include/linux/rmap.h

   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 #include <linux/highmem.h>
  13
  14 /*
  15  * The anon_vma heads a list of private "related" vmas, to scan if
  16  * an anonymous page pointing to this anon_vma needs to be unmapped:
  17  * the vmas on the list will be related by forking, or by splitting.
  18  *
  19  * Since vmas come and go as they are split and merged (particularly
  20  * in mprotect), the mapping field of an anonymous page cannot point
  21  * directly to a vma: instead it points to an anon_vma, on whose list
  22  * the related vmas can be easily linked or unlinked.
  23  *
  24  * After unlinking the last vma on the list, we must garbage collect
  25  * the anon_vma object itself: we're guaranteed no page can be
  26  * pointing to this anon_vma once its vma list is empty.
  27  */
  28 struct anon_vma {
  29         struct anon_vma *root;          /* Root of this anon_vma tree */
  30         struct rw_semaphore rwsem;      /* W: modification, R: walking the list */
  31         /*
  32          * The refcount is taken on an anon_vma when there is no
  33          * guarantee that the vma of page tables will exist for
  34          * the duration of the operation. A caller that takes
  35          * the reference is responsible for clearing up the
  36          * anon_vma if they are the last user on release
  37          */
  38         atomic_t refcount;
  39
  40         /*
  41          * Count of child anon_vmas and VMAs which points to this anon_vma.
  42          *
  43          * This counter is used for making decision about reusing anon_vma
  44          * instead of forking new one. See comments in function anon_vma_clone.
  45          */
  46         unsigned degree;
  47
  48         struct anon_vma *parent;        /* Parent of this anon_vma */
  49
  50         /*
  51          * NOTE: the LSB of the rb_root.rb_node is set by
  52          * mm_take_all_locks() _after_ taking the above lock. So the
  53          * rb_root must only be read/written after taking the above lock
  54          * to be sure to see a valid next pointer. The LSB bit itself
  55          * is serialized by a system wide lock only visible to
  56          * mm_take_all_locks() (mm_all_locks_mutex).
  57          */
  58         struct rb_root rb_root; /* Interval tree of private "related" vmas */
  59 };
  60
  61 /*
  62  * The copy-on-write semantics of fork mean that an anon_vma
  63  * can become associated with multiple processes. Furthermore,
  64  * each child process will have its own anon_vma, where new
  65  * pages for that process are instantiated.
  66  *
  67  * This structure allows us to find the anon_vmas associated
  68  * with a VMA, or the VMAs associated with an anon_vma.
  69  * The "same_vma" list contains the anon_vma_chains linking
  70  * all the anon_vmas associated with this VMA.
  71  * The "rb" field indexes on an interval tree the anon_vma_chains
  72  * which link all the VMAs associated with this anon_vma.
  73  */
  74 struct anon_vma_chain {
  75         struct vm_area_struct *vma;
  76         struct anon_vma *anon_vma;
  77         struct list_head same_vma;   /* locked by mmap_sem & page_table_lock */
  78         struct rb_node rb;                      /* locked by anon_vma->rwsem */
  79         unsigned long rb_subtree_last;
  80 #ifdef CONFIG_DEBUG_VM_RB
  81         unsigned long cached_vma_start, cached_vma_last;
  82 #endif
  83 };
  84
  85 enum ttu_flags {
  86         TTU_UNMAP = 1,                  /* unmap mode */
  87         TTU_MIGRATION = 2,              /* migration mode */
  88         TTU_MUNLOCK = 4,                /* munlock mode */
  89         TTU_LZFREE = 8,                 /* lazy free mode */
  90         TTU_SPLIT_HUGE_PMD = 16,        /* split huge PMD if any */
  91
  92         TTU_IGNORE_MLOCK = (1 << 8),    /* ignore mlock */
  93         TTU_IGNORE_ACCESS = (1 << 9),   /* don't age */
  94         TTU_IGNORE_HWPOISON = (1 << 10),/* corrupted page is recoverable */
  95         TTU_BATCH_FLUSH = (1 << 11),    /* Batch TLB flushes where possible
  96                                          * and caller guarantees they will
  97                                          * do a final flush if necessary */
  98         TTU_RMAP_LOCKED = (1 << 12)     /* do not grab rmap lock:
  99                                          * caller holds it */
 100 };
 101
 102 #ifdef CONFIG_MMU
 103 static inline void get_anon_vma(struct anon_vma *anon_vma)
 104 {
 105         atomic_inc(&anon_vma->refcount);
 106 }
 107
 108 void __put_anon_vma(struct anon_vma *anon_vma);
 109
 110 static inline void put_anon_vma(struct anon_vma *anon_vma)
 111 {
 112         if (atomic_dec_and_test(&anon_vma->refcount))
 113                 __put_anon_vma(anon_vma);
 114 }
 115
 116 static inline void anon_vma_lock_write(struct anon_vma *anon_vma)
 117 {
 118         down_write(&anon_vma->root->rwsem);
 119 }
 120
 121 static inline void anon_vma_unlock_write(struct anon_vma *anon_vma)
 122 {
 123         up_write(&anon_vma->root->rwsem);
 124 }
 125
 126 static inline void anon_vma_lock_read(struct anon_vma *anon_vma)
 127 {
 128         down_read(&anon_vma->root->rwsem);
 129 }
 130
 131 static inline void anon_vma_unlock_read(struct anon_vma *anon_vma)
 132 {
 133         up_read(&anon_vma->root->rwsem);
 134 }
 135
 136
 137 /*
 138  * anon_vma helper functions.
