1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _LINUX_RMAP_H
3 #define _LINUX_RMAP_H
4 /*
5 * Declarations for Reverse Mapping functions in mm/rmap.c
6 */
7
8 #include <linux/list.h>
9 #include <linux/slab.h>
10 #include <linux/mm.h>
11 #include <linux/rwsem.h>
12 #include <linux/memcontrol.h>
13 #include <linux/highmem.h>
14 #ifndef __GENKSYMS__
15 #define PROTECT_TRACE_INCLUDE_PATH
16 #include <trace/hooks/mm.h>
17 #endif
18
19 /*
20 * The anon_vma heads a list of private "related" vmas, to scan if
21 * an anonymous page pointing to this anon_vma needs to be unmapped:
22 * the vmas on the list will be related by forking, or by splitting.
23 *
24 * Since vmas come and go as they are split and merged (particularly
25 * in mprotect), the mapping field of an anonymous page cannot point
26 * directly to a vma: instead it points to an anon_vma, on whose list
27 * the related vmas can be easily linked or unlinked.
28 *
29 * After unlinking the last vma on the list, we must garbage collect
30 * the anon_vma object itself: we're guaranteed no page can be
31 * pointing to this anon_vma once its vma list is empty.
32 */
33 struct anon_vma {
34 struct anon_vma *root; /* Root of this anon_vma tree */
35 struct rw_semaphore rwsem; /* W: modification, R: walking the list */
36 /*
37 * The refcount is taken on an anon_vma when there is no
38 * guarantee that the vma of page tables will exist for
39 * the duration of the operation. A caller that takes
40 * the reference is responsible for clearing up the
41 * anon_vma if they are the last user on release
42 */
43 atomic_t refcount;
44
45 /*
46 * Count of child anon_vmas and VMAs which points to this anon_vma.
47 *
48 * This counter is used for making decision about reusing anon_vma
49 * instead of forking new one. See comments in function anon_vma_clone.
50 */
51 unsigned degree;
52
53 struct anon_vma *parent; /* Parent of this anon_vma */
54
55 /*
56 * NOTE: the LSB of the rb_root.rb_node is set by
57 * mm_take_all_locks() _after_ taking the above lock. So the
58 * rb_root must only be read/written after taking the above lock
59 * to be sure to see a valid next pointer. The LSB bit itself
60 * is serialized by a system wide lock only visible to
61 * mm_take_all_locks() (mm_all_locks_mutex).
62 */
63
64 /* Interval tree of private "related" vmas */
65 struct rb_root_cached rb_root;
66 };
67
68 /*
69 * The copy-on-write semantics of fork mean that an anon_vma
70 * can become associated with multiple processes. Furthermore,
71 * each child process will have its own anon_vma, where new
72 * pages for that process are instantiated.
73 *
74 * This structure allows us to find the anon_vmas associated
75 * with a VMA, or the VMAs associated with an anon_vma.
76 * The "same_vma" list contains the anon_vma_chains linking
77 * all the anon_vmas associated with this VMA.
78 * The "rb" field indexes on an interval tree the anon_vma_chains
79 * which link all the VMAs associated with this anon_vma.
80 */
81 struct anon_vma_chain {
82 struct vm_area_struct *vma;
83 struct anon_vma *anon_vma;
84 struct list_head same_vma; /* locked by mmap_lock & page_table_lock */
85 struct rb_node rb; /* locked by anon_vma->rwsem */
86 unsigned long rb_subtree_last;
87 #ifdef CONFIG_DEBUG_VM_RB
88 unsigned long cached_vma_start, cached_vma_last;
89 #endif
90 };
91
92 enum ttu_flags {
93 TTU_MIGRATION = 0x1, /* migration mode */
94 TTU_MUNLOCK = 0x2, /* munlock mode */
95
96 TTU_SPLIT_HUGE_PMD = 0x4, /* split huge PMD if any */
97 TTU_IGNORE_MLOCK = 0x8, /* ignore mlock */
98 TTU_SYNC = 0x10, /* avoid racy checks with PVMW_SYNC */
99 TTU_IGNORE_HWPOISON = 0x20, /* corrupted page is recoverable */
100 TTU_BATCH_FLUSH = 0x40, /* Batch TLB flushes where possible
101 * and caller guarantees they will
102 * do a final flush if necessary */
103 TTU_RMAP_LOCKED = 0x80, /* do not grab rmap lock:
