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_SPLIT_HUGE_PMD = 0x4, /* split huge PMD if any */ 90 TTU_IGNORE_MLOCK = 0x8, /* ignore mlock */ 91 TTU_SYNC = 0x10, /* avoid racy checks with PVMW_SYNC */ 92 TTU_IGNORE_HWPOISON = 0x20, /* corrupted page is recoverable */ 93 TTU_BATCH_FLUSH = 0x40, /* Batch TLB flushes where possible 94 * and caller guarantees they will 95 * do a final flush if necessary */ 96 TTU_RMAP_LOCKED = 0x80, /* do not grab rmap lock: 97 * caller holds it */ 98 }; 99 100 #ifdef CONFIG_MMU 101 static inline void get_anon_vma(struct anon_vma *anon_vma) 102 { 103 atomic_inc(&anon_vma->refcount); 104 } 105 106 void __put_anon_vma(struct anon_vma *anon_vma); 107 108 static inline void put_anon_vma(struct anon_vma *anon_vma) 109 { 110 if (atomic_dec_and_test(&anon_vma->refcount)) 111 __put_anon_vma(anon_vma); 112 } 113 114 static inline void anon_vma_lock_write(struct anon_vma *anon_vma) 115 { 116 down_write(&anon_vma->root->rwsem); 117 } 118 119 static inline void anon_vma_unlock_write(struct anon_vma *anon_vma) 120 { 121 up_write(&anon_vma->root->rwsem); 122 } 123 124 static inline void anon_vma_lock_read(struct anon_vma *anon_vma) 125 { 126 down_read(&anon_vma->root->rwsem); 127 } 128 129 static inline void anon_vma_unlock_read(struct anon_vma *anon_vma) 130 { 131 up_read(&anon_vma->root->rwsem); 132 } 133 134 135 /* 136 * anon_vma helper functions. 137 */ 138 void anon_vma_init(void); /* create anon_vma_cachep */ 139 int __anon_vma_prepare(struct vm_area_struct *); 140 void unlink_anon_vmas(struct vm_area_struct *); 141 int anon_vma_clone(struct vm_area_struct *, struct vm_area_struct *); 142 int anon_vma_fork(struct vm_area_struct *, struct vm_area_struct *); 143 144 static inline int anon_vma_prepare(struct vm_area_struct *vma) 145 { 146 if (likely(vma->anon_vma)) 147 return 0; 148 149 return __anon_vma_prepare(vma); 150 } 151 152 static inline void anon_vma_merge(struct vm_area_struct *vma, 153 struct vm_area_struct *next) 154 { 155 VM_BUG_ON_VMA(vma->anon_vma != next->anon_vma, vma); 156 unlink_anon_vmas(next); 157 } 158 159 struct anon_vma *page_get_anon_vma(struct page *page); 160 161 /* bitflags for do_page_add_anon_rmap() */ 162 #define RMAP_EXCLUSIVE 0x01 163 #define RMAP_COMPOUND 0x02 164 165 /* 166 * rmap interfaces called when adding or removing pte of page 167 */ 168 void page_move_anon_rmap(struct page *, struct vm_area_struct *); 169 void page_add_anon_rmap(struct page *, struct vm_area_struct *, 170 unsigned long, bool); 171 void do_page_add_anon_rmap(struct page *, struct vm_area_struct *, 172 unsigned long, int); 173 void page_add_new_anon_rmap(struct page *, struct vm_area_struct *, 174 unsigned long, bool); 175 void page_add_file_rmap(struct page *, bool); 176 void page_remove_rmap(struct page *, bool); 177 178 void hugepage_add_anon_rmap(struct page *, struct vm_area_struct *, 179 unsigned long); 180 void hugepage_add_new_anon_rmap(struct page *, struct vm_area_struct *, 181 unsigned long); 182 183 static inline void page_dup_rmap(struct page *page, bool compound) 184 { 185 atomic_inc(compound ? compound_mapcount_ptr(page) : &page->_mapcount); 186 } 187 188 /* 189 * Called from mm/vmscan.c to handle paging out 190 */ 191 int page_referenced(struct page *, int is_locked, 192 struct mem_cgroup *memcg, unsigned long *vm_flags); 193 194 void try_to_migrate(struct