rmap.h 8.4 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295
  1. #ifndef _LINUX_RMAP_H
  2. #define _LINUX_RMAP_H
  3. /*
  4. * Declarations for Reverse Mapping functions in mm/rmap.c
  5. */
  6. #include <linux/list.h>
  7. #include <linux/slab.h>
  8. #include <linux/mm.h>
  9. #include <linux/rwsem.h>
  10. #include <linux/memcontrol.h>
  11. /*
  12. * The anon_vma heads a list of private "related" vmas, to scan if
  13. * an anonymous page pointing to this anon_vma needs to be unmapped:
  14. * the vmas on the list will be related by forking, or by splitting.
  15. *
  16. * Since vmas come and go as they are split and merged (particularly
  17. * in mprotect), the mapping field of an anonymous page cannot point
  18. * directly to a vma: instead it points to an anon_vma, on whose list
  19. * the related vmas can be easily linked or unlinked.
  20. *
  21. * After unlinking the last vma on the list, we must garbage collect
  22. * the anon_vma object itself: we're guaranteed no page can be
  23. * pointing to this anon_vma once its vma list is empty.
  24. */
  25. struct anon_vma {
  26. struct anon_vma *root; /* Root of this anon_vma tree */
  27. struct rw_semaphore rwsem; /* W: modification, R: walking the list */
  28. /*
  29. * The refcount is taken on an anon_vma when there is no
  30. * guarantee that the vma of page tables will exist for
  31. * the duration of the operation. A caller that takes
  32. * the reference is responsible for clearing up the
  33. * anon_vma if they are the last user on release
  34. */
  35. atomic_t refcount;
  36. /*
  37. * Count of child anon_vmas and VMAs which points to this anon_vma.
  38. *
  39. * This counter is used for making decision about reusing anon_vma
  40. * instead of forking new one. See comments in function anon_vma_clone.
  41. */
  42. unsigned degree;
  43. struct anon_vma *parent; /* Parent of this anon_vma */
  44. /*
  45. * NOTE: the LSB of the rb_root.rb_node is set by
  46. * mm_take_all_locks() _after_ taking the above lock. So the
  47. * rb_root must only be read/written after taking the above lock
  48. * to be sure to see a valid next pointer. The LSB bit itself
  49. * is serialized by a system wide lock only visible to
  50. * mm_take_all_locks() (mm_all_locks_mutex).
  51. */
  52. struct rb_root rb_root; /* Interval tree of private "related" vmas */
  53. };
  54. /*
  55. * The copy-on-write semantics of fork mean that an anon_vma
  56. * can become associated with multiple processes. Furthermore,
  57. * each child process will have its own anon_vma, where new
  58. * pages for that process are instantiated.
  59. *
  60. * This structure allows us to find the anon_vmas associated
  61. * with a VMA, or the VMAs associated with an anon_vma.
  62. * The "same_vma" list contains the anon_vma_chains linking
  63. * all the anon_vmas associated with this VMA.
  64. * The "rb" field indexes on an interval tree the anon_vma_chains
  65. * which link all the VMAs associated with this anon_vma.
  66. */
  67. struct anon_vma_chain {
  68. struct vm_area_struct *vma;
  69. struct anon_vma *anon_vma;
  70. struct list_head same_vma; /* locked by mmap_sem & page_table_lock */
  71. struct rb_node rb; /* locked by anon_vma->rwsem */
  72. unsigned long rb_subtree_last;
  73. #ifdef CONFIG_DEBUG_VM_RB
  74. unsigned long cached_vma_start, cached_vma_last;
  75. #endif
  76. };
  77. enum ttu_flags {
  78. TTU_UNMAP = 1, /* unmap mode */
  79. TTU_MIGRATION = 2, /* migration mode */
  80. TTU_MUNLOCK = 4, /* munlock mode */
  81. TTU_IGNORE_MLOCK = (1 << 8), /* ignore mlock */
  82. TTU_IGNORE_ACCESS = (1 << 9), /* don't age */
  83. TTU_IGNORE_HWPOISON = (1 << 10),/* corrupted page is recoverable */
  84. };
  85. #ifdef CONFIG_MMU
  86. static inline void get_anon_vma(struct anon_vma *anon_vma)
  87. {
  88. atomic_inc(&anon_vma->refcount);
  89. }
  90. void __put_anon_vma(struct anon_vma *anon_vma);
  91. static inline void put_anon_vma(struct anon_vma *anon_vma)
  92. {
  93. if (atomic_dec_and_test(&anon_vma->refcount))
  94. __put_anon_vma(anon_vma);
  95. }
  96. static inline struct anon_vma *page_anon_vma(struct page *page)
  97. {
  98. if (((unsigned long)page->mapping & PAGE_MAPPING_FLAGS) !=
  99. PAGE_MAPPING_ANON)
  100. return NULL;
  101. return page_rmapping(page);
  102. }
  103. static inline void vma_lock_anon_vma(struct vm_area_struct *vma)
  104. {
  105. struct anon_vma *anon_vma = vma->anon_vma;
  106. if (anon_vma)
  107. down_write(&anon_vma->root->rwsem);
  108. }
  109. static inline void vma_unlock_anon_vma(struct vm_area_struct *vma)
  110. {
  111. struct anon_vma *anon_vma = vma->anon_vma;
  112. if (anon_vma)
  113. up_write(&anon_vma->root->rwsem);
  114. }
  115. static inline void anon_vma_lock_write(struct anon_vma *anon_vma)
  116. {
  117. down_write(&anon_vma->root->rwsem);
  118. }
  119. static inline void anon_vma_unlock_write(struct anon_vma *anon_vma)
  120. {
  121. up_write(&anon_vma->root->rwsem);
  122. }
  123. static inline void anon_vma_lock_read(struct anon_vma *anon_vma)
  124. {
  125. down_read(&anon_vma->root->rwsem);
  126. }
  127. static inline void anon_vma_unlock_read(struct anon_vma *anon_vma)
  128. {
  129. up_read(&anon_vma->root->rwsem);
  130. }
  131. /*
  132. * anon_vma helper functions.
