1 /*
2  * Filesystem access notification for Linux
3  *
4  *  Copyright (C) 2008 Red Hat, Inc., Eric Paris <[email protected]>
5  */
6 
7 #ifndef __LINUX_FSNOTIFY_BACKEND_H
8 #define __LINUX_FSNOTIFY_BACKEND_H
9 
10 #ifdef __KERNEL__
11 
12 #include <linux/idr.h> /* inotify uses this */
13 #include <linux/fs.h> /* struct inode */
14 #include <linux/list.h>
15 #include <linux/path.h> /* struct path */
16 #include <linux/spinlock.h>
17 #include <linux/types.h>
18 #include <linux/atomic.h>
19 #include <linux/user_namespace.h>
20 
21 /*
22  * IN_* from inotfy.h lines up EXACTLY with FS_*, this is so we can easily
23  * convert between them.  dnotify only needs conversion at watch creation
24  * so no perf loss there.  fanotify isn't defined yet, so it can use the
25  * wholes if it needs more events.
26  */
27 #define FS_ACCESS		0x00000001	/* File was accessed */
28 #define FS_MODIFY		0x00000002	/* File was modified */
29 #define FS_ATTRIB		0x00000004	/* Metadata changed */
30 #define FS_CLOSE_WRITE		0x00000008	/* Writtable file was closed */
31 #define FS_CLOSE_NOWRITE	0x00000010	/* Unwrittable file closed */
32 #define FS_OPEN			0x00000020	/* File was opened */
33 #define FS_MOVED_FROM		0x00000040	/* File was moved from X */
34 #define FS_MOVED_TO		0x00000080	/* File was moved to Y */
35 #define FS_CREATE		0x00000100	/* Subfile was created */
36 #define FS_DELETE		0x00000200	/* Subfile was deleted */
37 #define FS_DELETE_SELF		0x00000400	/* Self was deleted */
38 #define FS_MOVE_SELF		0x00000800	/* Self was moved */
39 
40 #define FS_UNMOUNT		0x00002000	/* inode on umount fs */
41 #define FS_Q_OVERFLOW		0x00004000	/* Event queued overflowed */
42 #define FS_IN_IGNORED		0x00008000	/* last inotify event here */
43 
44 #define FS_OPEN_PERM		0x00010000	/* open event in an permission hook */
45 #define FS_ACCESS_PERM		0x00020000	/* access event in a permissions hook */
46 
47 #define FS_EXCL_UNLINK		0x04000000	/* do not send events if object is unlinked */
48 #define FS_ISDIR		0x40000000	/* event occurred against dir */
49 #define FS_IN_ONESHOT		0x80000000	/* only send event once */
50 
51 #define FS_DN_RENAME		0x10000000	/* file renamed */
52 #define FS_DN_MULTISHOT		0x20000000	/* dnotify multishot */
53 
54 /* This inode cares about things that happen to its children.  Always set for
55  * dnotify and inotify. */
56 #define FS_EVENT_ON_CHILD	0x08000000
57 
58 /* This is a list of all events that may get sent to a parernt based on fs event
59  * happening to inodes inside that directory */
60 #define FS_EVENTS_POSS_ON_CHILD   (FS_ACCESS | FS_MODIFY | FS_ATTRIB |\
61 				   FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN |\
62 				   FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE |\
63 				   FS_DELETE | FS_OPEN_PERM | FS_ACCESS_PERM)
64 
65 #define FS_MOVE			(FS_MOVED_FROM | FS_MOVED_TO)
66 
67 #define ALL_FSNOTIFY_PERM_EVENTS (FS_OPEN_PERM | FS_ACCESS_PERM)
68 
69 #define ALL_FSNOTIFY_EVENTS (FS_ACCESS | FS_MODIFY | FS_ATTRIB | \
70 			     FS_CLOSE_WRITE | FS_CLOSE_NOWRITE | FS_OPEN | \
71 			     FS_MOVED_FROM | FS_MOVED_TO | FS_CREATE | \
72 			     FS_DELETE | FS_DELETE_SELF | FS_MOVE_SELF | \
73 			     FS_UNMOUNT | FS_Q_OVERFLOW | FS_IN_IGNORED | \
74 			     FS_OPEN_PERM | FS_ACCESS_PERM | FS_EXCL_UNLINK | \
75 			     FS_ISDIR | FS_IN_ONESHOT | FS_DN_RENAME | \
76 			     FS_DN_MULTISHOT | FS_EVENT_ON_CHILD)
77 
78 struct fsnotify_group;
79 struct fsnotify_event;
80 struct fsnotify_mark;
81 struct fsnotify_event_private_data;
82 struct fsnotify_fname;
83 
84 /*
85  * Each group much define these ops.  The fsnotify infrastructure will call
86  * these operations for each relevant group.