104 * caller holds it */
105 TTU_SPLIT_FREEZE = 0x100, /* freeze pte under splitting thp */
106 };
107
108 #ifdef CONFIG_MMU
get_anon_vma(struct anon_vma * anon_vma)109 static inline void get_anon_vma(struct anon_vma *anon_vma)
110 {
111 atomic_inc(&anon_vma->refcount);
112 }
113
114 void __put_anon_vma(struct anon_vma *anon_vma);
115
put_anon_vma(struct anon_vma * anon_vma)116 static inline void put_anon_vma(struct anon_vma *anon_vma)
117 {
118 if (atomic_dec_and_test(&anon_vma->refcount))
119 __put_anon_vma(anon_vma);
120 }
121
anon_vma_lock_write(struct anon_vma * anon_vma)122 static inline void anon_vma_lock_write(struct anon_vma *anon_vma)
123 {
124 down_write(&anon_vma->root->rwsem);
125 }
126
anon_vma_unlock_write(struct anon_vma * anon_vma)127 static inline void anon_vma_unlock_write(struct anon_vma *anon_vma)
128 {
129 up_write(&anon_vma->root->rwsem);
130 }
131
anon_vma_lock_read(struct anon_vma * anon_vma)132 static inline void anon_vma_lock_read(struct anon_vma *anon_vma)
133 {
134 down_read(&anon_vma->root->rwsem);
135 }
136
anon_vma_trylock_read(struct anon_vma * anon_vma)137 static inline int anon_vma_trylock_read(struct anon_vma *anon_vma)
138 {
139 return down_read_trylock(&anon_vma->root->rwsem);
140 }
141
anon_vma_unlock_read(struct anon_vma * anon_vma)142 static inline void anon_vma_unlock_read(struct anon_vma *anon_vma)
143 {
144 up_read(&anon_vma->root->rwsem);
145 }
146
147
148 /*
149 * anon_vma helper functions.
150 */
151 void anon_vma_init(void); /* create anon_vma_cachep */
152 int __anon_vma_prepare(struct vm_area_struct *);
153 void unlink_anon_vmas(struct vm_area_struct *);
154 int anon_vma_clone(struct vm_area_struct *, struct vm_area_struct *);
155 int anon_vma_fork(struct vm_area_struct *, struct vm_area_struct *);
156
anon_vma_prepare(struct vm_area_struct * vma)157 static inline int anon_vma_prepare(struct vm_area_struct *vma)
158 {
159 if (likely(vma->anon_vma))
160 return 0;
161
162 return __anon_vma_prepare(vma);
163 }
164
anon_vma_merge(struct vm_area_struct * vma,struct vm_area_struct * next)165 static inline void anon_vma_merge(struct vm_area_struct *vma,
166 struct vm_area_struct *next)
167 {
168 VM_BUG_ON_VMA(vma->anon_vma != next->anon_vma, vma);
169 unlink_anon_vmas(next);
170 }
171
172 struct anon_vma *page_get_anon_vma(struct page *page);
173
174 /* bitflags for do_page_add_anon_rmap() */
175 #define RMAP_EXCLUSIVE 0x01
176 #define RMAP_COMPOUND 0x02
177
178 /*
179 * rmap interfaces called when adding or removing pte of page
180 */
181 void page_move_anon_rmap(struct page *, struct vm_area_struct *);
182 void page_add_anon_rmap(struct page *, struct vm_area_struct *,
183 unsigned long, bool);
184 void do_page_add_anon_rmap(struct page *, struct vm_area_struct *,
185 unsigned long, int);
186 void __page_add_new_anon_rmap(struct page *page, struct vm_area_struct *vma,
187 unsigned long address, bool compound);
page_add_new_anon_rmap(struct page * page,struct vm_area_struct * vma,unsigned long address,bool compound)188 static inline void page_add_new_anon_rmap(struct page *page,
189 struct vm_area_struct *vma,
190 unsigned long address, bool compound)
191 {
192 VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
193 __page_add_new_anon_rmap(page, vma, address, compound);
194 }
195
196 void page_add_file_rmap(struct page *, bool);
197 void page_remove_rmap(struct page *, bool);
198
199 void hugepage_add_anon_rmap(struct page *, struct vm_area_struct *,
200 unsigned long);
201 void hugepage_add_new_anon_rmap(struct page *, struct vm_area_struct *,
202 unsigned long);
203
page_dup_rmap(struct page * page,bool compound)204 static inline void page_dup_rmap(struct page *page, bool compound)
205 {
206 bool success = false;
207
208 if (!compound)
209 trace_android_vh_update_page_mapcount(page, true, compound, NULL, &success);
210 if (!success)
