1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Filesystem access notification for Linux
8 #ifndef __LINUX_FSNOTIFY_BACKEND_H
9 #define __LINUX_FSNOTIFY_BACKEND_H
13 #include <linux/idr.h> /* inotify uses this */
14 #include <linux/fs.h> /* struct inode */
15 #include <linux/list.h>
16 #include <linux/path.h> /* struct path */
17 #include <linux/spinlock.h>
18 #include <linux/types.h>
19 #include <linux/atomic.h>
20 #include <linux/user_namespace.h>
21 #include <linux/refcount.h>
24 * IN_* from inotfy.h lines up EXACTLY with FS_*, this is so we can easily
25 * convert between them. dnotify only needs conversion at watch creation
26 * so no perf loss there. fanotify isn't defined yet, so it can use the
27 * wholes if it needs more events.
29 #define FS_ACCESS 0x00000001 /* File was accessed */
30 #define FS_MODIFY 0x00000002 /* File was modified */
31 #define FS_ATTRIB 0x00000004 /* Metadata changed */
32 #define FS_CLOSE_WRITE 0x00000008 /* Writtable file was closed */
33 #define FS_CLOSE_NOWRITE 0x00000010 /* Unwrittable file closed */
34 #define FS_OPEN 0x00000020 /* File was opened */
35 #define FS_MOVED_FROM 0x00000040 /* File was moved from X */
36 #define FS_MOVED_TO 0x00000080 /* File was moved to Y */
37 #define FS_CREATE 0x00000100 /* Subfile was created */
38 #define FS_DELETE 0x00000200 /* Subfile was deleted */
39 #define FS_DELETE_SELF 0x00000400 /* Self was deleted */
40 #define FS_MOVE_SELF 0x00000800 /* Self was moved */
41 #define FS_OPEN_EXEC 0x00001000 /* File was opened for exec */
43 #define FS_UNMOUNT 0x00002000 /* inode on umount fs */
44 #define FS_Q_OVERFLOW 0x00004000 /* Event queued overflowed */
45 #define FS_IN_IGNORED 0x00008000 /* last inotify event here */
47 #define FS_OPEN_PERM 0x00010000 /* open event in an permission hook */
48 #define FS_ACCESS_PERM 0x00020000 /* access event in a permissions hook */
49 #define FS_OPEN_EXEC_PERM 0x00040000 /* open/exec event in a permission hook */
51 #define FS_EXCL_UNLINK 0x04000000 /* do not send events if object is unlinked */
52 #define FS_ISDIR 0x40000000 /* event occurred against dir */
53 #define FS_IN_ONESHOT 0x80000000 /* only send event once */
55 #define FS_DN_RENAME 0x10000000 /* file renamed */
56 #define FS_DN_MULTISHOT 0x20000000 /* dnotify multishot */
58 /* This inode cares about things that happen to its children. Always set for
59 * dnotify and inotify. */
60 #define FS_EVENT_ON_CHILD 0x08000000
62 /* This is a list of all events that may get sent to a parernt based on fs event
63 * happening to inodes inside that directory */
64 #define FS_EVENTS_POSS_ON_CHILD (FS_ACCESS | FS_MODIFY | FS_ATTRIB |\
65 FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN |\
66 FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE |\
67 FS_DELETE | FS_OPEN_PERM | FS_ACCESS_PERM | \
68 FS_OPEN_EXEC | FS_OPEN_EXEC_PERM)
70 #define FS_MOVE (FS_MOVED_FROM | FS_MOVED_TO)
72 #define ALL_FSNOTIFY_PERM_EVENTS (FS_OPEN_PERM | FS_ACCESS_PERM | \
75 /* Events that can be reported to backends */
76 #define ALL_FSNOTIFY_EVENTS (FS_ACCESS | FS_MODIFY | FS_ATTRIB | \
77 FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN | \
78 FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE | \
79 FS_DELETE | FS_DELETE_SELF | FS_MOVE_SELF | \
80 FS_UNMOUNT | FS_Q_OVERFLOW | FS_IN_IGNORED | \
81 FS_OPEN_PERM | FS_ACCESS_PERM | FS_DN_RENAME | \
82 FS_OPEN_EXEC | FS_OPEN_EXEC_PERM)
84 /* Extra flags that may be reported with event or control handling of events */
85 #define ALL_FSNOTIFY_FLAGS (FS_EXCL_UNLINK | FS_ISDIR | FS_IN_ONESHOT | \
86 FS_DN_MULTISHOT | FS_EVENT_ON_CHILD)
88 #define ALL_FSNOTIFY_BITS (ALL_FSNOTIFY_EVENTS | ALL_FSNOTIFY_FLAGS)
90 struct fsnotify_group;
91 struct fsnotify_event;
93 struct fsnotify_event_private_data;
94 struct fsnotify_fname;
95 struct fsnotify_iter_info;
100 * Each group much define these ops. The fsnotify infrastructure will call
101 * these operations for each relevant group.
