include/linux/rmap.h

   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 extern int isolate_lru_page(struct page *page);
  16 extern void putback_lru_page(struct page *page);
  17 extern unsigned long reclaim_pages(struct list_head *page_list);
  18
  19 /*
  20  * The anon_vma heads a list of private "related" vmas, to scan if
  21  * an anonymous page pointing to this anon_vma needs to be unmapped:
  22  * the vmas on the list will be related by forking, or by splitting.
  23  *
  24  * Since vmas come and go as they are split and merged (particularly
  25  * in mprotect), the mapping field of an anonymous page cannot point
  26  * directly to a vma: instead it points to an anon_vma, on whose list
  27  * the related vmas can be easily linked or unlinked.
  28  *
  29  * After unlinking the last vma on the list, we must garbage collect
  30  * the anon_vma object itself: we're guaranteed no page can be
  31  * pointing to this anon_vma once its vma list is empty.
  32  */
  33 struct anon_vma {
  34         struct anon_vma *root;          /* Root of this anon_vma tree */
  35         struct rw_semaphore rwsem;      /* W: modification, R: walking the list */
  36         /*
  37          * The refcount is taken on an anon_vma when there is no
  38          * guarantee that the vma of page tables will exist for
  39          * the duration of the operation. A caller that takes
  40          * the reference is responsible for clearing up the
  41          * anon_vma if they are the last user on release
  42          */
  43         atomic_t refcount;
  44
  45         /*
  46          * Count of child anon_vmas and VMAs which points to this anon_vma.
  47          *
  48          * This counter is used for making decision about reusing anon_vma
  49          * instead of forking new one. See comments in function anon_vma_clone.
  50          */
  51         unsigned degree;
  52
  53         struct anon_vma *parent;        /* Parent of this anon_vma */
  54
  55         /*
  56          * NOTE: the LSB of the rb_root.rb_node is set by
  57          * mm_take_all_locks() _after_ taking the above lock. So the
  58          * rb_root must only be read/written after taking the above lock
  59          * to be sure to see a valid next pointer. The LSB bit itself
  60          * is serialized by a system wide lock only visible to
  61          * mm_take_all_locks() (mm_all_locks_mutex).
  62          */
  63
  64         /* Interval tree of private "related" vmas */
  65         struct rb_root_cached rb_root;
  66 };
  67
  68 /*
  69  * The copy-on-write semantics of fork mean that an anon_vma
  70  * can become associated with multiple processes. Furthermore,
  71  * each child process will have its own anon_vma, where new
  72  * pages for that process are instantiated.
  73  *
  74  * This structure allows us to find the anon_vmas associated
  75  * with a VMA, or the VMAs associated with an anon_vma.
  76  * The "same_vma" list contains the anon_vma_chains linking
  77  * all the anon_vmas associated with this VMA.
  78  * The "rb" field indexes on an interval tree the anon_vma_chains
  79  * which link all the VMAs associated with this anon_vma.
