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 unsigned degree; /* ANDROID: KABI preservation, DO NOT USE! */
46
47 struct anon_vma *parent; /* Parent of this anon_vma */
48
49 /*
50 * NOTE: the LSB of the rb_root.rb_node is set by
51 * mm_take_all_locks() _after_ taking the above lock. So the
52 * rb_root must only be read/written after taking the above lock
53 * to be sure to see a valid next pointer. The LSB bit itself
54 * is serialized by a system wide lock only visible to
55 * mm_take_all_locks() (mm_all_locks_mutex).
56 */
57
58 /* Interval tree of private "related" vmas */
59 struct rb_root_cached rb_root;
60
61 /*
62 * ANDROID: KABI preservation, it's safe to put these at the end of this structure as it's
63 * only passed by a pointer everywhere, the size and internal structures are local to the
64 * core kernel.
65 */
66 #ifndef __GENKSYMS__
67 /*
68 * Count of child anon_vmas. Equals to the count of all anon_vmas that
69 * have ->parent pointing to this one, including itself.
70 *
71 * This counter is used for making decision about reusing anon_vma
72 * instead of forking new one. See comments in function anon_vma_clone.
73 */
74 unsigned long num_children;
75 /* Count of VMAs whose ->anon_vma pointer points to this object. */
76 unsigned long num_active_vmas;
77 #endif
78
79 };
80
81 /*
82 * The copy-on-write semantics of fork mean that an anon_vma
83 * can become associated with multiple processes. Furthermore,
84 * each child process will have its own anon_vma, where new
85 * pages for that process are instantiated.
86 *
87 * This structure allows us to find the anon_vmas associated
88 * with a VMA, or the VMAs associated with an anon_vma.
89 * The "same_vma" list contains the anon_vma_chains linking
90 * all the anon_vmas associated with this VMA.
91 * The "rb" field indexes on an interval tree the anon_vma_chains
92 * which link all the VMAs associated with this anon_vma.
93 */
94 struct anon_vma_chain {
95 struct vm_area_struct *vma;
96 struct anon_vma *anon_vma;
97 struct list_head same_vma; /* locked by mmap_lock & page_table_lock */
98 struct rb_node rb; /* locked by anon_vma->rwsem */
99 unsigned long rb_subtree_last;
100 #ifdef CONFIG_DEBUG_VM_RB
101 unsigned long cached_vma_start, cached_vma_last;
102 #endif
103 };
104
105 enum ttu_flags {
106 TTU_MIGRATION = 0x1, /* migration mode */
107 TTU_MUNLOCK = 0x2, /* munlock mode */
108
109 TTU_SPLIT_HUGE_PMD = 0x4, /* split huge PMD if any */
110 TTU_IGNORE_MLOCK = 0x8, /* ignore mlock */
111 TTU_SYNC = 0x10, /* avoid racy checks with PVMW_SYNC */
112 TTU_IGNORE_HWPOISON = 0x20, /* corrupted page is recoverable */
113 TTU_BATCH_FLUSH = 0x40, /* Batch TLB flushes where possible
114 * and caller guarantees they will
115 * do a final flush if necessary */
116 TTU_RMAP_LOCKED = 0x80, /* do not grab rmap lock:
117 * caller holds it */
118 TTU_SPLIT_FREEZE = 0x100, /* freeze pte under splitting thp */
119 };
120
121 #ifdef CONFIG_MMU
get_anon_vma(struct anon_vma * anon_vma)122 static inline void get_anon_vma(struct anon_vma *anon_vma)
123 {
124 atomic_inc(&anon_vma->refcount);
125 }
126
127 void __put_anon_vma(struct anon_vma *anon_vma);
128
put_anon_vma(struct anon_vma * anon_vma)129 static inline void put_anon_vma(struct anon_vma *anon_vma)
130 {
131 if (atomic_dec_and_test(&anon_vma->refcount))
132 __put_anon_vma(anon_vma);
133 }
134
anon_vma_lock_write(struct anon_vma * anon_vma)135 static inline void anon_vma_lock_write(struct anon_vma *anon_vma)
136 {
137 down_write(&anon_vma->root->rwsem);
138 }
139
anon_vma_unlock_write(struct anon_vma * anon_vma)140 static inline void anon_vma_unlock_write(struct anon_vma *anon_vma)
141 {
142 up_write(&anon_vma->root->rwsem);
143 }
144
anon_vma_lock_read(struct anon_vma * anon_vma)145 static inline void anon_vma_lock_read(struct anon_vma *anon_vma)
146 {
147 down_read(&anon_vma->root->rwsem);
148 }
149
anon_vma_trylock_read(struct anon_vma * anon_vma)150 static inline int anon_vma_trylock_read(struct anon_vma *anon_vma)
151 {
152 return down_read_trylock(&anon_vma->root->rwsem);
153 }
154
anon_vma_unlock_read(struct anon_vma * anon_vma)155 static inline void anon_vma_unlock_read(struct anon_vma *anon_vma)
156 {
157 up_read(&anon_vma->root->rwsem);
158 }
159
160
161 /*
162 * anon_vma helper functions.