87  *
88  * handle_event - main call for a group to handle an fs event
89  * free_group_priv - called when a group refcnt hits 0 to clean up the private union
90  * freeing_mark - called when a mark is being destroyed for some reason.  The group
91  * 		MUST be holding a reference on each mark and that reference must be
92  * 		dropped in this function.  inotify uses this function to send
93  * 		userspace messages that marks have been removed.
94  */
95 struct fsnotify_ops {
96 	int (*handle_event)(struct fsnotify_group *group,
97 			    struct inode *inode,
98 			    struct fsnotify_mark *inode_mark,
99 			    struct fsnotify_mark *vfsmount_mark,
100 			    u32 mask, const void *data, int data_type,
101 			    const unsigned char *file_name, u32 cookie);
102 	void (*free_group_priv)(struct fsnotify_group *group);
103 	void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group);
104 	void (*free_event)(struct fsnotify_event *event);
105 };
106 
107 /*
108  * all of the information about the original object we want to now send to
109  * a group.  If you want to carry more info from the accessing task to the
110  * listener this structure is where you need to be adding fields.
111  */
112 struct fsnotify_event {
113 	struct list_head list;
114 	/* inode may ONLY be dereferenced during handle_event(). */
115 	struct inode *inode;	/* either the inode the event happened to or its parent */
116 	u32 mask;		/* the type of access, bitwise OR for FS_* event types */
117 };
118 
119 /*
120  * A group is a "thing" that wants to receive notification about filesystem
121  * events.  The mask holds the subset of event types this group cares about.
122  * refcnt on a group is up to the implementor and at any moment if it goes 0
123  * everything will be cleaned up.
124  */
125 struct fsnotify_group {
126 	/*
127 	 * How the refcnt is used is up to each group.  When the refcnt hits 0
128 	 * fsnotify will clean up all of the resources associated with this group.
129 	 * As an example, the dnotify group will always have a refcnt=1 and that
130 	 * will never change.  Inotify, on the other hand, has a group per
131 	 * inotify_init() and the refcnt will hit 0 only when that fd has been
132 	 * closed.
133 	 */
134 	atomic_t refcnt;		/* things with interest in this group */
135 
136 	const struct fsnotify_ops *ops;	/* how this group handles things */
137 
138 	/* needed to send notification to userspace */
139 	spinlock_t notification_lock;		/* protect the notification_list */
140 	struct list_head notification_list;	/* list of event_holder this group needs to send to userspace */
141 	wait_queue_head_t notification_waitq;	/* read() on the notification file blocks on this waitq */
142 	unsigned int q_len;			/* events on the queue */
143 	unsigned int max_events;		/* maximum events allowed on the list */
144 	/*
145 	 * Valid fsnotify group priorities.  Events are send in order from highest
146 	 * priority to lowest priority.  We default to the lowest priority.
147 	 */
148 	#define FS_PRIO_0	0 /* normal notifiers, no permissions */
149 	#define FS_PRIO_1	1 /* fanotify content based access control */
150 	#define FS_PRIO_2	2 /* fanotify pre-content access */
151 	unsigned int priority;
152 	bool shutdown;		/* group is being shut down, don't queue more events */
153 
154 	/* stores all fastpath marks assoc with this group so they can be cleaned on unregister */
155 	struct mutex mark_mutex;	/* protect marks_list */
156 	atomic_t num_marks;		/* 1 for each mark and 1 for not being
157 					 * past the point of no return when freeing
158 					 * a group */
159 	struct list_head marks_list;	/* all inode marks for this group */
160 
161 	struct fasync_struct *fsn_fa;    /* async notification */
162 
163 	struct fsnotify_event *overflow_event;	/* Event we queue when the
164 						 * notification list is too
165 						 * full */
166 
167 	/* groups can define private fields here or use the void *private */
168 	union {
169 		void *private;
170 #ifdef CONFIG_INOTIFY_USER
171 		struct inotify_group_private_data {
172 			spinlock_t	idr_lock;
173 			struct idr      idr;
174 			struct ucounts *ucounts;
175 		} inotify_data;
176 #endif
177 #ifdef CONFIG_FANOTIFY
178 		struct fanotify_group_private_data {
179 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
180 			/* allows a group to block waiting for a userspace response */
181 			struct list_head access_list;
182 			wait_queue_head_t access_waitq;
183 #endif /* CONFIG_FANOTIFY_ACCESS_PERMISSIONS */
184 			int f_flags;
185 			unsigned int max_marks;
186 			struct user_struct *user;
187 		} fanotify_data;
188 #endif /* CONFIG_FANOTIFY */
189 	};
190 };
191 
192 /* when calling fsnotify tell it if the data is a path or inode */
193 #define FSNOTIFY_EVENT_NONE	0
194 #define FSNOTIFY_EVENT_PATH	1
195 #define FSNOTIFY_EVENT_INODE	2
196 
197 /*
198  * A mark is simply an object attached to an in core inode which allows an
199  * fsnotify listener to indicate they are either no longer interested in events
200  * of a type matching mask or only interested in those events.
