xref: /linux/mm/mapping_dirty_helpers.c (revision f8e17c17b81070f38062dce79ca7f4541851dadd)
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 
114 /* wp_clean_pmd_entry - The pagewalk pmd callback. */
115 static int wp_clean_pmd_entry(pmd_t *pmd, unsigned long addr, unsigned long end,
116 			      struct mm_walk *walk)
117 {
118 	/* Dirty-tracking should be handled on the pte level */
119 	pmd_t pmdval = pmd_read_atomic(pmd);
120 
121 	if (pmd_trans_huge(pmdval) || pmd_devmap(pmdval))
122 		WARN_ON(pmd_write(pmdval) || pmd_dirty(pmdval));
123 
124 	return 0;
125 }
126 
127 /* wp_clean_pud_entry - The pagewalk pud callback. */
128 static int wp_clean_pud_entry(pud_t *pud, unsigned long addr, unsigned long end,
129 			      struct mm_walk *walk)
130 {
131 	/* Dirty-tracking should be handled on the pte level */
132 	pud_t pudval = READ_ONCE(*pud);
133 
134 	if (pud_trans_huge(pudval) || pud_devmap(pudval))
135 		WARN_ON(pud_write(pudval) || pud_dirty(pudval));
136 
137 	return 0;
138 }
139 
140 /*
141  * wp_clean_pre_vma - The pagewalk pre_vma callback.
142  *
143  * The pre_vma callback performs the cache flush, stages the tlb flush
144  * and calls the necessary mmu notifiers.
145  */
146 static int wp_clean_pre_vma(unsigned long start, unsigned long end,
147 			    struct mm_walk *walk)
148 {
149 	struct wp_walk *wpwalk = walk->private;
150 
151 	wpwalk->tlbflush_start = end;
152 	wpwalk->tlbflush_end = start;
153 
154 	mmu_notifier_range_init(&wpwalk->range, MMU_NOTIFY_PROTECTION_PAGE, 0,
155 				walk->vma, walk->mm, start, end);
156 	mmu_notifier_invalidate_range_start(&wpwalk->range);
157 	flush_cache_range(walk->vma, start, end);
158 
159 	/*
160 	 * We're not using tlb_gather_mmu() since typically
161 	 * only a small subrange of PTEs are affected, whereas
162 	 * tlb_gather_mmu() records the full range.
163 	 */
164 	inc_tlb_flush_pending(walk->mm);
165 
166 	return 0;
167 }
168 
169 /*
170  * wp_clean_post_vma - The pagewalk post_vma callback.
171  *
172  * The post_vma callback performs the tlb flush and calls necessary mmu
173  * notifiers.
174  */
175 static void wp_clean_post_vma(struct mm_walk *walk)
176 {
177 	struct wp_walk *wpwalk = walk->private;
178 
179 	if (mm_tlb_flush_nested(walk->mm))
180 		flush_tlb_range(walk->vma, wpwalk->range.start,
181 				wpwalk->range.end);
182 	else if (wpwalk->tlbflush_end > wpwalk->tlbflush_start)
183 		flush_tlb_range(walk->vma, wpwalk->tlbflush_start,
184 				wpwalk->tlbflush_end);
185 
186 	mmu_notifier_invalidate_range_end(&wpwalk->range);
187 	dec_tlb_flush_pending(walk->mm);
188 }
189 
190 /*
191  * wp_clean_test_walk - The pagewalk test_walk callback.
192  *
193  * Won't perform dirty-tracking on COW, read-only or HUGETLB vmas.
194  */
195 static int wp_clean_test_walk(unsigned long start, unsigned long end,
196 			      struct mm_walk *walk)
197 {
198 	unsigned long vm_flags = READ_ONCE(walk->vma->vm_flags);
199 
200 	/* Skip non-applicable VMAs */
201 	if ((vm_flags & (VM_SHARED | VM_MAYWRITE | VM_HUGETLB)) !=
202 	    (VM_SHARED | VM_MAYWRITE))
203 		return 1;
204 
205 	return 0;
206 }
207 
208 static const struct mm_walk_ops clean_walk_ops = {
209 	.pte_entry = clean_record_pte,
210 	.pmd_entry = wp_clean_pmd_entry,
211 	.pud_entry = wp_clean_pud_entry,
212 	.test_walk = wp_clean_test_walk,
213 	.pre_vma = wp_clean_pre_vma,
214 	.post_vma = wp_clean_post_vma
215 };
216 
217 static const struct mm_walk_ops wp_walk_ops = {
218 	.pte_entry = wp_pte,
219 	.pmd_entry = wp_clean_pmd_entry,
220 	.pud_entry = wp_clean_pud_entry,
221 	.test_walk = wp_clean_test_walk,
222 	.pre_vma = wp_clean_pre_vma,
223 	.post_vma = wp_clean_post_vma
224 };
225 
226 /**
227  * wp_shared_mapping_range - Write-protect all ptes in an address space range
228  * @mapping: The address_space we want to write protect
229  * @first_index: The first page offset in the range
230  * @nr: Number of incremental page offsets to cover
231  *
232  * Note: This function currently skips transhuge page-table entries, since
233  * it's intended for dirty-tracking on the PTE level. It will warn on
234  * encountering transhuge write-enabled entries, though, and can easily be
235  * extended to handle them as well.