163 */
164 void anon_vma_init(void); /* create anon_vma_cachep */
165 int __anon_vma_prepare(struct vm_area_struct *);
166 void unlink_anon_vmas(struct vm_area_struct *);
167 int anon_vma_clone(struct vm_area_struct *, struct vm_area_struct *);
168 int anon_vma_fork(struct vm_area_struct *, struct vm_area_struct *);
169
anon_vma_prepare(struct vm_area_struct * vma)170 static inline int anon_vma_prepare(struct vm_area_struct *vma)
171 {
172 if (likely(vma->anon_vma))
173 return 0;
174
175 return __anon_vma_prepare(vma);
176 }
177
anon_vma_merge(struct vm_area_struct * vma,struct vm_area_struct * next)178 static inline void anon_vma_merge(struct vm_area_struct *vma,
179 struct vm_area_struct *next)
180 {
181 VM_BUG_ON_VMA(vma->anon_vma != next->anon_vma, vma);
182 unlink_anon_vmas(next);
183 }
184
185 struct anon_vma *page_get_anon_vma(struct page *page);
186
187 /* bitflags for do_page_add_anon_rmap() */
188 #define RMAP_EXCLUSIVE 0x01
189 #define RMAP_COMPOUND 0x02
190
191 /*
192 * rmap interfaces called when adding or removing pte of page
193 */
194 void page_move_anon_rmap(struct page *, struct vm_area_struct *);
195 void page_add_anon_rmap(struct page *, struct vm_area_struct *,
196 unsigned long, bool);
197 void do_page_add_anon_rmap(struct page *, struct vm_area_struct *,
198 unsigned long, int);
199 void __page_add_new_anon_rmap(struct page *page, struct vm_area_struct *vma,
200 unsigned long address, bool compound);
page_add_new_anon_rmap(struct page * page,struct vm_area_struct * vma,unsigned long address,bool compound)201 static inline void page_add_new_anon_rmap(struct page *page,
202 struct vm_area_struct *vma,
203 unsigned long address, bool compound)
204 {
205 VM_BUG_ON_VMA(address < vma->vm_start || address >= vma->vm_end, vma);
206 __page_add_new_anon_rmap(page, vma, address, compound);
207 }
208
209 void page_add_file_rmap(struct page *, bool);
210 void page_remove_rmap(struct page *, bool);
211
212 void hugepage_add_anon_rmap(struct page *, struct vm_area_struct *,
213 unsigned long);
214 void hugepage_add_new_anon_rmap(struct page *, struct vm_area_struct *,
215 unsigned long);
216
page_dup_rmap(struct page * page,bool compound)217 static inline void page_dup_rmap(struct page *page, bool compound)
218 {
219 bool success = false;
220
221 if (!compound)
222 trace_android_vh_update_page_mapcount(page, true, compound, NULL, &success);
223 if (!success)
224 atomic_inc(compound ? compound_mapcount_ptr(page) : &page->_mapcount);
225 }
226
227 /*
228 * Called from mm/vmscan.c to handle paging out
229 */
230 int page_referenced(struct page *, int is_locked,
231 struct mem_cgroup *memcg, unsigned long *vm_flags);
232
233 bool try_to_unmap(struct page *, enum ttu_flags flags);
234
235 /* Avoid racy checks */
236 #define PVMW_SYNC (1 << 0)
237 /* Look for migarion entries rather than present PTEs */
238 #define PVMW_MIGRATION (1 << 1)
239
240 struct page_vma_mapped_walk {
241 struct page *page;
242 struct vm_area_struct *vma;
243 unsigned long address;
244 pmd_t *pmd;
245 pte_t *pte;
246 spinlock_t *ptl;
247 unsigned int flags;
248 };
249
page_vma_mapped_walk_done(struct page_vma_mapped_walk * pvmw)250 static inline void page_vma_mapped_walk_done(struct page_vma_mapped_walk *pvmw)
251 {
252 /* HugeTLB pte is set to the relevant page table entry without pte_mapped. */
253 if (pvmw->pte && !PageHuge(pvmw->page))
254 pte_unmap(pvmw->pte);
255 if (pvmw->ptl)
256 spin_unlock(pvmw->ptl);
257 }
258
259 bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw);
260
261 /*
262 * Used by swapoff to help locate where page is expected in vma.
