | flush_tlb_range(walk->vma, wpwalk->tlbflush_start, | |
223 | wpwalk->tlbflush_end); | |
224 | ||
225 | mmu_notifier_invalidate_range_end(&wpwalk->range); | |
226 | dec_tlb_flush_pending(walk->mm); | |
227 | } | |
228 | ||
229 | /* | |
230 | * wp_clean_test_walk - The pagewalk test_walk callback. | |
231 | * | |
232 | * Won't perform dirty-tracking on COW, read-only or HUGETLB vmas. | |
233 | */ | |
234 | static int wp_clean_test_walk(unsigned long start, unsigned long end, | |
235 | struct mm_walk *walk) | |
236 | { | |
237 | unsigned long vm_flags = READ_ONCE(walk->vma->vm_flags); | |
238 | ||
239 | /* Skip non-applicable VMAs */ | |
240 | if ((vm_flags & (VM_SHARED | VM_MAYWRITE | VM_HUGETLB)) != | |
241 | (VM_SHARED | VM_MAYWRITE)) | |
242 | return 1; | |
243 | ||
244 | return 0; | |
245 | } | |
246 | ||
247 | static const struct mm_walk_ops clean_walk_ops = { | |
248 | .pte_entry = clean_record_pte, | |
249 | .pmd_entry = wp_clean_pmd_entry, | |
250 | .pud_entry = wp_clean_pud_entry, | |
251 | .test_walk = wp_clean_test_walk, | |
252 | .pre_vma = wp_clean_pre_vma, | |
253 | .post_vma = wp_clean_post_vma | |
254 | }; | |
255 | ||
256 | static const struct mm_walk_ops wp_walk_ops = { | |
257 | .pte_entry = wp_pte, | |
258 | .pmd_entry = wp_clean_pmd_entry, | |
259 | .pud_entry = wp_clean_pud_entry, | |
260 | .test_walk = wp_clean_test_walk, | |
261 | .pre_vma = wp_clean_pre_vma, | |
262 | .post_vma = wp_clean_post_vma | |
263 | }; | |
264 | ||
265 | /** | |
266 | * wp_shared_mapping_range - Write-protect all ptes in an address space range | |
267 | * @mapping: The address_space we want to write protect | |
268 | * @first_index: The first page offset in the range | |
269 | * @nr: Number of incremental page offsets to cover | |
270 | * | |
271 | * Note: This function currently skips transhuge page-table entries, since | |
272 | * it's intended for dirty-tracking on the PTE level. It will warn on | |
273 | * encountering transhuge write-enabled entries, though, and can easily be | |
274 | * extended to handle them as well. | |
275 | * | |
276 | * Return: The number of ptes actually write-protected. Note that | |
277 | * already write-protected ptes are not counted. | |
278 | */ | |
279 | unsigned long wp_shared_mapping_range(struct address_space *mapping, | |
280 | pgoff_t first_index, pgoff_t nr) | |
281 | { | |
282 | struct wp_walk wpwalk = { .total = 0 }; | |
283 | ||
284 | i_mmap_lock_read(mapping); | |
285 | WARN_ON(walk_page_mapping(mapping, first_index, nr, &wp_walk_ops, | |
286 | &wpwalk)); | |
287 | i_mmap_unlock_read(mapping); | |
288 | ||
289 | return wpwalk.total; | |
290 | } | |
291 | EXPORT_SYMBOL_GPL(wp_shared_mapping_range); | |
292 | ||
293 | /** | |
294 | * clean_record_shared_mapping_range - Clean and record all ptes in an | |
295 | * address space range | |
296 | * @mapping: The address_space we want to clean | |
297 | * @first_index: The first page offset in the range | |
298 | * @nr: Number of incremental page offsets to cover | |
299 | * @bitmap_pgoff: The page offset of the first bit in @bitmap | |
300 | * @bitmap: Pointer to a bitmap of at least @nr bits. The bitmap needs to | |
301 | * cover the whole range @first_index..@first_index + @nr. | |
302 | * @start: Pointer to number of the first set bit in @bitmap. | |
303 | * is modified as new bits are set by the function. | |
304 | * @end: Pointer to the number of the last set bit in @bitmap. | |
305 | * none set. The value is modified as new bits are set by the function. | |
306 | * | |
307 | * Note: When this function returns there is no guarantee that a CPU has | |
308 | * not already dirtied new ptes. However it will not clean any ptes not | |
309 | * reported in the bitmap. The guarantees are as follows: | |
310 | * a) All ptes dirty when the function starts executing will end up recorded | |
311 | * in the bitmap. | |
312 | * b) All ptes dirtied after that will either remain dirty, be recorded in the | |
313 | * bitmap or both. | |
314 | * | |
315 | * If a caller needs to make sure all dirty ptes are picked up and none | |
316 | * additional are added, it first needs to write-protect the address-space | |
317 | * range and make sure new writers are blocked in page_mkwrite() or | |
318 | * pfn_mkwrite(). And then after a TLB flush following the write-protection | |
319 | * pick up all dirty bits. | |
320 | * | |
b417941f | 321 | * This function currently skips transhuge page-table entries, since |
c5acad84 TH |
322 | * it's intended for dirty-tracking on the PTE level. It will warn on |
323 | * encountering transhuge dirty entries, though, and can easily be extended | |
324 | * to handle them as well. | |
325 | * | |
326 | * Return: The number of dirty ptes actually cleaned. | |
327 | */ | |
328 | unsigned long clean_record_shared_mapping_range(struct address_space *mapping, | |
329 | pgoff_t first_index, pgoff_t nr, | |
330 | pgoff_t bitmap_pgoff, | |
331 | unsigned long *bitmap, | |
332 | pgoff_t *start, | |
333 | pgoff_t *end) | |
334 | { | |
335 | bool none_set = (*start >= *end); | |
336 | struct clean_walk cwalk = { | |
337 | .base = { .total = 0 }, | |
338 | .bitmap_pgoff = bitmap_pgoff, | |
339 | .bitmap = bitmap, | |
340 | .start = none_set ? nr : *start, | |
341 | .end = none_set ? 0 : *end, | |
342 | }; | |
343 | ||
344 | i_mmap_lock_read(mapping); | |
345 | WARN_ON(walk_page_mapping(mapping, first_index, nr, &clean_walk_ops, | |
346 | &cwalk.base)); | |
347 | i_mmap_unlock_read(mapping); | |
348 | ||
349 | *start = cwalk.start; | |
350 | *end = cwalk.end; | |
351 | ||
352 | return cwalk.base.total; | |
353 | } | |
354 | EXPORT_SYMBOL_GPL(clean_record_shared_mapping_range); |