Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
1da177e4 LT |
2 | #ifndef _LINUX_RMAP_H |
3 | #define _LINUX_RMAP_H | |
4 | /* | |
5 | * Declarations for Reverse Mapping functions in mm/rmap.c | |
6 | */ | |
7 | ||
1da177e4 LT |
8 | #include <linux/list.h> |
9 | #include <linux/slab.h> | |
10 | #include <linux/mm.h> | |
5a505085 | 11 | #include <linux/rwsem.h> |
bed7161a | 12 | #include <linux/memcontrol.h> |
ace71a19 | 13 | #include <linux/highmem.h> |
1da177e4 LT |
14 | |
15 | /* | |
16 | * The anon_vma heads a list of private "related" vmas, to scan if | |
17 | * an anonymous page pointing to this anon_vma needs to be unmapped: | |
18 | * the vmas on the list will be related by forking, or by splitting. | |
19 | * | |
20 | * Since vmas come and go as they are split and merged (particularly | |
21 | * in mprotect), the mapping field of an anonymous page cannot point | |
22 | * directly to a vma: instead it points to an anon_vma, on whose list | |
23 | * the related vmas can be easily linked or unlinked. | |
24 | * | |
25 | * After unlinking the last vma on the list, we must garbage collect | |
26 | * the anon_vma object itself: we're guaranteed no page can be | |
27 | * pointing to this anon_vma once its vma list is empty. | |
28 | */ | |
29 | struct anon_vma { | |
5a505085 IM |
30 | struct anon_vma *root; /* Root of this anon_vma tree */ |
31 | struct rw_semaphore rwsem; /* W: modification, R: walking the list */ | |
7f60c214 | 32 | /* |
83813267 | 33 | * The refcount is taken on an anon_vma when there is no |
7f60c214 MG |
34 | * guarantee that the vma of page tables will exist for |
35 | * the duration of the operation. A caller that takes | |
36 | * the reference is responsible for clearing up the | |
37 | * anon_vma if they are the last user on release | |
38 | */ | |
83813267 PZ |
39 | atomic_t refcount; |
40 | ||
7a3ef208 KK |
41 | /* |
42 | * Count of child anon_vmas and VMAs which points to this anon_vma. | |
43 | * | |
44 | * This counter is used for making decision about reusing anon_vma | |
45 | * instead of forking new one. See comments in function anon_vma_clone. | |
46 | */ | |
47 | unsigned degree; | |
48 | ||
49 | struct anon_vma *parent; /* Parent of this anon_vma */ | |
50 | ||
7906d00c | 51 | /* |
bf181b9f | 52 | * NOTE: the LSB of the rb_root.rb_node is set by |
7906d00c | 53 | * mm_take_all_locks() _after_ taking the above lock. So the |
bf181b9f | 54 | * rb_root must only be read/written after taking the above lock |
7906d00c AA |
55 | * to be sure to see a valid next pointer. The LSB bit itself |
56 | * is serialized by a system wide lock only visible to | |
57 | * mm_take_all_locks() (mm_all_locks_mutex). | |
58 | */ | |
f808c13f DB |
59 | |
60 | /* Interval tree of private "related" vmas */ | |
61 | struct rb_root_cached rb_root; | |
5beb4930 RR |
62 | }; |
63 | ||
64 | /* | |
65 | * The copy-on-write semantics of fork mean that an anon_vma | |
66 | * can become associated with multiple processes. Furthermore, | |
67 | * each child process will have its own anon_vma, where new | |
68 | * pages for that process are instantiated. | |
69 | * | |
70 | * This structure allows us to find the anon_vmas associated | |