page *page, enum ttu_flags flags); 195 void try_to_unmap(struct page *, enum ttu_flags flags); 196 197 int make_device_exclusive_range(struct mm_struct *mm, unsigned long start, 198 unsigned long end, struct page **pages, 199 void *arg); 200 201 /* Avoid racy checks */ 202 #define PVMW_SYNC (1 << 0) 203 /* Look for migarion entries rather than present PTEs */ 204 #define PVMW_MIGRATION (1 << 1) 205 206 struct page_vma_mapped_walk { 207 struct page *page; 208 struct vm_area_struct *vma; 209 unsigned long address; 210 pmd_t *pmd; 211 pte_t *pte; 212 spinlock_t *ptl; 213 unsigned int flags; 214 }; 215 216 static inline void page_vma_mapped_walk_done(struct page_vma_mapped_walk *pvmw) 217 { 218 /* HugeTLB pte is set to the relevant page table entry without pte_mapped. */ 219 if (pvmw->pte && !PageHuge(pvmw->page)) 220 pte_unmap(pvmw->pte); 221 if (pvmw->ptl) 222 spin_unlock(pvmw->ptl); 223 } 224 225 bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw); 226 227 /* 228 * Used by swapoff to help locate where page is expected in vma. 229 */ 230 unsigned long page_address_in_vma(struct page *, struct vm_area_struct *); 231 232 /* 233 * Cleans the PTEs of shared mappings. 234 * (and since clean PTEs should also be readonly, write protects them too) 235 * 236 * returns the number of cleaned PTEs. 237 */ 238 int page_mkclean(struct page *); 239 240 /* 241 * called in munlock()/munmap() path to check for other vmas holding 242 * the page mlocked. 243 */ 244 void page_mlock(struct page *page); 245 246 void remove_migration_ptes(struct page *old, struct page *new, bool locked); 247 248 /* 249 * Called by memory-failure.c to kill processes. 250 */ 251 struct anon_vma *page_lock_anon_vma_read(struct page *page); 252 void page_unlock_anon_vma_read(struct anon_vma *anon_vma); 253 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma); 254 255 /* 256 * rmap_walk_control: To control rmap traversing for specific needs 257 * 258 * arg: passed to rmap_one() and invalid_vma() 259 * rmap_one: executed on each vma where page is mapped 260 * done: for checking traversing termination condition 261 * anon_lock: for getting anon_lock by optimized way rather than default 262 * invalid_vma: for skipping uninterested vma 263 */ 264 struct rmap_walk_control { 265 void *arg; 266 /* 267 * Return false if page table scanning in rmap_walk should be stopped. 268 * Otherwise, return true. 269 */ 270 bool (*rmap_one)(struct page *page, struct vm_area_struct *vma, 271 unsigned long addr, void *arg); 272 int (*done)(struct page *page); 273 struct anon_vma *(*anon_lock)(struct page *page); 274 bool (*invalid_vma)(struct vm_area_struct *vma, void *arg); 275 }; 276 277 void rmap_walk(struct page *page, struct rmap_walk_control *rwc); 278 void rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc); 279 280 #else /* !CONFIG_MMU */ 281 282 #define anon_vma_init() do {} while (0) 283 #define anon_vma_prepare(vma) (0) 284 #define anon_vma_link(vma) do {} while (0) 285 286 static inline int page_referenced(struct page *page, int is_locked, 287 struct mem_cgroup *memcg, 288 unsigned long *vm_flags) 289 { 290 *vm_flags = 0; 291 return 0; 292 } 293 294 #define try_to_unmap(page, refs) false 295 296 static inline int page_mkclean(struct page *page) 297 { 298 return 0; 299 } 300 301 302 #endif /* CONFIG_MMU */ 303 304 #endif /* _LINUX_RMAP_H */ 305