  80  */
  81 struct anon_vma_chain {
  82         struct vm_area_struct *vma;
  83         struct anon_vma *anon_vma;
  84         struct list_head same_vma;   /* locked by mmap_sem & page_table_lock */
  85         struct rb_node rb;                      /* locked by anon_vma->rwsem */
  86         unsigned long rb_subtree_last;
  87 #ifdef CONFIG_DEBUG_VM_RB
  88         unsigned long cached_vma_start, cached_vma_last;
  89 #endif
  90 };
  91
  92 enum ttu_flags {
  93         TTU_MIGRATION           = 0x1,  /* migration mode */
  94         TTU_MUNLOCK             = 0x2,  /* munlock mode */
  95
  96         TTU_SPLIT_HUGE_PMD      = 0x4,  /* split huge PMD if any */
  97         TTU_IGNORE_MLOCK        = 0x8,  /* ignore mlock */
  98         TTU_IGNORE_ACCESS       = 0x10, /* don't age */
  99         TTU_IGNORE_HWPOISON     = 0x20, /* corrupted page is recoverable */
 100         TTU_BATCH_FLUSH         = 0x40, /* Batch TLB flushes where possible
 101                                          * and caller guarantees they will
 102                                          * do a final flush if necessary */
 103         TTU_RMAP_LOCKED         = 0x80, /* do not grab rmap lock:
 104                                          * caller holds it */
 105         TTU_SPLIT_FREEZE        = 0x100,                /* freeze pte under splitting thp */
 106 };
 107
 108 #ifdef CONFIG_MMU
 109 static inline void get_anon_vma(struct anon_vma *anon_vma)
 110 {
 111         atomic_inc(&anon_vma->refcount);
 112 }
 113
 114 void __put_anon_vma(struct anon_vma *anon_vma);
 115
 116 static inline void put_anon_vma(struct anon_vma *anon_vma)
 117 {
 118         if (atomic_dec_and_test(&anon_vma->refcount))
 119                 __put_anon_vma(anon_vma);
 120 }
 121
 122 static inline void anon_vma_lock_write(struct anon_vma *anon_vma)
 123 {
 124         down_write(&anon_vma->root->rwsem);
 125 }
 126
 127 static inline void anon_vma_unlock_write(struct anon_vma *anon_vma)
 128 {
 129         up_write(&anon_vma->root->rwsem);
 130 }
 131
 132 static inline void anon_vma_lock_read(struct anon_vma *anon_vma)
 133 {
 134         down_read(&anon_vma->root->rwsem);
 135 }
 136
 137 static inline void anon_vma_unlock_read(struct anon_vma *anon_vma)
 138 {
 139         up_read(&anon_vma->root->rwsem);
 140 }
 141
 142
 143 /*
 144  * anon_vma helper functions.
 145  */
 146 void anon_vma_init(void);       /* create anon_vma_cachep */
 147 int  __anon_vma_prepare(struct vm_area_struct *);
 148 void unlink_anon_vmas(struct vm_area_struct *);
 149 int anon_vma_clone(struct vm_area_struct *, struct vm_area_struct *);
 150 int anon_vma_fork(struct vm_area_struct *, struct vm_area_struct *);
 151
 152 static inline int anon_vma_prepare(struct vm_area_struct *vma)
 153 {
 154         if (likely(vma->anon_vma))
 155                 return 0;
 156
 157         return __anon_vma_prepare(vma);
 158 }
 159
 160 static inline void anon_vma_merge(struct vm_area_struct *vma,
 161                                   struct vm_area_struct *next)
 162 {
 163         VM_BUG_ON_VMA(vma->anon_vma != next->anon_vma, vma);
 164         unlink_anon_vmas(next);
 165 }
 166
 167 struct anon_vma *page_get_anon_vma(struct page *page);
 168
 169 /* bitflags for do_page_add_anon_rmap() */
 170 #define RMAP_EXCLUSIVE 0x01
 171 #define RMAP_COMPOUND 0x02
 172
 173 /*
 174  * rmap interfaces called when adding or removing pte of page
 175  */
 176 void page_move_anon_rmap(struct page *, struct vm_area_struct *);
 177 void page_add_anon_rmap(struct page *, struct vm_area_struct *,
 178                 unsigned long, bool);
 179 void do_page_add_anon_rmap(struct page *, struct vm_area_struct *,
 180                            unsigned long, int);
 181 void page_add_new_anon_rmap(struct page *, struct vm_area_struct *,
 182                 unsigned long, bool);
 183 void page_add_file_rmap(struct page *, bool);
 184 void page_remove_rmap(struct page *, bool);
 185
 186 void hugepage_add_anon_rmap(struct page *, struct vm_area_struct *,
 187                             unsigned long);