71 | * with a VMA, or the VMAs associated with an anon_vma. | |
72 | * The "same_vma" list contains the anon_vma_chains linking | |
73 | * all the anon_vmas associated with this VMA. | |
bf181b9f | 74 | * The "rb" field indexes on an interval tree the anon_vma_chains |
5beb4930 RR |
75 | * which link all the VMAs associated with this anon_vma. |
76 | */ | |
77 | struct anon_vma_chain { | |
78 | struct vm_area_struct *vma; | |
79 | struct anon_vma *anon_vma; | |
c1e8d7c6 | 80 | struct list_head same_vma; /* locked by mmap_lock & page_table_lock */ |
5a505085 | 81 | struct rb_node rb; /* locked by anon_vma->rwsem */ |
bf181b9f | 82 | unsigned long rb_subtree_last; |
ed8ea815 ML |
83 | #ifdef CONFIG_DEBUG_VM_RB |
84 | unsigned long cached_vma_start, cached_vma_last; | |
85 | #endif | |
1da177e4 LT |
86 | }; |
87 | ||
02c6de8d | 88 | enum ttu_flags { |
a128ca71 SL |
89 | TTU_MIGRATION = 0x1, /* migration mode */ |
90 | TTU_MUNLOCK = 0x2, /* munlock mode */ | |
91 | ||
92 | TTU_SPLIT_HUGE_PMD = 0x4, /* split huge PMD if any */ | |
93 | TTU_IGNORE_MLOCK = 0x8, /* ignore mlock */ | |
732ed558 | 94 | TTU_SYNC = 0x10, /* avoid racy checks with PVMW_SYNC */ |
a128ca71 SL |
95 | TTU_IGNORE_HWPOISON = 0x20, /* corrupted page is recoverable */ |
96 | TTU_BATCH_FLUSH = 0x40, /* Batch TLB flushes where possible | |
72b252ae MG |
97 | * and caller guarantees they will |
98 | * do a final flush if necessary */ | |
b5ff8161 | 99 | TTU_RMAP_LOCKED = 0x80, /* do not grab rmap lock: |
2a52bcbc | 100 | * caller holds it */ |
b5ff8161 | 101 | TTU_SPLIT_FREEZE = 0x100, /* freeze pte under splitting thp */ |
02c6de8d MK |
102 | }; |
103 | ||
1da177e4 | 104 | #ifdef CONFIG_MMU |
76545066 RR |
105 | static inline void get_anon_vma(struct anon_vma *anon_vma) |
106 | { | |
83813267 | 107 | atomic_inc(&anon_vma->refcount); |
76545066 RR |
108 | } |
109 | ||
01d8b20d PZ |
110 | void __put_anon_vma(struct anon_vma *anon_vma); |
111 | ||
112 | static inline void put_anon_vma(struct anon_vma *anon_vma) | |
113 | { | |
114 | if (atomic_dec_and_test(&anon_vma->refcount)) | |
115 | __put_anon_vma(anon_vma); | |
116 | } | |
1da177e4 | 117 | |
4fc3f1d6 | 118 | static inline void anon_vma_lock_write(struct anon_vma *anon_vma) |
cba48b98 | 119 | { |
5a505085 | 120 | down_write(&anon_vma->root->rwsem); |
cba48b98 RR |
121 | } |
122 | ||
08b52706 | 123 | static inline void anon_vma_unlock_write(struct anon_vma *anon_vma) |
cba48b98 | 124 | { |
5a505085 | 125 | up_write(&anon_vma->root->rwsem); |
cba48b98 RR |
126 | } |
127 | ||
4fc3f1d6 IM |
128 | static inline void anon_vma_lock_read(struct anon_vma *anon_vma) |
129 | { | |
130 | down_read(&anon_vma->root->rwsem); | |
131 | } | |
132 | ||
133 | static inline void anon_vma_unlock_read(struct anon_vma *anon_vma) | |
134 | { | |
135 | up_read(&anon_vma->root->rwsem); | |
136 | } | |
137 | ||
138 | ||
1da177e4 LT |
139 | /* |
140 | * anon_vma helper functions. | |
141 | */ | |
142 | void anon_vma_init(void); /* create anon_vma_cachep */ | |
d5a187da | 143 | int __anon_vma_prepare(struct vm_area_struct *); |
5beb4930 RR |
144 | void unlink_anon_vmas(struct vm_area_struct *); |
145 | int anon_vma_clone(struct vm_area_struct *, struct vm_area_struct *); | |