263 */
264 unsigned long page_address_in_vma(struct page *, struct vm_area_struct *);
265
266 /*
267 * Cleans the PTEs of shared mappings.
268 * (and since clean PTEs should also be readonly, write protects them too)
269 *
270 * returns the number of cleaned PTEs.
271 */
272 int page_mkclean(struct page *);
273
274 /*
275 * called in munlock()/munmap() path to check for other vmas holding
276 * the page mlocked.
277 */
278 void try_to_munlock(struct page *);
279
280 void remove_migration_ptes(struct page *old, struct page *new, bool locked);
281
282 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma);
283
284 /*
285 * rmap_walk_control: To control rmap traversing for specific needs
286 *
287 * arg: passed to rmap_one() and invalid_vma()
288 * try_lock: bail out if the rmap lock is contended
289 * contended: indicate the rmap traversal bailed out due to lock contention
290 * rmap_one: executed on each vma where page is mapped
291 * done: for checking traversing termination condition
292 * anon_lock: for getting anon_lock by optimized way rather than default
293 * invalid_vma: for skipping uninterested vma
294 */
295 struct rmap_walk_control {
296 void *arg;
297 bool try_lock;
298 bool contended;
299 /*
300 * Return false if page table scanning in rmap_walk should be stopped.
301 * Otherwise, return true.
302 */
303 bool (*rmap_one)(struct page *page, struct vm_area_struct *vma,
304 unsigned long addr, void *arg);
305 int (*done)(struct page *page);
306 struct anon_vma *(*anon_lock)(struct page *page,
307 struct rmap_walk_control *rwc);
308 bool (*invalid_vma)(struct vm_area_struct *vma, void *arg);
309 };
310
311 void rmap_walk(struct page *page, struct rmap_walk_control *rwc);
312 void rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc);
313
314 /*
315 * Called by memory-failure.c to kill processes.
316 */
317 struct anon_vma *page_lock_anon_vma_read(struct page *page,
318 struct rmap_walk_control *rwc);
319 void page_unlock_anon_vma_read(struct anon_vma *anon_vma);
320
321 #else /* !CONFIG_MMU */
322
323 #define anon_vma_init() do {} while (0)
324 #define anon_vma_prepare(vma) (0)
325 #define anon_vma_link(vma) do {} while (0)
326
page_referenced(struct page * page,int is_locked,struct mem_cgroup * memcg,unsigned long * vm_flags)327 static inline int page_referenced(struct page *page, int is_locked,
328 struct mem_cgroup *memcg,
329 unsigned long *vm_flags)
330 {
331 *vm_flags = 0;
332 return 0;
333 }
334
335 #define try_to_unmap(page, refs) false
336
page_mkclean(struct page * page)337 static inline int page_mkclean(struct page *page)
338 {
339 return 0;
340 }
341
342
343 #endif /* CONFIG_MMU */
344
345 #endif /* _LINUX_RMAP_H */
346