201  *
202  * These are flushed when an inode is evicted from core and may be flushed
203  * when the inode is modified (as seen by fsnotify_access).  Some fsnotify
204  * users (such as dnotify) will flush these when the open fd is closed and not
205  * at inode eviction or modification.
206  *
207  * Text in brackets is showing the lock(s) protecting modifications of a
208  * particular entry. obj_lock means either inode->i_lock or
209  * mnt->mnt_root->d_lock depending on the mark type.
210  */
211 struct fsnotify_mark {
212 	/* Mask this mark is for [mark->lock, group->mark_mutex] */
213 	__u32 mask;
214 	/* We hold one for presence in g_list. Also one ref for each 'thing'
215 	 * in kernel that found and may be using this mark. */
216 	atomic_t refcnt;
217 	/* Group this mark is for. Set on mark creation, stable until last ref
218 	 * is dropped */
219 	struct fsnotify_group *group;
220 	/* List of marks by group->i_fsnotify_marks. Also reused for queueing
221 	 * mark into destroy_list when it's waiting for the end of SRCU period
222 	 * before it can be freed. [group->mark_mutex] */
223 	struct list_head g_list;
224 	/* Protects inode / mnt pointers, flags, masks */
225 	spinlock_t lock;
226 	/* List of marks for inode / vfsmount [obj_lock] */
227 	struct hlist_node obj_list;
228 	union {	/* Object pointer [mark->lock, group->mark_mutex] */
229 		struct inode *inode;	/* inode this mark is associated with */
230 		struct vfsmount *mnt;	/* vfsmount this mark is associated with */
231 	};
232 	/* Events types to ignore [mark->lock, group->mark_mutex] */
233 	__u32 ignored_mask;
234 #define FSNOTIFY_MARK_FLAG_INODE		0x01
235 #define FSNOTIFY_MARK_FLAG_VFSMOUNT		0x02
236 #define FSNOTIFY_MARK_FLAG_OBJECT_PINNED	0x04
237 #define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY	0x08
238 #define FSNOTIFY_MARK_FLAG_ALIVE		0x10
239 #define FSNOTIFY_MARK_FLAG_ATTACHED		0x20
240 	unsigned int flags;		/* flags [mark->lock] */
241 	void (*free_mark)(struct fsnotify_mark *mark); /* called on final put+free */
242 };
243 
244 #ifdef CONFIG_FSNOTIFY
245 
246 /* called from the vfs helpers */
247 
248 /* main fsnotify call to send events */
249 extern int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
250 		    const unsigned char *name, u32 cookie);
251 extern int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask);
252 extern void __fsnotify_inode_delete(struct inode *inode);
253 extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
254 extern u32 fsnotify_get_cookie(void);
255 
256 static inline int fsnotify_inode_watches_children(struct inode *inode)
257 {
258 	/* FS_EVENT_ON_CHILD is set if the inode may care */
259 	if (!(inode->i_fsnotify_mask & FS_EVENT_ON_CHILD))
260 		return 0;
261 	/* this inode might care about child events, does it care about the
262 	 * specific set of events that can happen on a child? */
263 	return inode->i_fsnotify_mask & FS_EVENTS_POSS_ON_CHILD;
264 }
265 
266 /*
267  * Update the dentry with a flag indicating the interest of its parent to receive
268  * filesystem events when those events happens to this dentry->d_inode.
269  */
270 static inline void fsnotify_update_flags(struct dentry *dentry)
271 {
272 	assert_spin_locked(&dentry->d_lock);
273 
274 	/*
275 	 * Serialisation of setting PARENT_WATCHED on the dentries is provided
276 	 * by d_lock. If inotify_inode_watched changes after we have taken
277 	 * d_lock, the following __fsnotify_update_child_dentry_flags call will
278 	 * find our entry, so it will spin until we complete here, and update
279 	 * us with the new state.