146 | int anon_vma_fork(struct vm_area_struct *, struct vm_area_struct *); | |
1da177e4 | 147 | |
d5a187da VB |
148 | static inline int anon_vma_prepare(struct vm_area_struct *vma) |
149 | { | |
150 | if (likely(vma->anon_vma)) | |
151 | return 0; | |
152 | ||
153 | return __anon_vma_prepare(vma); | |
154 | } | |
155 | ||
5beb4930 RR |
156 | static inline void anon_vma_merge(struct vm_area_struct *vma, |
157 | struct vm_area_struct *next) | |
158 | { | |
81d1b09c | 159 | VM_BUG_ON_VMA(vma->anon_vma != next->anon_vma, vma); |
5beb4930 RR |
160 | unlink_anon_vmas(next); |
161 | } | |
162 | ||
01d8b20d PZ |
163 | struct anon_vma *page_get_anon_vma(struct page *page); |
164 | ||
d281ee61 KS |
165 | /* bitflags for do_page_add_anon_rmap() */ |
166 | #define RMAP_EXCLUSIVE 0x01 | |
167 | #define RMAP_COMPOUND 0x02 | |
168 | ||
1da177e4 LT |
169 | /* |
170 | * rmap interfaces called when adding or removing pte of page | |
171 | */ | |
5a49973d | 172 | void page_move_anon_rmap(struct page *, struct vm_area_struct *); |
d281ee61 KS |
173 | void page_add_anon_rmap(struct page *, struct vm_area_struct *, |
174 | unsigned long, bool); | |
ad8c2ee8 RR |
175 | void do_page_add_anon_rmap(struct page *, struct vm_area_struct *, |
176 | unsigned long, int); | |
d281ee61 KS |
177 | void page_add_new_anon_rmap(struct page *, struct vm_area_struct *, |
178 | unsigned long, bool); | |
dd78fedd | 179 | void page_add_file_rmap(struct page *, bool); |
d281ee61 | 180 | void page_remove_rmap(struct page *, bool); |
1da177e4 | 181 | |
0fe6e20b NH |
182 | void hugepage_add_anon_rmap(struct page *, struct vm_area_struct *, |
183 | unsigned long); | |
184 | void hugepage_add_new_anon_rmap(struct page *, struct vm_area_struct *, | |
185 | unsigned long); | |
186 | ||
53f9263b | 187 | static inline void page_dup_rmap(struct page *page, bool compound) |
1da177e4 | 188 | { |
53f9263b | 189 | atomic_inc(compound ? compound_mapcount_ptr(page) : &page->_mapcount); |
1da177e4 LT |
190 | } |
191 | ||
192 | /* | |
193 | * Called from mm/vmscan.c to handle paging out | |
194 | */ | |
6fe6b7e3 | 195 | int page_referenced(struct page *, int is_locked, |
72835c86 | 196 | struct mem_cgroup *memcg, unsigned long *vm_flags); |
5ad64688 | 197 | |
1fb08ac6 | 198 | void try_to_unmap(struct page *, enum ttu_flags flags); |
1da177e4 | 199 | |
ace71a19 KS |
200 | /* Avoid racy checks */ |
201 | #define PVMW_SYNC (1 << 0) | |
202 | /* Look for migarion entries rather than present PTEs */ | |
203 | #define PVMW_MIGRATION (1 << 1) | |
204 | ||
205 | struct page_vma_mapped_walk { | |
206 | struct page *page; | |
207 | struct vm_area_struct *vma; | |
208 | unsigned long address; | |
209 | pmd_t *pmd; | |
210 | pte_t *pte; | |
211 | spinlock_t *ptl; | |
212 | unsigned int flags; | |
213 | }; | |
214 | ||
215 | static inline void page_vma_mapped_walk_done(struct page_vma_mapped_walk *pvmw) | |
216 | { | |
5d5d19ed ML |
217 | /* HugeTLB pte is set to the relevant page table entry without pte_mapped. */ |
218 | if (pvmw->pte && !PageHuge(pvmw->page)) | |
ace71a19 KS |
219 | pte_unmap(pvmw->pte); |
220 | if (pvmw->ptl) | |
221 | spin_unlock(pvmw->ptl); | |
222 | } | |
223 | ||
224 | bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw); | |
