fsnotify_backend.h 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418
  1. /*
  2. * Filesystem access notification for Linux
  3. *
  4. * Copyright (C) 2008 Red Hat, Inc., Eric Paris <eparis@redhat.com>
  5. */
  6. #ifndef __LINUX_FSNOTIFY_BACKEND_H
  7. #define __LINUX_FSNOTIFY_BACKEND_H
  8. #ifdef __KERNEL__
  9. #include <linux/idr.h> /* inotify uses this */
  10. #include <linux/fs.h> /* struct inode */
  11. #include <linux/list.h>
  12. #include <linux/path.h> /* struct path */
  13. #include <linux/spinlock.h>
  14. #include <linux/types.h>
  15. #include <linux/atomic.h>
  16. #include <linux/user_namespace.h>
  17. /*
  18. * IN_* from inotfy.h lines up EXACTLY with FS_*, this is so we can easily
  19. * convert between them. dnotify only needs conversion at watch creation
  20. * so no perf loss there. fanotify isn't defined yet, so it can use the
  21. * wholes if it needs more events.
  22. */
  23. #define FS_ACCESS 0x00000001 /* File was accessed */
  24. #define FS_MODIFY 0x00000002 /* File was modified */
  25. #define FS_ATTRIB 0x00000004 /* Metadata changed */
  26. #define FS_CLOSE_WRITE 0x00000008 /* Writtable file was closed */
  27. #define FS_CLOSE_NOWRITE 0x00000010 /* Unwrittable file closed */
  28. #define FS_OPEN 0x00000020 /* File was opened */
  29. #define FS_MOVED_FROM 0x00000040 /* File was moved from X */
  30. #define FS_MOVED_TO 0x00000080 /* File was moved to Y */
  31. #define FS_CREATE 0x00000100 /* Subfile was created */
  32. #define FS_DELETE 0x00000200 /* Subfile was deleted */
  33. #define FS_DELETE_SELF 0x00000400 /* Self was deleted */
  34. #define FS_MOVE_SELF 0x00000800 /* Self was moved */
  35. #define FS_UNMOUNT 0x00002000 /* inode on umount fs */
  36. #define FS_Q_OVERFLOW 0x00004000 /* Event queued overflowed */
  37. #define FS_IN_IGNORED 0x00008000 /* last inotify event here */
  38. #define FS_OPEN_PERM 0x00010000 /* open event in an permission hook */
  39. #define FS_ACCESS_PERM 0x00020000 /* access event in a permissions hook */
  40. #define FS_EXCL_UNLINK 0x04000000 /* do not send events if object is unlinked */
  41. #define FS_ISDIR 0x40000000 /* event occurred against dir */
  42. #define FS_IN_ONESHOT 0x80000000 /* only send event once */
  43. #define FS_DN_RENAME 0x10000000 /* file renamed */
  44. #define FS_DN_MULTISHOT 0x20000000 /* dnotify multishot */
  45. /* This inode cares about things that happen to its children. Always set for
  46. * dnotify and inotify. */
  47. #define FS_EVENT_ON_CHILD 0x08000000
  48. /* This is a list of all events that may get sent to a parernt based on fs event
  49. * happening to inodes inside that directory */
  50. #define FS_EVENTS_POSS_ON_CHILD (FS_ACCESS | FS_MODIFY | FS_ATTRIB |\
  51. FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN |\
  52. FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE |\
  53. FS_DELETE | FS_OPEN_PERM | FS_ACCESS_PERM)
  54. #define FS_MOVE (FS_MOVED_FROM | FS_MOVED_TO)
  55. #define ALL_FSNOTIFY_PERM_EVENTS (FS_OPEN_PERM | FS_ACCESS_PERM)
  56. #define ALL_FSNOTIFY_EVENTS (FS_ACCESS | FS_MODIFY | FS_ATTRIB | \
  57. FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN | \
  58. FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE | \
  59. FS_DELETE | FS_DELETE_SELF | FS_MOVE_SELF | \
  60. FS_UNMOUNT | FS_Q_OVERFLOW | FS_IN_IGNORED | \
  61. FS_OPEN_PERM | FS_ACCESS_PERM | FS_EXCL_UNLINK | \
  62. FS_ISDIR | FS_IN_ONESHOT | FS_DN_RENAME | \
  63. FS_DN_MULTISHOT | FS_EVENT_ON_CHILD)
  64. struct fsnotify_group;
  65. struct fsnotify_event;
  66. struct fsnotify_mark;
  67. struct fsnotify_event_private_data;
  68. struct fsnotify_fname;
  69. struct fsnotify_iter_info;
  70. /*
  71. * Each group much define these ops. The fsnotify infrastructure will call
  72. * these operations for each relevant group.
