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