103 * handle_event - main call for a group to handle an fs event
104 * free_group_priv - called when a group refcnt hits 0 to clean up the private union
105 * freeing_mark - called when a mark is being destroyed for some reason. The group
106 * MUST be holding a reference on each mark and that reference must be
107 * dropped in this function. inotify uses this function to send
108 * userspace messages that marks have been removed.
110 struct fsnotify_ops {
111 int (*handle_event)(struct fsnotify_group *group,
113 u32 mask, const void *data, int data_type,
114 const unsigned char *file_name, u32 cookie,
115 struct fsnotify_iter_info *iter_info);
116 void (*free_group_priv)(struct fsnotify_group *group);
117 void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group);
118 void (*free_event)(struct fsnotify_event *event);
119 /* called on final put+free to free memory */
120 void (*free_mark)(struct fsnotify_mark *mark);
124 * all of the information about the original object we want to now send to
125 * a group. If you want to carry more info from the accessing task to the
126 * listener this structure is where you need to be adding fields.
128 struct fsnotify_event {
129 struct list_head list;
130 /* inode may ONLY be dereferenced during handle_event(). */
131 struct inode *inode; /* either the inode the event happened to or its parent */
132 u32 mask; /* the type of access, bitwise OR for FS_* event types */
136 * A group is a "thing" that wants to receive notification about filesystem
137 * events. The mask holds the subset of event types this group cares about.
138 * refcnt on a group is up to the implementor and at any moment if it goes 0
139 * everything will be cleaned up.
141 struct fsnotify_group {
142 const struct fsnotify_ops *ops; /* how this group handles things */
145 * How the refcnt is used is up to each group. When the refcnt hits 0
146 * fsnotify will clean up all of the resources associated with this group.
147 * As an example, the dnotify group will always have a refcnt=1 and that
148 * will never change. Inotify, on the other hand, has a group per
149 * inotify_init() and the refcnt will hit 0 only when that fd has been
152 refcount_t refcnt; /* things with interest in this group */
154 /* needed to send notification to userspace */
155 spinlock_t notification_lock; /* protect the notification_list */
156 struct list_head notification_list; /* list of event_holder this group needs to send to userspace */
157 wait_queue_head_t notification_waitq; /* read() on the notification file blocks on this waitq */
158 unsigned int q_len; /* events on the queue */
159 unsigned int max_events; /* maximum events allowed on the list */
161 * Valid fsnotify group priorities. Events are send in order from highest
162 * priority to lowest priority. We default to the lowest priority.