  133. */
  134. void anon_vma_init(void); /* create anon_vma_cachep */
  135. int anon_vma_prepare(struct vm_area_struct *);
  136. void unlink_anon_vmas(struct vm_area_struct *);
  137. int anon_vma_clone(struct vm_area_struct *, struct vm_area_struct *);
  138. int anon_vma_fork(struct vm_area_struct *, struct vm_area_struct *);
  139. static inline void anon_vma_merge(struct vm_area_struct *vma,
  140. struct vm_area_struct *next)
  141. {
  142. VM_BUG_ON_VMA(vma->anon_vma != next->anon_vma, vma);
  143. unlink_anon_vmas(next);
  144. }
  145. struct anon_vma *page_get_anon_vma(struct page *page);
  146. /*
  147. * rmap interfaces called when adding or removing pte of page
  148. */
  149. void page_move_anon_rmap(struct page *, struct vm_area_struct *, unsigned long);
  150. void page_add_anon_rmap(struct page *, struct vm_area_struct *, unsigned long);
  151. void do_page_add_anon_rmap(struct page *, struct vm_area_struct *,
  152. unsigned long, int);
  153. void page_add_new_anon_rmap(struct page *, struct vm_area_struct *, unsigned long);
  154. void page_add_file_rmap(struct page *);
  155. void page_remove_rmap(struct page *);
  156. void hugepage_add_anon_rmap(struct page *, struct vm_area_struct *,
  157. unsigned long);
  158. void hugepage_add_new_anon_rmap(struct page *, struct vm_area_struct *,
  159. unsigned long);
  160. static inline void page_dup_rmap(struct page *page)
  161. {
  162. atomic_inc(&page->_mapcount);
  163. }
  164. /*
  165. * Called from mm/vmscan.c to handle paging out
  166. */
  167. int page_referenced(struct page *, int is_locked,
  168. struct mem_cgroup *memcg, unsigned long *vm_flags);
  169. #define TTU_ACTION(x) ((x) & TTU_ACTION_MASK)
  170. int try_to_unmap(struct page *, enum ttu_flags flags);
  171. /*
  172. * Called from mm/filemap_xip.c to unmap empty zero page
  173. */
  174. pte_t *__page_check_address(struct page *, struct mm_struct *,
  175. unsigned long, spinlock_t **, int);
  176. static inline pte_t *page_check_address(struct page *page, struct mm_struct *mm,
  177. unsigned long address,
  178. spinlock_t **ptlp, int sync)
  179. {
  180. pte_t *ptep;
  181. __cond_lock(*ptlp, ptep = __page_check_address(page, mm, address,
  182. ptlp, sync));
  183. return ptep;
  184. }
  185. /*
  186. * Used by swapoff to help locate where page is expected in vma.
  187. */
  188. unsigned long page_address_in_vma(struct page *, struct vm_area_struct *);
  189. /*
  190. * Cleans the PTEs of shared mappings.
  191. * (and since clean PTEs should also be readonly, write protects them too)
  192. *
  193. * returns the number of cleaned PTEs.
  194. */
  195. int page_mkclean(struct page *);
  196. /*
  197. * called in munlock()/munmap() path to check for other vmas holding
  198. * the page mlocked.
  199. */
  200. int try_to_munlock(struct page *);
  201. /*
  202. * Called by memory-failure.c to kill processes.
  203. */
  204. struct anon_vma *page_lock_anon_vma_read(struct page *page);
  205. void page_unlock_anon_vma_read(struct anon_vma *anon_vma);
  206. int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma);
  207. /*
  208. * rmap_walk_control: To control rmap traversing for specific needs
  209. *
  210. * arg: passed to rmap_one() and invalid_vma()
  211. * rmap_one: executed on each vma where page is mapped
  212. * done: for checking traversing termination condition
  213. * anon_lock: for getting anon_lock by optimized way rather than default
  214. * invalid_vma: for skipping uninterested vma
  215. */
  216. struct rmap_walk_control {
  217. void *arg;
  218. int (*rmap_one)(struct page *page, struct vm_area_struct *vma,
  219. unsigned long addr, void *arg);
  220. int (*done)(struct page *page);
  221. struct anon_vma *(*anon_lock)(struct page *page);
  222. bool (*invalid_vma)(struct vm_area_struct *vma, void *arg);
  223. };
  224. int rmap_walk(struct page *page, struct rmap_walk_control *rwc);
  225. #else /* !CONFIG_MMU */
  226. #define anon_vma_init() do {} while (0)
  227. #define anon_vma_prepare(vma) (0)
  228. #define anon_vma_link(vma) do {} while (0)
  229. static inline int page_referenced(struct page *page, int is_locked,
  230. struct mem_cgroup *memcg,
  231. unsigned long *vm_flags)
  232. {
  233. *vm_flags = 0;
  234. return 0;
  235. }
  236. #define try_to_unmap(page, refs) SWAP_FAIL
  237. static inline int page_mkclean(struct page *page)
  238. {
  239. return 0;
  240. }
  241. #endif /* CONFIG_MMU */
  242. /*
  243. * Return values of try_to_unmap
  244. */
  245. #define SWAP_SUCCESS 0
  246. #define SWAP_AGAIN 1
  247. #define SWAP_FAIL 2
  248. #define SWAP_MLOCK 3
  249. #endif /* _LINUX_RMAP_H */