211 atomic_inc(compound ? compound_mapcount_ptr(page) : &page->_mapcount);
212 }
213
214 /*
215 * Called from mm/vmscan.c to handle paging out
216 */
217 int page_referenced(struct page *, int is_locked,
218 struct mem_cgroup *memcg, unsigned long *vm_flags);
219
220 bool try_to_unmap(struct page *, enum ttu_flags flags);
221
222 /* Avoid racy checks */
223 #define PVMW_SYNC (1 << 0)
224 /* Look for migarion entries rather than present PTEs */
225 #define PVMW_MIGRATION (1 << 1)
226
227 struct page_vma_mapped_walk {
228 struct page *page;
229 struct vm_area_struct *vma;
230 unsigned long address;
231 pmd_t *pmd;
232 pte_t *pte;
233 spinlock_t *ptl;
234 unsigned int flags;
235 };
236
page_vma_mapped_walk_done(struct page_vma_mapped_walk * pvmw)237 static inline void page_vma_mapped_walk_done(struct page_vma_mapped_walk *pvmw)
238 {
239 /* HugeTLB pte is set to the relevant page table entry without pte_mapped. */
240 if (pvmw->pte && !PageHuge(pvmw->page))
241 pte_unmap(pvmw->pte);
242 if (pvmw->ptl)
243 spin_unlock(pvmw->ptl);
244 }
245
246 bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw);
247
248 /*
249 * Used by swapoff to help locate where page is expected in vma.
250 */
251 unsigned long page_address_in_vma(struct page *, struct vm_area_struct *);
252
253 /*
254 * Cleans the PTEs of shared mappings.
255 * (and since clean PTEs should also be readonly, write protects them too)
256 *
257 * returns the number of cleaned PTEs.
258 */
259 int page_mkclean(struct page *);
260
261 /*
262 * called in munlock()/munmap() path to check for other vmas holding
263 * the page mlocked.
264 */
265 void try_to_munlock(struct page *);
266
267 void remove_migration_ptes(struct page *old, struct page *new, bool locked);
268
269 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma);
270
271 /*
272 * rmap_walk_control: To control rmap traversing for specific needs
273 *
274 * arg: passed to rmap_one() and invalid_vma()
275 * try_lock: bail out if the rmap lock is contended
276 * contended: indicate the rmap traversal bailed out due to lock contention
277 * rmap_one: executed on each vma where page is mapped
278 * done: for checking traversing termination condition
279 * anon_lock: for getting anon_lock by optimized way rather than default
280 * invalid_vma: for skipping uninterested vma
281 */
282 struct rmap_walk_control {
283 void *arg;
284 bool try_lock;
285 bool contended;
286 /*
287 * Return false if page table scanning in rmap_walk should be stopped.
288 * Otherwise, return true.
289 */
290 bool (*rmap_one)(struct page *page, struct vm_area_struct *vma,
291 unsigned long addr, void *arg);
292 int (*done)(struct page *page);
293 struct anon_vma *(*anon_lock)(struct page *page,
294 struct rmap_walk_control *rwc);
295 bool (*invalid_vma)(struct vm_area_struct *vma, void *arg);
296 };
297
298 void rmap_walk(struct page *page, struct rmap_walk_control *rwc);
299 void rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc);
300
301 /*
302 * Called by memory-failure.c to kill processes.
303 */
304 struct anon_vma *page_lock_anon_vma_read(struct page *page,
305 struct rmap_walk_control *rwc);
306 void page_unlock_anon_vma_read(struct anon_vma *anon_vma);
307
308 #else /* !CONFIG_MMU */
309
310 #define anon_vma_init() do {} while (0)
311 #define anon_vma_prepare(vma) (0)
312 #define anon_vma_link(vma) do {} while (0)
313
page_referenced(struct page * page,int is_locked,struct mem_cgroup * memcg,unsigned long * vm_flags)314 static inline int page_referenced(struct page *page, int is_locked,
315 struct mem_cgroup *memcg,
316 unsigned long *vm_flags)
317 {
318 *vm_flags = 0;
319 return 0;
320 }
321
322 #define try_to_unmap(page, refs) false
323
page_mkclean(struct page * page)324 static inline int page_mkclean(struct page *page)
325 {
326 return 0;
327 }
328
329
330 #endif /* CONFIG_MMU */
331
332 #endif /* _LINUX_RMAP_H */
333