164 #define FS_PRIO_0 0 /* normal notifiers, no permissions */
165 #define FS_PRIO_1 1 /* fanotify content based access control */
166 #define FS_PRIO_2 2 /* fanotify pre-content access */
167 unsigned int priority;
168 bool shutdown; /* group is being shut down, don't queue more events */
170 /* stores all fastpath marks assoc with this group so they can be cleaned on unregister */
171 struct mutex mark_mutex; /* protect marks_list */
172 atomic_t num_marks; /* 1 for each mark and 1 for not being
173 * past the point of no return when freeing
175 atomic_t user_waits; /* Number of tasks waiting for user
177 struct list_head marks_list; /* all inode marks for this group */
179 struct fasync_struct *fsn_fa; /* async notification */
181 struct fsnotify_event *overflow_event; /* Event we queue when the
182 * notification list is too
185 struct mem_cgroup *memcg; /* memcg to charge allocations */
187 /* groups can define private fields here or use the void *private */
190 #ifdef CONFIG_INOTIFY_USER
191 struct inotify_group_private_data {
194 struct ucounts *ucounts;
197 #ifdef CONFIG_FANOTIFY
198 struct fanotify_group_private_data {
199 /* allows a group to block waiting for a userspace response */
200 struct list_head access_list;
201 wait_queue_head_t access_waitq;
202 int flags; /* flags from fanotify_init() */
203 int f_flags; /* event_f_flags from fanotify_init() */
204 unsigned int max_marks;
205 struct user_struct *user;
207 #endif /* CONFIG_FANOTIFY */
211 /* when calling fsnotify tell it if the data is a path or inode */
212 #define FSNOTIFY_EVENT_NONE 0
213 #define FSNOTIFY_EVENT_PATH 1
214 #define FSNOTIFY_EVENT_INODE 2
216 enum fsnotify_obj_type {
217 FSNOTIFY_OBJ_TYPE_INODE,
218 FSNOTIFY_OBJ_TYPE_VFSMOUNT,
219 FSNOTIFY_OBJ_TYPE_SB,
220 FSNOTIFY_OBJ_TYPE_COUNT,
221 FSNOTIFY_OBJ_TYPE_DETACHED = FSNOTIFY_OBJ_TYPE_COUNT
224 #define FSNOTIFY_OBJ_TYPE_INODE_FL (1U << FSNOTIFY_OBJ_TYPE_INODE)
225 #define FSNOTIFY_OBJ_TYPE_VFSMOUNT_FL (1U << FSNOTIFY_OBJ_TYPE_VFSMOUNT)
226 #define FSNOTIFY_OBJ_TYPE_SB_FL (1U << FSNOTIFY_OBJ_TYPE_SB)
227 #define FSNOTIFY_OBJ_ALL_TYPES_MASK ((1U << FSNOTIFY_OBJ_TYPE_COUNT) - 1)
229 static inline bool fsnotify_valid_obj_type(unsigned int type)
231 return (type < FSNOTIFY_OBJ_TYPE_COUNT);
234 struct fsnotify_iter_info {
235 struct fsnotify_mark *marks[FSNOTIFY_OBJ_TYPE_COUNT];
236 unsigned int report_mask;
240 static inline bool fsnotify_iter_should_report_type(
241 struct fsnotify_iter_info *iter_info, int type)
243 return (iter_info->report_mask & (1U << type));
246 static inline void fsnotify_iter_set_report_type(
247 struct fsnotify_iter_info *iter_info, int type)
249 iter_info->report_mask |= (1U << type);
252 static inline void fsnotify_iter_set_report_type_mark(
253 struct fsnotify_iter_info *iter_info, int type,
254 struct fsnotify_mark *mark)
256 iter_info->marks[type] = mark;
257 iter_info->report_mask |= (1U << type);
260 #define FSNOTIFY_ITER_FUNCS(name, NAME) \
261 static inline struct fsnotify_mark *fsnotify_iter_##name##_mark( \
262 struct fsnotify_iter_info *iter_info) \
264 return (iter_info->report_mask & FSNOTIFY_OBJ_TYPE_##NAME##_FL) ? \
265 iter_info->marks[FSNOTIFY_OBJ_TYPE_##NAME] : NULL; \
268 FSNOTIFY_ITER_FUNCS(inode, INODE)
269 FSNOTIFY_ITER_FUNCS(vfsmount, VFSMOUNT)
270 FSNOTIFY_ITER_FUNCS(sb, SB)
272 #define fsnotify_foreach_obj_type(type) \
273 for (type = 0; type < FSNOTIFY_OBJ_TYPE_COUNT; type++)
276 * fsnotify_connp_t is what we embed in objects which connector can be attached
277 * to. fsnotify_connp_t * is how we refer from connector back to object.
279 struct fsnotify_mark_connector;
280 typedef struct fsnotify_mark_connector __rcu *fsnotify_connp_t;
283 * Inode/vfsmount/sb point to this structure which tracks all marks attached to
284 * the inode/vfsmount/sb. The reference to inode/vfsmount/sb is held by this
285 * structure. We destroy this structure when there are no more marks attached
286 * to it. The structure is protected by fsnotify_mark_srcu.
