xref: /linux-6.15/include/linux/rmap.h (revision 91feabc2)
1 #ifndef _LINUX_RMAP_H
2 #define _LINUX_RMAP_H
3 /*
4  * Declarations for Reverse Mapping functions in mm/rmap.c
5  */
6 
7 #include <linux/list.h>
8 #include <linux/slab.h>
9 #include <linux/mm.h>
10 #include <linux/rwsem.h>
11 #include <linux/memcontrol.h>
12 
13 /*
14  * The anon_vma heads a list of private "related" vmas, to scan if
15  * an anonymous page pointing to this anon_vma needs to be unmapped:
16  * the vmas on the list will be related by forking, or by splitting.
17  *
18  * Since vmas come and go as they are split and merged (particularly
19  * in mprotect), the mapping field of an anonymous page cannot point
20  * directly to a vma: instead it points to an anon_vma, on whose list
21  * the related vmas can be easily linked or unlinked.
22  *
23  * After unlinking the last vma on the list, we must garbage collect
24  * the anon_vma object itself: we're guaranteed no page can be
25  * pointing to this anon_vma once its vma list is empty.
26  */
27 struct anon_vma {
28 	struct anon_vma *root;		/* Root of this anon_vma tree */
29 	struct rw_semaphore rwsem;	/* W: modification, R: walking the list */
30 	/*
31 	 * The refcount is taken on an anon_vma when there is no
32 	 * guarantee that the vma of page tables will exist for
33 	 * the duration of the operation. A caller that takes
34 	 * the reference is responsible for clearing up the
35 	 * anon_vma if they are the last user on release
36 	 */
37 	atomic_t refcount;
38 
39 	/*
40 	 * Count of child anon_vmas and VMAs which points to this anon_vma.
41 	 *
42 	 * This counter is used for making decision about reusing anon_vma
43 	 * instead of forking new one. See comments in function anon_vma_clone.
44 	 */
45 	unsigned degree;
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 	struct rb_root rb_root;	/* Interval tree of private "related" vmas */
58 };
59 
60 /*
61  * The copy-on-write semantics of fork mean that an anon_vma
62  * can become associated with multiple processes. Furthermore,
63  * each child process will have its own anon_vma, where new
64  * pages for that process are instantiated.
65  *
66  * This structure allows us to find the anon_vmas associated
67  * with a VMA, or the VMAs associated with an anon_vma.
68  * The "same_vma" list contains the anon_vma_chains linking
69  * all the anon_vmas associated with this VMA.
70  * The "rb" field indexes on an interval tree the anon_vma_chains
71  * which link all the VMAs associated with this anon_vma.
72  */
73 struct anon_vma_chain {
74 	struct vm_area_struct *vma;
75 	struct anon_vma *anon_vma;
76 	struct list_head same_vma;   /* locked by mmap_sem & page_table_lock */
77 	struct rb_node rb;			/* locked by anon_vma->rwsem */
78 	unsigned long rb_subtree_last;
79 #ifdef CONFIG_DEBUG_VM_RB
80 	unsigned long cached_vma_start, cached_vma_last;
81 #endif
82 };
83 
84 enum ttu_flags {
85 	TTU_UNMAP = 1,			/* unmap mode */
86 	TTU_MIGRATION = 2,		/* migration mode */
87 	TTU_MUNLOCK = 4,		/* munlock mode */
88 	TTU_LZFREE = 8,			/* lazy free mode */
89 
90 	TTU_IGNORE_MLOCK = (1 << 8),	/* ignore mlock */
91 	TTU_IGNORE_ACCESS = (1 << 9),	/* don't age */
92 	TTU_IGNORE_HWPOISON = (1 << 10),/* corrupted page is recoverable */
93 	TTU_BATCH_FLUSH = (1 << 11),	/* Batch TLB flushes where possible
94 					 * and caller guarantees they will
95 					 * do a final flush if necessary */
96 };
97 
98 #ifdef CONFIG_MMU
99 static inline void get_anon_vma(struct anon_vma *anon_vma)
100 {
101 	atomic_inc(&anon_vma->refcount);
102 }