236  *
237  * Return: The number of ptes actually write-protected. Note that
238  * already write-protected ptes are not counted.
239  */
240 unsigned long wp_shared_mapping_range(struct address_space *mapping,
241 				      pgoff_t first_index, pgoff_t nr)
242 {
243 	struct wp_walk wpwalk = { .total = 0 };
244 
245 	i_mmap_lock_read(mapping);
246 	WARN_ON(walk_page_mapping(mapping, first_index, nr, &wp_walk_ops,
247 				  &wpwalk));
248 	i_mmap_unlock_read(mapping);
249 
250 	return wpwalk.total;
251 }
252 EXPORT_SYMBOL_GPL(wp_shared_mapping_range);
253 
254 /**
255  * clean_record_shared_mapping_range - Clean and record all ptes in an
256  * address space range
257  * @mapping: The address_space we want to clean
258  * @first_index: The first page offset in the range
259  * @nr: Number of incremental page offsets to cover
260  * @bitmap_pgoff: The page offset of the first bit in @bitmap
261  * @bitmap: Pointer to a bitmap of at least @nr bits. The bitmap needs to
262  * cover the whole range @first_index..@first_index + @nr.
263  * @start: Pointer to number of the first set bit in @bitmap.
264  * is modified as new bits are set by the function.
265  * @end: Pointer to the number of the last set bit in @bitmap.
266  * none set. The value is modified as new bits are set by the function.
267  *
268  * Note: When this function returns there is no guarantee that a CPU has
269  * not already dirtied new ptes. However it will not clean any ptes not
270  * reported in the bitmap. The guarantees are as follows:
271  * a) All ptes dirty when the function starts executing will end up recorded
272  *    in the bitmap.
273  * b) All ptes dirtied after that will either remain dirty, be recorded in the
274  *    bitmap or both.
275  *
276  * If a caller needs to make sure all dirty ptes are picked up and none
277  * additional are added, it first needs to write-protect the address-space
278  * range and make sure new writers are blocked in page_mkwrite() or
279  * pfn_mkwrite(). And then after a TLB flush following the write-protection
280  * pick up all dirty bits.
281  *
282  * Note: This function currently skips transhuge page-table entries, since
283  * it's intended for dirty-tracking on the PTE level. It will warn on
284  * encountering transhuge dirty entries, though, and can easily be extended
285  * to handle them as well.
286  *
287  * Return: The number of dirty ptes actually cleaned.
288  */
289 unsigned long clean_record_shared_mapping_range(struct address_space *mapping,
290 						pgoff_t first_index, pgoff_t nr,
291 						pgoff_t bitmap_pgoff,
292 						unsigned long *bitmap,
293 						pgoff_t *start,
294 						pgoff_t *end)
295 {
296 	bool none_set = (*start >= *end);
297 	struct clean_walk cwalk = {
298 		.base = { .total = 0 },
299 		.bitmap_pgoff = bitmap_pgoff,
300 		.bitmap = bitmap,
301 		.start = none_set ? nr : *start,
302 		.end = none_set ? 0 : *end,
303 	};
304 
305 	i_mmap_lock_read(mapping);
306 	WARN_ON(walk_page_mapping(mapping, first_index, nr, &clean_walk_ops,
307 				  &cwalk.base));
308 	i_mmap_unlock_read(mapping);
309 
310 	*start = cwalk.start;
311 	*end = cwalk.end;
312 
313 	return cwalk.base.total;
314 }
315 EXPORT_SYMBOL_GPL(clean_record_shared_mapping_range);
316