  73. *
  74. * handle_event - main call for a group to handle an fs event
  75. * free_group_priv - called when a group refcnt hits 0 to clean up the private union
  76. * freeing_mark - called when a mark is being destroyed for some reason. The group
  77. * MUST be holding a reference on each mark and that reference must be
  78. * dropped in this function. inotify uses this function to send
  79. * userspace messages that marks have been removed.
  80. */
  81. struct fsnotify_ops {
  82. int (*handle_event)(struct fsnotify_group *group,
  83. struct inode *inode,
  84. struct fsnotify_mark *inode_mark,
  85. struct fsnotify_mark *vfsmount_mark,
  86. u32 mask, const void *data, int data_type,
  87. const unsigned char *file_name, u32 cookie,
  88. struct fsnotify_iter_info *iter_info);
  89. void (*free_group_priv)(struct fsnotify_group *group);
  90. void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group);
  91. void (*free_event)(struct fsnotify_event *event);
  92. /* called on final put+free to free memory */
  93. void (*free_mark)(struct fsnotify_mark *mark);
  94. };
  95. /*
  96. * all of the information about the original object we want to now send to
  97. * a group. If you want to carry more info from the accessing task to the
  98. * listener this structure is where you need to be adding fields.
  99. */
  100. struct fsnotify_event {
  101. struct list_head list;
  102. /* inode may ONLY be dereferenced during handle_event(). */
  103. struct inode *inode; /* either the inode the event happened to or its parent */
  104. u32 mask; /* the type of access, bitwise OR for FS_* event types */
  105. };
  106. /*
  107. * A group is a "thing" that wants to receive notification about filesystem
  108. * events. The mask holds the subset of event types this group cares about.
  109. * refcnt on a group is up to the implementor and at any moment if it goes 0
  110. * everything will be cleaned up.
  111. */
  112. struct fsnotify_group {
  113. /*
  114. * How the refcnt is used is up to each group. When the refcnt hits 0
  115. * fsnotify will clean up all of the resources associated with this group.
  116. * As an example, the dnotify group will always have a refcnt=1 and that
  117. * will never change. Inotify, on the other hand, has a group per
  118. * inotify_init() and the refcnt will hit 0 only when that fd has been
  119. * closed.
  120. */
  121. atomic_t refcnt; /* things with interest in this group */
  122. const struct fsnotify_ops *ops; /* how this group handles things */
  123. /* needed to send notification to userspace */
  124. spinlock_t notification_lock; /* protect the notification_list */
  125. struct list_head notification_list; /* list of event_holder this group needs to send to userspace */
  126. wait_queue_head_t notification_waitq; /* read() on the notification file blocks on this waitq */
  127. unsigned int q_len; /* events on the queue */
  128. unsigned int max_events; /* maximum events allowed on the list */
  129. /*
  130. * Valid fsnotify group priorities. Events are send in order from highest
  131. * priority to lowest priority. We default to the lowest priority.
  132. */
  133. #define FS_PRIO_0 0 /* normal notifiers, no permissions */
  134. #define FS_PRIO_1 1 /* fanotify content based access control */
  135. #define FS_PRIO_2 2 /* fanotify pre-content access */
  136. unsigned int priority;
  137. bool shutdown; /* group is being shut down, don't queue more events */
  138. /* stores all fastpath marks assoc with this group so they can be cleaned on unregister */
  139. struct mutex mark_mutex; /* protect marks_list */
  140. atomic_t num_marks; /* 1 for each mark and 1 for not being
  141. * past the point of no return when freeing
  142. * a group */
  143. struct list_head marks_list; /* all inode marks for this group */
  144. struct fasync_struct *fsn_fa; /* async notification */
  145. struct fsnotify_event *overflow_event; /* Event we queue when the
  146. * notification list is too
  147. * full */
  148. atomic_t user_waits; /* Number of tasks waiting for user
  149. * response */
  150. /* groups can define private fields here or use the void *private */
  151. union {
  152. void *private;
  153. #ifdef CONFIG_INOTIFY_USER
  154. struct inotify_group_private_data {
  155. spinlock_t idr_lock;
  156. struct idr idr;
  157. struct ucounts *ucounts;
  158. } inotify_data;
  159. #endif
  160. #ifdef CONFIG_FANOTIFY
  161. struct fanotify_group_private_data {
  162. #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
  163. /* allows a group to block waiting for a userspace response */
  164. struct list_head access_list;
  165. wait_queue_head_t access_waitq;
  166. #endif /* CONFIG_FANOTIFY_ACCESS_PERMISSIONS */
  167. int f_flags;
  168. unsigned int max_marks;
  169. struct user_struct *user;
  170. } fanotify_data;
  171. #endif /* CONFIG_FANOTIFY */
  172. };
  173. };
  174. /* when calling fsnotify tell it if the data is a path or inode */
  175. #define FSNOTIFY_EVENT_NONE 0
  176. #define FSNOTIFY_EVENT_PATH 1
  177. #define FSNOTIFY_EVENT_INODE 2
  178. /*
  179. * Inode / vfsmount point to this structure which tracks all marks attached to
  180. * the inode / vfsmount. The reference to inode / vfsmount is held by this
  181. * structure. We destroy this structure when there are no more marks attached
  182. * to it. The structure is protected by fsnotify_mark_srcu.