103 
104 void __put_anon_vma(struct anon_vma *anon_vma);
105 
106 static inline void put_anon_vma(struct anon_vma *anon_vma)
107 {
108 	if (atomic_dec_and_test(&anon_vma->refcount))
109 		__put_anon_vma(anon_vma);
110 }
111 
112 static inline void vma_lock_anon_vma(struct vm_area_struct *vma)
113 {
114 	struct anon_vma *anon_vma = vma->anon_vma;
115 	if (anon_vma)
116 		down_write(&anon_vma->root->rwsem);
117 }
118 
119 static inline void vma_unlock_anon_vma(struct vm_area_struct *vma)
120 {
121 	struct anon_vma *anon_vma = vma->anon_vma;
122 	if (anon_vma)
123 		up_write(&anon_vma->root->rwsem);
124 }
125 
126 static inline void anon_vma_lock_write(struct anon_vma *anon_vma)
127 {
128 	down_write(&anon_vma->root->rwsem);
129 }
130 
131 static inline void anon_vma_unlock_write(struct anon_vma *anon_vma)
132 {
133 	up_write(&anon_vma->root->rwsem);
134 }
135 
136 static inline void anon_vma_lock_read(struct anon_vma *anon_vma)
137 {
138 	down_read(&anon_vma->root->rwsem);
139 }
140 
141 static inline void anon_vma_unlock_read(struct anon_vma *anon_vma)
142 {
143 	up_read(&anon_vma->root->rwsem);
144 }
145 
146 
147 /*
148  * anon_vma helper functions.
149  */
150 void anon_vma_init(void);	/* create anon_vma_cachep */
151 int  anon_vma_prepare(struct vm_area_struct *);
152 void unlink_anon_vmas(struct vm_area_struct *);
153 int anon_vma_clone(struct vm_area_struct *, struct vm_area_struct *);
154 int anon_vma_fork(struct vm_area_struct *, struct vm_area_struct *);
155 
156 static inline void anon_vma_merge(struct vm_area_struct *vma,
157 				  struct vm_area_struct *next)
158 {
159 	VM_BUG_ON_VMA(vma->anon_vma != next->anon_vma, vma);
160 	unlink_anon_vmas(next);
161 }
162 
163 struct anon_vma *page_get_anon_vma(struct page *page);
164 
165 /* bitflags for do_page_add_anon_rmap() */
166 #define RMAP_EXCLUSIVE 0x01
167 #define RMAP_COMPOUND 0x02
168 
169 /*
170  * rmap interfaces called when adding or removing pte of page
171  */
172 void page_move_anon_rmap(struct page *, struct vm_area_struct *, unsigned long);
173 void page_add_anon_rmap(struct page *, struct vm_area_struct *,
174 		unsigned long, bool);
175 void do_page_add_anon_rmap(struct page *, struct vm_area_struct *,
176 			   unsigned long, int);
177 void page_add_new_anon_rmap(struct page *, struct vm_area_struct *,
178 		unsigned long, bool);
179 void page_add_file_rmap(struct page *);
180 void page_remove_rmap(struct page *, bool);
181 
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 
187 static inline void page_dup_rmap(struct page *page, bool compound)
188 {
189 	atomic_inc(compound ? compound_mapcount_ptr(page) : &page->_mapcount);
190 }
191 
192 /*
193  * Called from mm/vmscan.c to handle paging out
194  */
195 int page_referenced(struct page *, int is_locked,
196 			struct mem_cgroup *memcg, unsigned long *vm_flags);
197 
198 #define TTU_ACTION(x) ((x) & TTU_ACTION_MASK)
199 
200 int try_to_unmap(struct page *, enum ttu_flags flags);
201 
202 /*
203  * Used by uprobes to replace a userspace page safely
204  */
205 pte_t *__page_check_address(struct page *, struct mm_struct *,
206 				unsigned long, spinlock_t **, int);
207 
208 static inline pte_t *page_check_address(struct page *page, struct mm_struct *mm,
209 					unsigned long address,
210 					spinlock_t **ptlp, int sync)
211 {
212 	pte_t *ptep;
213 
214 	__cond_lock(*ptlp, ptep = __page_check_address(page, mm, address,
215 						       ptlp, sync));
216 	return ptep;
217 }
218 
219 /*
220  * Used by idle page tracking to check if a page was referenced via page
221  * tables.