225 | ||
1da177e4 LT |
226 | /* |
227 | * Used by swapoff to help locate where page is expected in vma. | |
228 | */ | |
229 | unsigned long page_address_in_vma(struct page *, struct vm_area_struct *); | |
230 | ||
d08b3851 PZ |
231 | /* |
232 | * Cleans the PTEs of shared mappings. | |
233 | * (and since clean PTEs should also be readonly, write protects them too) | |
234 | * | |
235 | * returns the number of cleaned PTEs. | |
236 | */ | |
237 | int page_mkclean(struct page *); | |
238 | ||
b291f000 NP |
239 | /* |
240 | * called in munlock()/munmap() path to check for other vmas holding | |
241 | * the page mlocked. | |
242 | */ | |
192d7232 | 243 | void try_to_munlock(struct page *); |
b291f000 | 244 | |
e388466d KS |
245 | void remove_migration_ptes(struct page *old, struct page *new, bool locked); |
246 | ||
10be22df AK |
247 | /* |
248 | * Called by memory-failure.c to kill processes. | |
249 | */ | |
4fc3f1d6 IM |
250 | struct anon_vma *page_lock_anon_vma_read(struct page *page); |
251 | void page_unlock_anon_vma_read(struct anon_vma *anon_vma); | |
6a46079c | 252 | int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma); |
10be22df | 253 | |
0dd1c7bb JK |
254 | /* |
255 | * rmap_walk_control: To control rmap traversing for specific needs | |
256 | * | |
257 | * arg: passed to rmap_one() and invalid_vma() | |
258 | * rmap_one: executed on each vma where page is mapped | |
259 | * done: for checking traversing termination condition | |
0dd1c7bb JK |
260 | * anon_lock: for getting anon_lock by optimized way rather than default |
261 | * invalid_vma: for skipping uninterested vma | |
262 | */ | |
051ac83a JK |
263 | struct rmap_walk_control { |
264 | void *arg; | |
e4b82222 MK |
265 | /* |
266 | * Return false if page table scanning in rmap_walk should be stopped. | |
267 | * Otherwise, return true. | |
268 | */ | |
269 | bool (*rmap_one)(struct page *page, struct vm_area_struct *vma, | |
051ac83a | 270 | unsigned long addr, void *arg); |
0dd1c7bb | 271 | int (*done)(struct page *page); |
0dd1c7bb JK |
272 | struct anon_vma *(*anon_lock)(struct page *page); |
273 | bool (*invalid_vma)(struct vm_area_struct *vma, void *arg); | |
051ac83a JK |
274 | }; |
275 | ||
1df631ae MK |
276 | void rmap_walk(struct page *page, struct rmap_walk_control *rwc); |
277 | void rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc); | |
e9995ef9 | 278 | |
1da177e4 LT |
279 | #else /* !CONFIG_MMU */ |
280 | ||
281 | #define anon_vma_init() do {} while (0) | |
282 | #define anon_vma_prepare(vma) (0) | |
283 | #define anon_vma_link(vma) do {} while (0) | |
284 | ||
01ff53f4 | 285 | static inline int page_referenced(struct page *page, int is_locked, |
72835c86 | 286 | struct mem_cgroup *memcg, |
01ff53f4 MF |
287 | unsigned long *vm_flags) |
288 | { | |
289 | *vm_flags = 0; | |
64574746 | 290 | return 0; |
01ff53f4 MF |
291 | } |
292 | ||
666e5a40 | 293 | #define try_to_unmap(page, refs) false |
1da177e4 | 294 | |
d08b3851 PZ |
295 | static inline int page_mkclean(struct page *page) |
296 | { | |
297 | return 0; | |
298 | } | |
299 | ||
300 | ||
1da177e4 LT |
301 | #endif /* CONFIG_MMU */ |
302 | ||
1da177e4 | 303 | #endif /* _LINUX_RMAP_H */ |