  183. */
  184. struct fsnotify_mark_connector {
  185. spinlock_t lock;
  186. #define FSNOTIFY_OBJ_TYPE_INODE 0x01
  187. #define FSNOTIFY_OBJ_TYPE_VFSMOUNT 0x02
  188. #define FSNOTIFY_OBJ_ALL_TYPES (FSNOTIFY_OBJ_TYPE_INODE | \
  189. FSNOTIFY_OBJ_TYPE_VFSMOUNT)
  190. unsigned int flags; /* Type of object [lock] */
  191. union { /* Object pointer [lock] */
  192. struct inode *inode;
  193. struct vfsmount *mnt;
  194. };
  195. union {
  196. struct hlist_head list;
  197. /* Used listing heads to free after srcu period expires */
  198. struct fsnotify_mark_connector *destroy_next;
  199. };
  200. };
  201. /*
  202. * A mark is simply an object attached to an in core inode which allows an
  203. * fsnotify listener to indicate they are either no longer interested in events
  204. * of a type matching mask or only interested in those events.
  205. *
  206. * These are flushed when an inode is evicted from core and may be flushed
  207. * when the inode is modified (as seen by fsnotify_access). Some fsnotify
  208. * users (such as dnotify) will flush these when the open fd is closed and not
  209. * at inode eviction or modification.
  210. *
  211. * Text in brackets is showing the lock(s) protecting modifications of a
  212. * particular entry. obj_lock means either inode->i_lock or
  213. * mnt->mnt_root->d_lock depending on the mark type.
  214. */
  215. struct fsnotify_mark {
  216. /* Mask this mark is for [mark->lock, group->mark_mutex] */
  217. __u32 mask;
  218. /* We hold one for presence in g_list. Also one ref for each 'thing'
  219. * in kernel that found and may be using this mark. */
  220. atomic_t refcnt;
  221. /* Group this mark is for. Set on mark creation, stable until last ref
  222. * is dropped */
  223. struct fsnotify_group *group;
  224. /* List of marks by group->i_fsnotify_marks. Also reused for queueing
  225. * mark into destroy_list when it's waiting for the end of SRCU period
  226. * before it can be freed. [group->mark_mutex] */
  227. struct list_head g_list;
  228. /* Protects inode / mnt pointers, flags, masks */
  229. spinlock_t lock;
  230. /* List of marks for inode / vfsmount [connector->lock, mark ref] */
  231. struct hlist_node obj_list;
  232. /* Head of list of marks for an object [mark ref] */
  233. struct fsnotify_mark_connector *connector;
  234. /* Events types to ignore [mark->lock, group->mark_mutex] */
  235. __u32 ignored_mask;
  236. #define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY 0x01
  237. #define FSNOTIFY_MARK_FLAG_ALIVE 0x02
  238. #define FSNOTIFY_MARK_FLAG_ATTACHED 0x04
  239. unsigned int flags; /* flags [mark->lock] */
  240. };
  241. #ifdef CONFIG_FSNOTIFY
  242. /* called from the vfs helpers */
  243. /* main fsnotify call to send events */
  244. extern int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
  245. const unsigned char *name, u32 cookie);
  246. extern int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask);
  247. extern void __fsnotify_inode_delete(struct inode *inode);
  248. extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
  249. extern u32 fsnotify_get_cookie(void);
  250. static inline int fsnotify_inode_watches_children(struct inode *inode)
  251. {
  252. /* FS_EVENT_ON_CHILD is set if the inode may care */
  253. if (!(inode->i_fsnotify_mask & FS_EVENT_ON_CHILD))
  254. return 0;
  255. /* this inode might care about child events, does it care about the
  256. * specific set of events that can happen on a child? */
  257. return inode->i_fsnotify_mask & FS_EVENTS_POSS_ON_CHILD;
  258. }
  259. /*
  260. * Update the dentry with a flag indicating the interest of its parent to receive
  261. * filesystem events when those events happens to this dentry->d_inode.
  262. */
  263. static inline void fsnotify_update_flags(struct dentry *dentry)
  264. {
  265. assert_spin_locked(&dentry->d_lock);
  266. /*
  267. * Serialisation of setting PARENT_WATCHED on the dentries is provided
  268. * by d_lock. If inotify_inode_watched changes after we have taken
  269. * d_lock, the following __fsnotify_update_child_dentry_flags call will
  270. * find our entry, so it will spin until we complete here, and update
  271. * us with the new state.