 139  */
 140 void anon_vma_init(void);       /* create anon_vma_cachep */
 141 int  __anon_vma_prepare(struct vm_area_struct *);
 142 void unlink_anon_vmas(struct vm_area_struct *);
 143 int anon_vma_clone(struct vm_area_struct *, struct vm_area_struct *);
 144 int anon_vma_fork(struct vm_area_struct *, struct vm_area_struct *);
 145
 146 static inline int anon_vma_prepare(struct vm_area_struct *vma)
 147 {
 148         if (likely(vma->anon_vma))
 149                 return 0;
 150
 151         return __anon_vma_prepare(vma);
 152 }
 153
 154 static inline void anon_vma_merge(struct vm_area_struct *vma,
 155                                   struct vm_area_struct *next)
 156 {
 157         VM_BUG_ON_VMA(vma->anon_vma != next->anon_vma, vma);
 158         unlink_anon_vmas(next);
 159 }
 160
 161 struct anon_vma *page_get_anon_vma(struct page *page);
 162
 163 /* bitflags for do_page_add_anon_rmap() */
 164 #define RMAP_EXCLUSIVE 0x01
 165 #define RMAP_COMPOUND 0x02
 166
 167 /*
 168  * rmap interfaces called when adding or removing pte of page
 169  */
 170 void page_move_anon_rmap(struct page *, struct vm_area_struct *);
 171 void page_add_anon_rmap(struct page *, struct vm_area_struct *,
 172                 unsigned long, bool);
 173 void do_page_add_anon_rmap(struct page *, struct vm_area_struct *,
 174                            unsigned long, int);
 175 void page_add_new_anon_rmap(struct page *, struct vm_area_struct *,
 176                 unsigned long, bool);
 177 void page_add_file_rmap(struct page *, bool);
 178 void page_remove_rmap(struct page *, bool);
 179
 180 void hugepage_add_anon_rmap(struct page *, struct vm_area_struct *,
 181                             unsigned long);
 182 void hugepage_add_new_anon_rmap(struct page *, struct vm_area_struct *,
 183                                 unsigned long);
 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 page_referenced(struct page *, int is_locked,
 194                         struct mem_cgroup *memcg, unsigned long *vm_flags);
 195
 196 #define TTU_ACTION(x) ((x) & TTU_ACTION_MASK)
 197
 198 int try_to_unmap(struct page *, enum ttu_flags flags);
 199
 200 /*
 201  * Used by uprobes to replace a userspace page safely
 202  */
 203 pte_t *__page_check_address(struct page *, struct mm_struct *,
 204                                 unsigned long, spinlock_t **, int);
 205
 206 static inline pte_t *page_check_address(struct page *page, struct mm_struct *mm,
 207                                         unsigned long address,
 208                                         spinlock_t **ptlp, int sync)
 209 {
 210         pte_t *ptep;
 211
 212         __cond_lock(*ptlp, ptep = __page_check_address(page, mm, address,
 213                                                        ptlp, sync));
 214         return ptep;
 215 }
 216
 217 /*
 218  * Used by idle page tracking to check if a page was referenced via page
 219  * tables.
