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
  13 /*
  14  * The anon_vma heads a list of private "related" vmas, to scan if
  15  * an anonymous page pointing to this anon_vma needs to be unmapped:
  16  * the vmas on the list will be related by forking, or by splitting.
  17  *
  18  * Since vmas come and go as they are split and merged (particularly
  19  * in mprotect), the mapping field of an anonymous page cannot point
  20  * directly to a vma: instead it points to an anon_vma, on whose list
  21  * the related vmas can be easily linked or unlinked.
  22  *
  23  * After unlinking the last vma on the list, we must garbage collect
  24  * the anon_vma object itself: we're guaranteed no page can be
  25  * pointing to this anon_vma once its vma list is empty.
  26  */
  27 struct anon_vma {
  28         struct anon_vma *root;          /* Root of this anon_vma tree */
  29         struct rw_semaphore rwsem;      /* W: modification, R: walking the list */
  30         /*
  31          * The refcount is taken on an anon_vma when there is no
  32          * guarantee that the vma of page tables will exist for
  33          * the duration of the operation. A caller that takes
  34          * the reference is responsible for clearing up the
  35          * anon_vma if they are the last user on release
  36          */
  37         atomic_t refcount;
  38
  39         /*
  40          * Count of child anon_vmas and VMAs which points to this anon_vma.
  41          *
  42          * This counter is used for making decision about reusing anon_vma
  43          * instead of forking new one. See comments in function anon_vma_clone.
  44          */
  45         unsigned degree;
  46
  47         struct anon_vma *parent;        /* Parent of this anon_vma */
  48
  49         /*
  50          * NOTE: the LSB of the rb_root.rb_node is set by
  51          * mm_take_all_locks() _after_ taking the above lock. So the
  52          * rb_root must only be read/written after taking the above lock
  53          * to be sure to see a valid next pointer. The LSB bit itself
  54          * is serialized by a system wide lock only visible to
  55          * mm_take_all_locks() (mm_all_locks_mutex).
  56          */
  57         struct rb_root rb_root; /* Interval tree of private "related" vmas */
  58 };
  59
  60 /*
  61  * The copy-on-write semantics of fork mean that an anon_vma
  62  * can become associated with multiple processes. Furthermore,
  63  * each child process will have its own anon_vma, where new
  64  * pages for that process are instantiated.
  65  *
  66  * This structure allows us to find the anon_vmas associated
  67  * with a VMA, or the VMAs associated with an anon_vma.
  68  * The "same_vma" list contains the anon_vma_chains linking
  69  * all the anon_vmas associated with this VMA.
  70  * The "rb" field indexes on an interval tree the anon_vma_chains
  71  * which link all the VMAs associated with this anon_vma.
  72  */
  73 struct anon_vma_chain {
  74         struct vm_area_struct *vma;
  75         struct anon_vma *anon_vma;
  76         struct list_head same_vma;   /* locked by mmap_sem & page_table_lock */
  77         struct rb_node rb;                      /* locked by anon_vma->rwsem */
  78         unsigned long rb_subtree_last;
  79 #ifdef CONFIG_DEBUG_VM_RB
  80         unsigned long cached_vma_start, cached_vma_last;
  81 #endif
  82 };
  83
  84 enum ttu_flags {
  85         TTU_UNMAP = 1,                  /* unmap mode */
  86         TTU_MIGRATION = 2,              /* migration mode */
  87         TTU_MUNLOCK = 4,                /* munlock mode */
  88         TTU_LZFREE = 8,                 /* lazy free mode */
  89         TTU_SPLIT_HUGE_PMD = 16,        /* split huge PMD if any */
  90
  91         TTU_IGNORE_MLOCK = (1 << 8),    /* ignore mlock */
  92         TTU_IGNORE_ACCESS = (1 << 9),   /* don't age */
  93         TTU_IGNORE_HWPOISON = (1 << 10),/* corrupted page is recoverable */
  94         TTU_BATCH_FLUSH = (1 << 11),    /* Batch TLB flushes where possible
  95                                          * and caller guarantees they will
  96                                          * do a final flush if necessary */
  97         TTU_RMAP_LOCKED = (1 << 12)     /* 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 void anon_vma_merge(struct vm_area_struct *vma,
 146                                   struct vm_area_struct *next)
 147 {
 148         VM_BUG_ON_VMA(vma->anon_vma != next->anon_vma, vma);
 149         unlink_anon_vmas(next);