288 struct fsnotify_mark_connector {
290 unsigned int type; /* Type of object [lock] */
292 /* Object pointer [lock] */
293 fsnotify_connp_t *obj;
294 /* Used listing heads to free after srcu period expires */
295 struct fsnotify_mark_connector *destroy_next;
297 struct hlist_head list;
301 * A mark is simply an object attached to an in core inode which allows an
302 * fsnotify listener to indicate they are either no longer interested in events
303 * of a type matching mask or only interested in those events.
305 * These are flushed when an inode is evicted from core and may be flushed
306 * when the inode is modified (as seen by fsnotify_access). Some fsnotify
307 * users (such as dnotify) will flush these when the open fd is closed and not
308 * at inode eviction or modification.
310 * Text in brackets is showing the lock(s) protecting modifications of a
311 * particular entry. obj_lock means either inode->i_lock or
312 * mnt->mnt_root->d_lock depending on the mark type.
314 struct fsnotify_mark {
315 /* Mask this mark is for [mark->lock, group->mark_mutex] */
317 /* We hold one for presence in g_list. Also one ref for each 'thing'
318 * in kernel that found and may be using this mark. */
320 /* Group this mark is for. Set on mark creation, stable until last ref
322 struct fsnotify_group *group;
323 /* List of marks by group->marks_list. Also reused for queueing
324 * mark into destroy_list when it's waiting for the end of SRCU period
325 * before it can be freed. [group->mark_mutex] */
326 struct list_head g_list;
327 /* Protects inode / mnt pointers, flags, masks */
329 /* List of marks for inode / vfsmount [connector->lock, mark ref] */
330 struct hlist_node obj_list;
331 /* Head of list of marks for an object [mark ref] */
332 struct fsnotify_mark_connector *connector;
333 /* Events types to ignore [mark->lock, group->mark_mutex] */
335 #define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY 0x01
336 #define FSNOTIFY_MARK_FLAG_ALIVE 0x02
337 #define FSNOTIFY_MARK_FLAG_ATTACHED 0x04
338 unsigned int flags; /* flags [mark->lock] */
341 #ifdef CONFIG_FSNOTIFY
343 /* called from the vfs helpers */
345 /* main fsnotify call to send events */
346 extern int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
347 const unsigned char *name, u32 cookie);
348 extern int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask);
349 extern void __fsnotify_inode_delete(struct inode *inode);
350 extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
351 extern void fsnotify_sb_delete(struct super_block *sb);
352 extern u32 fsnotify_get_cookie(void);
354 static inline int fsnotify_inode_watches_children(struct inode *inode)
356 /* FS_EVENT_ON_CHILD is set if the inode may care */
357 if (!(inode->i_fsnotify_mask & FS_EVENT_ON_CHILD))
359 /* this inode might care about child events, does it care about the
360 * specific set of events that can happen on a child? */
361 return inode->i_fsnotify_mask & FS_EVENTS_POSS_ON_CHILD;
365 * Update the dentry with a flag indicating the interest of its parent to receive
366 * filesystem events when those events happens to this dentry->d_inode.
368 static inline void fsnotify_update_flags(struct dentry *dentry)
370 assert_spin_locked(&dentry->d_lock);
373 * Serialisation of setting PARENT_WATCHED on the dentries is provided
374 * by d_lock. If inotify_inode_watched changes after we have taken
375 * d_lock, the following __fsnotify_update_child_dentry_flags call will
376 * find our entry, so it will spin until we complete here, and update
377 * us with the new state.
