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 */
29struct 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 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
51 /*
52 * NOTE: the LSB of the rb_root.rb_node is set by
53 * mm_take_all_locks() _after_ taking the above lock. So the
54 * rb_root must only be read/written after taking the above lock
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 */
59
60 /* Interval tree of private "related" vmas */
61 struct rb_root_cached rb_root;
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.
74 * The "rb" field indexes on an interval tree the anon_vma_chains
75 * which link all the VMAs associated with this anon_vma.
76 */
77struct anon_vma_chain {
78 struct vm_area_struct *vma;
79 struct anon_vma *anon_vma;
80 struct list_head same_vma; /* locked by mmap_sem & page_table_lock */
81 struct rb_node rb; /* locked by anon_vma->rwsem */
82 unsigned long rb_subtree_last;
83#ifdef CONFIG_DEBUG_VM_RB
84 unsigned long cached_vma_start, cached_vma_last;
85#endif
86};
87
88enum ttu_flags {
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 */
94 TTU_IGNORE_ACCESS = 0x10, /* don't age */
95 TTU_IGNORE_HWPOISON = 0x20, /* corrupted page is recoverable */
96 TTU_BATCH_FLUSH = 0x40, /* Batch TLB flushes where possible
97 * and caller guarantees they will
98 * do a final flush if necessary */
99 TTU_RMAP_LOCKED = 0x80, /* do not grab rmap lock:
100 * caller holds it */
101 TTU_SPLIT_FREEZE = 0x100, /* freeze pte under splitting thp */
102};
103
104#ifdef CONFIG_MMU
105static inline void get_anon_vma(struct anon_vma *anon_vma)
106{
107 atomic_inc(&anon_vma->refcount);
108}
109
110void __put_anon_vma(struct anon_vma *anon_vma);
111
112static 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}
117
118static inline void anon_vma_lock_write(struct anon_vma *anon_vma)
119{
120 down_write(&anon_vma->root->rwsem);
121}
122
123static inline void anon_vma_unlock_write(struct anon_vma *anon_vma)
124{
125 up_write(&anon_vma->root->rwsem);
126}
127
128static inline void anon_vma_lock_read(struct anon_vma *anon_vma)
129{
130 down_read(&anon_vma->root->rwsem);
131}
132
133static inline void anon_vma_unlock_read(struct anon_vma *anon_vma)
134{
135 up_read(&anon_vma->root->rwsem);
136}
137
138
139/*
140 * anon_vma helper functions.
141 */
142void anon_vma_init(void); /* create anon_vma_cachep */
143int __anon_vma_prepare(struct vm_area_struct *);
144void unlink_anon_vmas(struct vm_area_struct *);
145int anon_vma_clone(struct vm_area_struct *, struct vm_area_struct *);
146int anon_vma_fork(struct vm_area_struct *, struct vm_area_struct *);
147
148static 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
156static 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
163struct 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 */
172void page_move_anon_rmap(struct page *, struct vm_area_struct *);
173void page_add_anon_rmap(struct page *, struct vm_area_struct *,
174 unsigned long, bool);
175void do_page_add_anon_rmap(struct page *, struct vm_area_struct *,
176 unsigned long, int);
177void page_add_new_anon_rmap(struct page *, struct vm_area_struct *,
178 unsigned long, bool);
179void page_add_file_rmap(struct page *, bool);
180void page_remove_rmap(struct page *, bool);
181
182void hugepage_add_anon_rmap(struct page *, struct vm_area_struct *,
183 unsigned long);
184void hugepage_add_new_anon_rmap(struct page *, struct vm_area_struct *,
185 unsigned long);
186
187static 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 */
195int page_referenced(struct page *, int is_locked,
196 struct mem_cgroup *memcg, unsigned long *vm_flags);
197
198bool try_to_unmap(struct page *, enum ttu_flags flags);
199
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
205struct 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
215static inline void page_vma_mapped_walk_done(struct page_vma_mapped_walk *pvmw)
216{
217 if (pvmw->pte)
218 pte_unmap(pvmw->pte);
219 if (pvmw->ptl)
220 spin_unlock(pvmw->ptl);
221}
222
223bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw);
224
225/*
226 * Used by swapoff to help locate where page is expected in vma.
227 */
228unsigned long page_address_in_vma(struct page *, struct vm_area_struct *);
229
230/*
231 * Cleans the PTEs of shared mappings.
232 * (and since clean PTEs should also be readonly, write protects them too)
233 *
234 * returns the number of cleaned PTEs.
235 */
236int page_mkclean(struct page *);
237
238/*
239 * called in munlock()/munmap() path to check for other vmas holding
240 * the page mlocked.
241 */
242void try_to_munlock(struct page *);
243
244void remove_migration_ptes(struct page *old, struct page *new, bool locked);
245
246/*
247 * Called by memory-failure.c to kill processes.
248 */
249struct anon_vma *page_lock_anon_vma_read(struct page *page);
250void page_unlock_anon_vma_read(struct anon_vma *anon_vma);
251int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma);
252
253/*
254 * rmap_walk_control: To control rmap traversing for specific needs
255 *
256 * arg: passed to rmap_one() and invalid_vma()
257 * rmap_one: executed on each vma where page is mapped
258 * done: for checking traversing termination condition
259 * anon_lock: for getting anon_lock by optimized way rather than default
260 * invalid_vma: for skipping uninterested vma
261 */
262struct rmap_walk_control {
263 void *arg;
264 /*
265 * Return false if page table scanning in rmap_walk should be stopped.
266 * Otherwise, return true.
267 */
268 bool (*rmap_one)(struct page *page, struct vm_area_struct *vma,
269 unsigned long addr, void *arg);
270 int (*done)(struct page *page);
271 struct anon_vma *(*anon_lock)(struct page *page);
272 bool (*invalid_vma)(struct vm_area_struct *vma, void *arg);
273};
274
275void rmap_walk(struct page *page, struct rmap_walk_control *rwc);
276void rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc);
277
278#else /* !CONFIG_MMU */
279
280#define anon_vma_init() do {} while (0)
281#define anon_vma_prepare(vma) (0)
282#define anon_vma_link(vma) do {} while (0)
283
284static inline int page_referenced(struct page *page, int is_locked,
285 struct mem_cgroup *memcg,
286 unsigned long *vm_flags)
287{
288 *vm_flags = 0;
289 return 0;
290}
291
292#define try_to_unmap(page, refs) false
293
294static inline int page_mkclean(struct page *page)
295{
296 return 0;
297}
298
299
300#endif /* CONFIG_MMU */
301
302#endif /* _LINUX_RMAP_H */
303