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