 220  */
 221 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 222 bool page_check_address_transhuge(struct page *page, struct mm_struct *mm,
 223                                   unsigned long address, pmd_t **pmdp,
 224                                   pte_t **ptep, spinlock_t **ptlp);
 225 #else
 226 static inline bool page_check_address_transhuge(struct page *page,
 227                                 struct mm_struct *mm, unsigned long address,
 228                                 pmd_t **pmdp, pte_t **ptep, spinlock_t **ptlp)
 229 {
 230         *ptep = page_check_address(page, mm, address, ptlp, 0);
 231         *pmdp = NULL;
 232         return !!*ptep;
 233 }
 234 #endif
 235
 236 /* Avoid racy checks */
 237 #define PVMW_SYNC               (1 << 0)
 238 /* Look for migarion entries rather than present PTEs */
 239 #define PVMW_MIGRATION          (1 << 1)
 240
 241 struct page_vma_mapped_walk {
 242         struct page *page;
 243         struct vm_area_struct *vma;
 244         unsigned long address;
 245         pmd_t *pmd;
 246         pte_t *pte;
 247         spinlock_t *ptl;
 248         unsigned int flags;
 249 };
 250
 251 static inline void page_vma_mapped_walk_done(struct page_vma_mapped_walk *pvmw)
 252 {
 253         if (pvmw->pte)
 254                 pte_unmap(pvmw->pte);
 255         if (pvmw->ptl)
 256                 spin_unlock(pvmw->ptl);
 257 }
 258
 259 bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw);
 260
 261 /*
 262  * Used by swapoff to help locate where page is expected in vma.
 263  */
 264 unsigned long page_address_in_vma(struct page *, struct vm_area_struct *);
 265
 266 /*
 267  * Cleans the PTEs of shared mappings.
 268  * (and since clean PTEs should also be readonly, write protects them too)
 269  *
 270  * returns the number of cleaned PTEs.
 271  */
 272 int page_mkclean(struct page *);
 273
 274 /*
 275  * called in munlock()/munmap() path to check for other vmas holding
 276  * the page mlocked.
 277  */
 278 int try_to_munlock(struct page *);
 279
 280 void remove_migration_ptes(struct page *old, struct page *new, bool locked);
 281
 282 /*
 283  * Called by memory-failure.c to kill processes.
 284  */
 285 struct anon_vma *page_lock_anon_vma_read(struct page *page);
 286 void page_unlock_anon_vma_read(struct anon_vma *anon_vma);
 287 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma);
 288
 289 /*
 290  * rmap_walk_control: To control rmap traversing for specific needs
 291  *
 292  * arg: passed to rmap_one() and invalid_vma()
 293  * rmap_one: executed on each vma where page is mapped
 294  * done: for checking traversing termination condition
 295  * anon_lock: for getting anon_lock by optimized way rather than default
 296  * invalid_vma: for skipping uninterested vma
 297  */
 298 struct rmap_walk_control {
 299         void *arg;
 300         int (*rmap_one)(struct page *page, struct vm_area_struct *vma,
 301                                         unsigned long addr, void *arg);
 302         int (*done)(struct page *page);
 303         struct anon_vma *(*anon_lock)(struct page *page);
 304         bool (*invalid_vma)(struct vm_area_struct *vma, void *arg);
 305 };
 306
 307 int rmap_walk(struct page *page, struct rmap_walk_control *rwc);
 308 int rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc);
 309
 310 #else   /* !CONFIG_MMU */
 311
 312 #define anon_vma_init()         do {} while (0)
 313 #define anon_vma_prepare(vma)   (0)
 314 #define anon_vma_link(vma)      do {} while (0)
 315
 316 static inline int page_referenced(struct page *page, int is_locked,
 317                                   struct mem_cgroup *memcg,
 318                                   unsigned long *vm_flags)
 319 {
 320         *vm_flags = 0;
 321         return 0;
 322 }
 323
 324 #define try_to_unmap(page, refs) SWAP_FAIL
 325
 326 static inline int page_mkclean(struct page *page)
 327 {
 328         return 0;
 329 }
 330
 331
 332 #endif  /* CONFIG_MMU */
 333
 334 /*
 335  * Return values of try_to_unmap
 336  */
 337 #define SWAP_SUCCESS    0
 338 #define SWAP_AGAIN      1
 339 #define SWAP_FAIL       2
 340 #define SWAP_MLOCK      3
 341 #define SWAP_LZFREE     4
 342
 343 #endif  /* _LINUX_RMAP_H */