379 if (fsnotify_inode_watches_children(dentry->d_parent->d_inode))
380 dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
382 dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
385 /* called from fsnotify listeners, such as fanotify or dnotify */
387 /* create a new group */
388 extern struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops);
389 /* get reference to a group */
390 extern void fsnotify_get_group(struct fsnotify_group *group);
391 /* drop reference on a group from fsnotify_alloc_group */
392 extern void fsnotify_put_group(struct fsnotify_group *group);
393 /* group destruction begins, stop queuing new events */
394 extern void fsnotify_group_stop_queueing(struct fsnotify_group *group);
396 extern void fsnotify_destroy_group(struct fsnotify_group *group);
397 /* fasync handler function */
398 extern int fsnotify_fasync(int fd, struct file *file, int on);
399 /* Free event from memory */
400 extern void fsnotify_destroy_event(struct fsnotify_group *group,
401 struct fsnotify_event *event);
402 /* attach the event to the group notification queue */
403 extern int fsnotify_add_event(struct fsnotify_group *group,
404 struct fsnotify_event *event,
405 int (*merge)(struct list_head *,
406 struct fsnotify_event *));
407 /* Queue overflow event to a notification group */
408 static inline void fsnotify_queue_overflow(struct fsnotify_group *group)
410 fsnotify_add_event(group, group->overflow_event, NULL);
413 /* true if the group notification queue is empty */
414 extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group);
415 /* return, but do not dequeue the first event on the notification queue */
416 extern struct fsnotify_event *fsnotify_peek_first_event(struct fsnotify_group *group);
417 /* return AND dequeue the first event on the notification queue */
418 extern struct fsnotify_event *fsnotify_remove_first_event(struct fsnotify_group *group);
420 /* functions used to manipulate the marks attached to inodes */
422 /* Get mask of events for a list of marks */
423 extern __u32 fsnotify_conn_mask(struct fsnotify_mark_connector *conn);
424 /* Calculate mask of events for a list of marks */
425 extern void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn);
426 extern void fsnotify_init_mark(struct fsnotify_mark *mark,
427 struct fsnotify_group *group);
428 /* Find mark belonging to given group in the list of marks */
429 extern struct fsnotify_mark *fsnotify_find_mark(fsnotify_connp_t *connp,
430 struct fsnotify_group *group);
431 /* attach the mark to the object */
432 extern int fsnotify_add_mark(struct fsnotify_mark *mark,
433 fsnotify_connp_t *connp, unsigned int type,
435 extern int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
436 fsnotify_connp_t *connp, unsigned int type,
438 /* attach the mark to the inode */
439 static inline int fsnotify_add_inode_mark(struct fsnotify_mark *mark,
443 return fsnotify_add_mark(mark, &inode->i_fsnotify_marks,
444 FSNOTIFY_OBJ_TYPE_INODE, allow_dups);
446 static inline int fsnotify_add_inode_mark_locked(struct fsnotify_mark *mark,
450 return fsnotify_add_mark_locked(mark, &inode->i_fsnotify_marks,
451 FSNOTIFY_OBJ_TYPE_INODE, allow_dups);
453 /* given a group and a mark, flag mark to be freed when all references are dropped */
454 extern void fsnotify_destroy_mark(struct fsnotify_mark *mark,
455 struct fsnotify_group *group);
456 /* detach mark from inode / mount list, group list, drop inode reference */
457 extern void fsnotify_detach_mark(struct fsnotify_mark *mark);
459 extern void fsnotify_free_mark(struct fsnotify_mark *mark);
460 /* run all the marks in a group, and clear all of the marks attached to given object type */
461 extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group, unsigned int type);
462 /* run all the marks in a group, and clear all of the vfsmount marks */
463 static inline void fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group *group)
465 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_VFSMOUNT_FL);
467 /* run all the marks in a group, and clear all of the inode marks */
468 static inline void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group)
470 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_INODE_FL);
472 /* run all the marks in a group, and clear all of the sn marks */
473 static inline void fsnotify_clear_sb_marks_by_group(struct fsnotify_group *group)
475 fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_SB_FL);
477 extern void fsnotify_get_mark(struct fsnotify_mark *mark);
478 extern void fsnotify_put_mark(struct fsnotify_mark *mark);
479 extern void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info);
480 extern bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info);
482 /* put here because inotify does some weird stuff when destroying watches */
483 extern void fsnotify_init_event(struct fsnotify_event *event,
484 struct inode *to_tell, u32 mask);
488 static inline int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
489 const unsigned char *name, u32 cookie)
494 static inline int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask)
499 static inline void __fsnotify_inode_delete(struct inode *inode)
502 static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
505 static inline void fsnotify_sb_delete(struct super_block *sb)
508 static inline void fsnotify_update_flags(struct dentry *dentry)
511 static inline u32 fsnotify_get_cookie(void)
516 static inline void fsnotify_unmount_inodes(struct super_block *sb)
519 #endif /* CONFIG_FSNOTIFY */
521 #endif /* __KERNEL __ */
523 #endif /* __LINUX_FSNOTIFY_BACKEND_H */