 150 }
 151
 152 struct anon_vma *page_get_anon_vma(struct page *page);
 153
 154 /* bitflags for do_page_add_anon_rmap() */
 155 #define RMAP_EXCLUSIVE 0x01
 156 #define RMAP_COMPOUND 0x02
 157
 158 /*
 159  * rmap interfaces called when adding or removing pte of page
 160  */
 161 void page_move_anon_rmap(struct page *, struct vm_area_struct *, unsigned long);
 162 void page_add_anon_rmap(struct page *, struct vm_area_struct *,
 163                 unsigned long, bool);
 164 void do_page_add_anon_rmap(struct page *, struct vm_area_struct *,
 165                            unsigned long, int);
 166 void page_add_new_anon_rmap(struct page *, struct vm_area_struct *,
 167                 unsigned long, bool);
 168 void page_add_file_rmap(struct page *);
 169 void page_remove_rmap(struct page *, bool);
 170
 171 void hugepage_add_anon_rmap(struct page *, struct vm_area_struct *,
 172                             unsigned long);
 173 void hugepage_add_new_anon_rmap(struct page *, struct vm_area_struct *,
 174                                 unsigned long);
 175
 176 static inline void page_dup_rmap(struct page *page, bool compound)
 177 {
 178         atomic_inc(compound ? compound_mapcount_ptr(page) : &page->_mapcount);
 179 }
 180
 181 /*
 182  * Called from mm/vmscan.c to handle paging out
 183  */
 184 int page_referenced(struct page *, int is_locked,
 185                         struct mem_cgroup *memcg, unsigned long *vm_flags);
 186
 187 #define TTU_ACTION(x) ((x) & TTU_ACTION_MASK)
 188
 189 int try_to_unmap(struct page *, enum ttu_flags flags);
 190
 191 /*
 192  * Used by uprobes to replace a userspace page safely
 193  */
 194 pte_t *__page_check_address(struct page *, struct mm_struct *,
 195                                 unsigned long, spinlock_t **, int);
 196
 197 static inline pte_t *page_check_address(struct page *page, struct mm_struct *mm,
 198                                         unsigned long address,
 199                                         spinlock_t **ptlp, int sync)
 200 {
 201         pte_t *ptep;
 202
 203         __cond_lock(*ptlp, ptep = __page_check_address(page, mm, address,
 204                                                        ptlp, sync));
 205         return ptep;
 206 }
 207
 208 /*
 209  * Used by idle page tracking to check if a page was referenced via page
 210  * tables.
 211  */
 212 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 213 bool page_check_address_transhuge(struct page *page, struct mm_struct *mm,
 214                                   unsigned long address, pmd_t **pmdp,
 215                                   pte_t **ptep, spinlock_t **ptlp);
 216 #else
 217 static inline bool page_check_address_transhuge(struct page *page,
 218                                 struct mm_struct *mm, unsigned long address,
 219                                 pmd_t **pmdp, pte_t **ptep, spinlock_t **ptlp)
 220 {
 221         *ptep = page_check_address(page, mm, address, ptlp, 0);
 222         *pmdp = NULL;
 223         return !!*ptep;
 224 }
 225 #endif
 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 int try_to_munlock(struct 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         int (*rmap_one)(struct page *page, struct vm_area_struct *vma,
 267                                         unsigned long addr, void *arg);
 268         int (*done)(struct page *page);
 269         struct anon_vma *(*anon_lock)(struct page *page);
 270         bool (*invalid_vma)(struct vm_area_struct *vma, void *arg);
 271 };
 272
 273 int rmap_walk(struct page *page, struct rmap_walk_control *rwc);
 274 int rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc);
 275
 276 #else   /* !CONFIG_MMU */
 277
 278 #define anon_vma_init()         do {} while (0)
 279 #define anon_vma_prepare(vma)   (0)
 280 #define anon_vma_link(vma)      do {} while (0)
 281
 282 static inline int page_referenced(struct page *page, int is_locked,
 283                                   struct mem_cgroup *memcg,
 284                                   unsigned long *vm_flags)
 285 {
 286         *vm_flags = 0;
 287         return 0;
 288 }
 289
 290 #define try_to_unmap(page, refs) SWAP_FAIL
 291
 292 static inline int page_mkclean(struct page *page)
 293 {
 294         return 0;
 295 }
 296
 297
 298 #endif  /* CONFIG_MMU */
 299
 300 /*
 301  * Return values of try_to_unmap
 302  */
 303 #define SWAP_SUCCESS    0
 304 #define SWAP_AGAIN      1
 305 #define SWAP_FAIL       2
 306 #define SWAP_MLOCK      3
 307 #define SWAP_LZFREE     4
 308
 309 #endif  /* _LINUX_RMAP_H */