 188 void hugepage_add_new_anon_rmap(struct page *, struct vm_area_struct *,
 189                                 unsigned long);
 190
 191 static inline void page_dup_rmap(struct page *page, bool compound)
 192 {
 193         atomic_inc(compound ? compound_mapcount_ptr(page) : &page->_mapcount);
 194 }
 195
 196 /*
 197  * Called from mm/vmscan.c to handle paging out
 198  */
 199 int page_referenced(struct page *, int is_locked,
 200                         struct mem_cgroup *memcg, unsigned long *vm_flags);
 201
 202 bool try_to_unmap(struct page *, enum ttu_flags flags);
 203
 204 /* Avoid racy checks */
 205 #define PVMW_SYNC               (1 << 0)
 206 /* Look for migarion entries rather than present PTEs */
 207 #define PVMW_MIGRATION          (1 << 1)
 208
 209 struct page_vma_mapped_walk {
 210         struct page *page;
 211         struct vm_area_struct *vma;
 212         unsigned long address;
 213         pmd_t *pmd;
 214         pte_t *pte;
 215         spinlock_t *ptl;
 216         unsigned int flags;
 217 };
 218
 219 static inline void page_vma_mapped_walk_done(struct page_vma_mapped_walk *pvmw)
 220 {
 221         if (pvmw->pte)
 222                 pte_unmap(pvmw->pte);
 223         if (pvmw->ptl)
 224                 spin_unlock(pvmw->ptl);
 225 }
 226
 227 bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw);
 228
 229 /*
 230  * Used by swapoff to help locate where page is expected in vma.
 231  */
 232 unsigned long page_address_in_vma(struct page *, struct vm_area_struct *);
 233
 234 /*
 235  * Cleans the PTEs of shared mappings.
 236  * (and since clean PTEs should also be readonly, write protects them too)
 237  *
 238  * returns the number of cleaned PTEs.
 239  */
 240 int page_mkclean(struct page *);
 241
 242 /*
 243  * called in munlock()/munmap() path to check for other vmas holding
 244  * the page mlocked.
 245  */
 246 void try_to_munlock(struct page *);
 247
 248 void remove_migration_ptes(struct page *old, struct page *new, bool locked);
 249
 250 /*
 251  * Called by memory-failure.c to kill processes.
 252  */
 253 struct anon_vma *page_lock_anon_vma_read(struct page *page);
 254 void page_unlock_anon_vma_read(struct anon_vma *anon_vma);
 255 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma);
 256
 257 /*
 258  * rmap_walk_control: To control rmap traversing for specific needs
 259  *
 260  * arg: passed to rmap_one() and invalid_vma()
 261  * rmap_one: executed on each vma where page is mapped
 262  * done: for checking traversing termination condition
 263  * anon_lock: for getting anon_lock by optimized way rather than default
 264  * invalid_vma: for skipping uninterested vma
 265  */
 266 struct rmap_walk_control {
 267         void *arg;
 268         /*
 269          * Return false if page table scanning in rmap_walk should be stopped.
 270          * Otherwise, return true.
 271          */
 272         bool (*rmap_one)(struct page *page, struct vm_area_struct *vma,
 273                                         unsigned long addr, void *arg);
 274         int (*done)(struct page *page);
 275         struct anon_vma *(*anon_lock)(struct page *page);
 276         bool (*invalid_vma)(struct vm_area_struct *vma, void *arg);
 277 };
 278
 279 void rmap_walk(struct page *page, struct rmap_walk_control *rwc);
 280 void rmap_walk_locked(struct page *page, struct rmap_walk_control *rwc);
 281
 282 #else   /* !CONFIG_MMU */
 283
 284 #define anon_vma_init()         do {} while (0)
 285 #define anon_vma_prepare(vma)   (0)
 286 #define anon_vma_link(vma)      do {} while (0)
 287
 288 static inline int page_referenced(struct page *page, int is_locked,
 289                                   struct mem_cgroup *memcg,
 290                                   unsigned long *vm_flags)
 291 {
 292         *vm_flags = 0;
 293         return 0;
 294 }
 295
 296 #define try_to_unmap(page, refs) false
 297
 298 static inline int page_mkclean(struct page *page)
 299 {
 300         return 0;
 301 }
 302
 303
 304 #endif  /* CONFIG_MMU */
 305
 306 #endif  /* _LINUX_RMAP_H */