  272. */
  273. if (fsnotify_inode_watches_children(dentry->d_parent->d_inode))
  274. dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
  275. else
  276. dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
  277. }
  278. /* called from fsnotify listeners, such as fanotify or dnotify */
  279. /* create a new group */
  280. extern struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops);
  281. /* get reference to a group */
  282. extern void fsnotify_get_group(struct fsnotify_group *group);
  283. /* drop reference on a group from fsnotify_alloc_group */
  284. extern void fsnotify_put_group(struct fsnotify_group *group);
  285. /* group destruction begins, stop queuing new events */
  286. extern void fsnotify_group_stop_queueing(struct fsnotify_group *group);
  287. /* destroy group */
  288. extern void fsnotify_destroy_group(struct fsnotify_group *group);
  289. /* fasync handler function */
  290. extern int fsnotify_fasync(int fd, struct file *file, int on);
  291. /* Free event from memory */
  292. extern void fsnotify_destroy_event(struct fsnotify_group *group,
  293. struct fsnotify_event *event);
  294. /* attach the event to the group notification queue */
  295. extern int fsnotify_add_event(struct fsnotify_group *group,
  296. struct fsnotify_event *event,
  297. int (*merge)(struct list_head *,
  298. struct fsnotify_event *));
  299. /* true if the group notification queue is empty */
  300. extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group);
  301. /* return, but do not dequeue the first event on the notification queue */
  302. extern struct fsnotify_event *fsnotify_peek_first_event(struct fsnotify_group *group);
  303. /* return AND dequeue the first event on the notification queue */
  304. extern struct fsnotify_event *fsnotify_remove_first_event(struct fsnotify_group *group);
  305. /* functions used to manipulate the marks attached to inodes */
  306. /* Calculate mask of events for a list of marks */
  307. extern void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn);
  308. extern void fsnotify_init_mark(struct fsnotify_mark *mark,
  309. struct fsnotify_group *group);
  310. /* Find mark belonging to given group in the list of marks */
  311. extern struct fsnotify_mark *fsnotify_find_mark(
  312. struct fsnotify_mark_connector __rcu **connp,
  313. struct fsnotify_group *group);
  314. /* attach the mark to the inode or vfsmount */
  315. extern int fsnotify_add_mark(struct fsnotify_mark *mark, struct inode *inode,
  316. struct vfsmount *mnt, int allow_dups);
  317. extern int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
  318. struct inode *inode, struct vfsmount *mnt, int allow_dups);
  319. /* given a group and a mark, flag mark to be freed when all references are dropped */
  320. extern void fsnotify_destroy_mark(struct fsnotify_mark *mark,
  321. struct fsnotify_group *group);
  322. /* detach mark from inode / mount list, group list, drop inode reference */
  323. extern void fsnotify_detach_mark(struct fsnotify_mark *mark);
  324. /* free mark */
  325. extern void fsnotify_free_mark(struct fsnotify_mark *mark);
  326. /* run all the marks in a group, and clear all of the marks attached to given object type */
  327. extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group, unsigned int type);
  328. /* run all the marks in a group, and clear all of the vfsmount marks */
  329. static inline void fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group *group)
  330. {
  331. fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_VFSMOUNT);
  332. }
  333. /* run all the marks in a group, and clear all of the inode marks */
  334. static inline void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group)
  335. {
  336. fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_INODE);
  337. }
  338. extern void fsnotify_get_mark(struct fsnotify_mark *mark);
  339. extern void fsnotify_put_mark(struct fsnotify_mark *mark);
  340. extern void fsnotify_unmount_inodes(struct super_block *sb);
  341. extern void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info);
  342. extern bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info);
  343. /* put here because inotify does some weird stuff when destroying watches */
  344. extern void fsnotify_init_event(struct fsnotify_event *event,
  345. struct inode *to_tell, u32 mask);
  346. #else
  347. static inline int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
  348. const unsigned char *name, u32 cookie)
  349. {
  350. return 0;
  351. }
  352. static inline int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask)
  353. {
  354. return 0;
  355. }
  356. static inline void __fsnotify_inode_delete(struct inode *inode)
  357. {}
  358. static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
  359. {}
  360. static inline void fsnotify_update_flags(struct dentry *dentry)
  361. {}
  362. static inline u32 fsnotify_get_cookie(void)
  363. {
  364. return 0;
  365. }
  366. static inline void fsnotify_unmount_inodes(struct super_block *sb)
  367. {}
  368. #endif /* CONFIG_FSNOTIFY */
  369. #endif /* __KERNEL __ */
  370. #endif /* __LINUX_FSNOTIFY_BACKEND_H */