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 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 */ 77 struct anon_vma_chain { 78 struct vm_area_struct *vma; 79 struct anon_vma *anon_vma; 80 struct list_head same_vma; /* locked by mmap_lock & 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 88 enum 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_HWPOISON = 0x20, /* corrupted page is recoverable */ 95 TTU_BATCH_FLUSH = 0x40, /* Batch TLB flushes where possible 96 * and caller guarantees they will 97 * do a final flush if necessary */ 98 TTU_RMAP_LOCKED = 0x80, /* do not grab rmap lock: 99 * caller holds it */ 100 TTU_SPLIT_FREEZE = 0x100, /* freeze pte under splitting thp */ 101 }; 102 103 #ifdef CONFIG_MMU 104 static inline void get_anon_vma(struct anon_vma *anon_vma) 105 { 106 atomic_inc(&anon_vma->refcount); 107 } 108 109 void __put_anon_vma(struct anon_vma *anon_vma); 110 111 static inline void put_anon_vma(struct anon_vma *anon_vma) 112 { 113 if (atomic_dec_and_test(&anon_vma->refcount)) 114 __put_anon_vma(anon_vma); 115 } 116 117 static inline void anon_vma_lock_write(struct anon_vma *anon_vma) 118 { 119 down_write(&anon_vma->root->rwsem); 120 } 121 122 static inline void anon_vma_unlock_write(struct anon_vma *anon_vma) 123 { 124 up_write(&anon_vma->root->rwsem); 125 } 126 127 static inline void anon_vma_lock_read(struct anon_vma *anon_vma) 128 { 129 down_read(&anon_vma->root->rwsem); 130 } 131 132 static inline void anon_vma_unlock_read(struct anon_vma *anon_vma) 133 { 134 up_read(&anon_vma->root->rwsem); 135 } 136 137 138 /* 139 * anon_vma helper functions. 140 */ 141 void anon_vma_init(void); /* create anon_vma_cachep */ 142 int __anon_vma_prepare(struct vm_area_struct *); 143 void unlink_anon_vmas(struct vm_area_struct *); 144 int anon_vma_clone(struct vm_area_struct *, struct vm_area_struct *); 145 int anon_vma_fork(struct vm_area_struct *, struct vm_area_struct *); 146 147 static inline int anon_vma_prepare(struct vm_area_struct *vma) 148 { 149 if (likely(vma->anon_vma)) 150 return 0; 151 152 return __anon_vma_prepare(vma); 153 } 154 155 static inline void anon_vma_merge(struct vm_area_struct *vma, 156 struct vm_area_struct *next) 157 { 158 VM_BUG_ON_VMA(vma->anon_vma != next->anon_vma, vma); 159 unlink_anon_vmas(next); 160 } 161 162 struct anon_vma *page_get_anon_vma(struct page *page); 163 164 /* bitflags for do_page_add_anon_rmap() */ 165 #define RMAP_EXCLUSIVE 0x01 166 #define RMAP_COMPOUND 0x02 167 168 /* 169 * rmap interfaces called when adding or removing pte of page 170 */ 171 void page_move_anon_rmap(struct page *, struct vm_area_struct *); 172 void page_add_anon_rmap(struct page *, struct vm_area_struct *, 173 unsigned long, bool); 174 void do_page_add_anon_rmap(struct page *, struct vm_area_struct *, 175 unsigned long, int); 176 void page_add_new_anon_rmap(struct page *, struct vm_area_struct *, 177 unsigned long, bool); 178 void page_add_file_rmap(struct page *, bool); 179 void page_remove_rmap(struct page *, bool); 180 181 void hugepage_add_anon_rmap(struct page *, struct vm_area_struct *, 182 unsigned long); 183 void hugepage_add_new_anon_rmap(struct page *, struct vm_area_struct *, 184 unsigned long); 185 186 static inline void page_dup_rmap(struct page *page, bool compound) 187 { 188 atomic_inc(compound ? compound_mapcount_ptr(page) : &page->_mapcount); 189 } 190 191 /* 192 * Called from mm/vmscan.c to handle paging out 193 */ 194 int page_referenced(struct page *, int is_locked, 195 struct mem_cgroup *memcg, unsigned long *vm_flags); 196 197 bool try_to_unmap(struct page *, enum ttu_flags flags); 198 199 /* Avoid racy checks */ 200 #define PVMW_SYNC (1 << 0) 201 /* Look for migarion entries rather than present PTEs */ 202 #define PVMW_MIGRATION (1 << 1) 203 204 struct page_vma_mapped_walk { 205 struct page *page; 206 struct vm_area_struct *vma; 207 unsigned long address; 208 pmd_t *pmd; 209 pte_t *pte; 210 spinlock_t *ptl; 211 unsigned int flags; 212 }; 213 214 static inline void page_vma_mapped_walk_done(struct page_vma_mapped_walk *pvmw) 215 { 216 if (pvmw->pte) 217 pte_unmap(pvmw->pte); 218 if (pvmw->ptl) 219 spin_unlock(pvmw->ptl); 220 } 221 222 bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw); 223 224 /* 225 * Used by swapoff to help locate where page is expected in vma. 226 */ 227 unsigned long page_address_in_vma(struct page *, struct vm_area_struct *); 228 229 /* 230 * Cleans the PTEs of shared mappings. 231 * (and since clean PTEs should also be readonly, write protects them too) 232 * 233 * returns the number of cleaned PTEs. 234 */ 235 int page_mkclean(struct page *); 236 237 /* 238 * called in munlock()/munmap() path to check for other vmas holding 239 * the page mlocked. 240 */ 241 void try_to_munlock(struct page *); 242 243 void remove_migration_ptes(struct page *old, struct page *new, bool locked); 244 245 /* 246 * Called by memory-failure.c to kill processes. 247 */ 248 struct anon_vma *page_lock_anon_vma_read(struct page *page); 249 void page_unlock_anon_vma_read(struct anon_vma *anon_vma); 250 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma); 251 252 /* 253 * rmap_walk_control: To control rmap traversing for specific needs 254 * 255 * arg: passed to rmap_one() and invalid_vma() 256 * rmap_one: executed on each vma where page is mapped 257 * done: for checking traversing termination condition 258 * anon_lock: for getting anon_lock by optimized way rather than default 259 * invalid_vma: for skipping uninterested vma 260 */ 261 struct rmap_walk_control { 262 void *arg; 263 /* 264 * Return false if page table scanning in rmap_walk should be stopped. 265 * Otherwise, return true. 266 */ 267 bool (*rmap_one)(struct page *page, struct vm_area_struct *vma, 268 unsigned long addr, void *arg); 269 int (*done)(struct page *page); 270 struct anon_vma *(*anon_lock)(struct page *page); 271 bool (*invalid_vma)(struct vm_area_struct *vma, void *arg); 272 }; 273 274 void rmap_walk(struct page *page, struct rmap_walk_control *rwc); 275 void rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc); 276 277 #else /* !CONFIG_MMU */ 278 279 #define anon_vma_init() do {} while (0) 280 #define anon_vma_prepare(vma) (0) 281 #define anon_vma_link(vma) do {} while (0) 282 283 static inline int page_referenced(struct page *page, int is_locked, 284 struct mem_cgroup *memcg, 285 unsigned long *vm_flags) 286 { 287 *vm_flags = 0; 288 return 0; 289 } 290 291 #define try_to_unmap(page, refs) false 292 293 static inline int page_mkclean(struct page *page) 294 { 295 return 0; 296 } 297 298 299 #endif /* CONFIG_MMU */ 300 301 #endif /* _LINUX_RMAP_H */ 302