222  */
223 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
224 bool page_check_address_transhuge(struct page *page, struct mm_struct *mm,
225 				  unsigned long address, pmd_t **pmdp,
226 				  pte_t **ptep, spinlock_t **ptlp);
227 #else
228 static inline bool page_check_address_transhuge(struct page *page,
229 				struct mm_struct *mm, unsigned long address,
230 				pmd_t **pmdp, pte_t **ptep, spinlock_t **ptlp)
231 {
232 	*ptep = page_check_address(page, mm, address, ptlp, 0);
233 	*pmdp = NULL;
234 	return !!*ptep;
235 }
236 #endif
237 
238 /*
239  * Used by swapoff to help locate where page is expected in vma.
240  */
241 unsigned long page_address_in_vma(struct page *, struct vm_area_struct *);
242 
243 /*
244  * Cleans the PTEs of shared mappings.
245  * (and since clean PTEs should also be readonly, write protects them too)
246  *
247  * returns the number of cleaned PTEs.
248  */
249 int page_mkclean(struct page *);
250 
251 /*
252  * called in munlock()/munmap() path to check for other vmas holding
253  * the page mlocked.
254  */
255 int try_to_munlock(struct page *);
256 
257 /*
258  * Called by memory-failure.c to kill processes.
259  */
260 struct anon_vma *page_lock_anon_vma_read(struct page *page);
261 void page_unlock_anon_vma_read(struct anon_vma *anon_vma);
262 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma);
263 
264 /*
265  * rmap_walk_control: To control rmap traversing for specific needs
266  *
267  * arg: passed to rmap_one() and invalid_vma()
268  * rmap_one: executed on each vma where page is mapped
269  * done: for checking traversing termination condition
270  * anon_lock: for getting anon_lock by optimized way rather than default
271  * invalid_vma: for skipping uninterested vma
272  */
273 struct rmap_walk_control {
274 	void *arg;
275 	int (*rmap_one)(struct page *page, struct vm_area_struct *vma,
276 					unsigned long addr, void *arg);
277 	int (*done)(struct page *page);
278 	struct anon_vma *(*anon_lock)(struct page *page);
279 	bool (*invalid_vma)(struct vm_area_struct *vma, void *arg);
280 };
281 
282 int rmap_walk(struct page *page, struct rmap_walk_control *rwc);
283 
284 #else	/* !CONFIG_MMU */
285 
286 #define anon_vma_init()		do {} while (0)
287 #define anon_vma_prepare(vma)	(0)
288 #define anon_vma_link(vma)	do {} while (0)
289 
290 static inline int page_referenced(struct page *page, int is_locked,
291 				  struct mem_cgroup *memcg,
292 				  unsigned long *vm_flags)
293 {
294 	*vm_flags = 0;
295 	return 0;
296 }
297 
298 #define try_to_unmap(page, refs) SWAP_FAIL
299 
300 static inline int page_mkclean(struct page *page)
301 {
302 	return 0;
303 }
304 
305 
306 #endif	/* CONFIG_MMU */
307 
308 /*
309  * Return values of try_to_unmap
310  */
311 #define SWAP_SUCCESS	0
312 #define SWAP_AGAIN	1
313 #define SWAP_FAIL	2
314 #define SWAP_MLOCK	3
315 #define SWAP_LZFREE	4
316 
317 #endif	/* _LINUX_RMAP_H */
318