280 	 */
281 	if (fsnotify_inode_watches_children(dentry->d_parent->d_inode))
282 		dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
283 	else
284 		dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
285 }
286 
287 /* called from fsnotify listeners, such as fanotify or dnotify */
288 
289 /* create a new group */
290 extern struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops);
291 /* get reference to a group */
292 extern void fsnotify_get_group(struct fsnotify_group *group);
293 /* drop reference on a group from fsnotify_alloc_group */
294 extern void fsnotify_put_group(struct fsnotify_group *group);
295 /* group destruction begins, stop queuing new events */
296 extern void fsnotify_group_stop_queueing(struct fsnotify_group *group);
297 /* destroy group */
298 extern void fsnotify_destroy_group(struct fsnotify_group *group);
299 /* fasync handler function */
300 extern int fsnotify_fasync(int fd, struct file *file, int on);
301 /* Free event from memory */
302 extern void fsnotify_destroy_event(struct fsnotify_group *group,
303 				   struct fsnotify_event *event);
304 /* attach the event to the group notification queue */
305 extern int fsnotify_add_event(struct fsnotify_group *group,
306 			      struct fsnotify_event *event,
307 			      int (*merge)(struct list_head *,
308 					   struct fsnotify_event *));
309 /* true if the group notification queue is empty */
310 extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group);
311 /* return, but do not dequeue the first event on the notification queue */
312 extern struct fsnotify_event *fsnotify_peek_first_event(struct fsnotify_group *group);
313 /* return AND dequeue the first event on the notification queue */
314 extern struct fsnotify_event *fsnotify_remove_first_event(struct fsnotify_group *group);
315 
316 /* functions used to manipulate the marks attached to inodes */
317 
318 /* run all marks associated with a vfsmount and update mnt->mnt_fsnotify_mask */
319 extern void fsnotify_recalc_vfsmount_mask(struct vfsmount *mnt);
320 /* run all marks associated with an inode and update inode->i_fsnotify_mask */
321 extern void fsnotify_recalc_inode_mask(struct inode *inode);
322 extern void fsnotify_init_mark(struct fsnotify_mark *mark, void (*free_mark)(struct fsnotify_mark *mark));
323 /* find (and take a reference) to a mark associated with group and inode */
324 extern struct fsnotify_mark *fsnotify_find_inode_mark(struct fsnotify_group *group, struct inode *inode);
325 /* find (and take a reference) to a mark associated with group and vfsmount */
326 extern struct fsnotify_mark *fsnotify_find_vfsmount_mark(struct fsnotify_group *group, struct vfsmount *mnt);
327 /* set the ignored_mask of a mark */
328 extern void fsnotify_set_mark_ignored_mask_locked(struct fsnotify_mark *mark, __u32 mask);
329 /* set the mask of a mark (might pin the object into memory */
330 extern void fsnotify_set_mark_mask_locked(struct fsnotify_mark *mark, __u32 mask);
331 /* attach the mark to both the group and the inode */
332 extern int fsnotify_add_mark(struct fsnotify_mark *mark, struct fsnotify_group *group,
333 			     struct inode *inode, struct vfsmount *mnt, int allow_dups);
334 extern int fsnotify_add_mark_locked(struct fsnotify_mark *mark, struct fsnotify_group *group,
335 				    struct inode *inode, struct vfsmount *mnt, int allow_dups);
336 /* given a group and a mark, flag mark to be freed when all references are dropped */
337 extern void fsnotify_destroy_mark(struct fsnotify_mark *mark,
338 				  struct fsnotify_group *group);
339 /* detach mark from inode / mount list, group list, drop inode reference */
340 extern void fsnotify_detach_mark(struct fsnotify_mark *mark);
341 /* free mark */
342 extern void fsnotify_free_mark(struct fsnotify_mark *mark);
343 /* run all the marks in a group, and clear all of the vfsmount marks */
344 extern void fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group *group);
345 /* run all the marks in a group, and clear all of the inode marks */
346 extern void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group);
347 /* run all the marks in a group, and clear all of the marks where mark->flags & flags is true*/
348 extern void fsnotify_clear_marks_by_group_flags(struct fsnotify_group *group, unsigned int flags);
349 extern void fsnotify_get_mark(struct fsnotify_mark *mark);
350 extern void fsnotify_put_mark(struct fsnotify_mark *mark);
351 extern void fsnotify_unmount_inodes(struct super_block *sb);
352 
353 /* put here because inotify does some weird stuff when destroying watches */
354 extern void fsnotify_init_event(struct fsnotify_event *event,
355 				struct inode *to_tell, u32 mask);
356 
357 #else
358 
359 static inline int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
360 			   const unsigned char *name, u32 cookie)
361 {
362 	return 0;
363 }
364 
365 static inline int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask)
366 {
367 	return 0;
368 }
369 
370 static inline void __fsnotify_inode_delete(struct inode *inode)
371 {}
372 
373 static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
374 {}
375 
376 static inline void fsnotify_update_flags(struct dentry *dentry)
377 {}
378 
379 static inline u32 fsnotify_get_cookie(void)
380 {
381 	return 0;
382 }
383 
384 static inline void fsnotify_unmount_inodes(struct super_block *sb)
385 {}
386 
387 #endif	/* CONFIG_FSNOTIFY */
388 
389 #endif	/* __KERNEL __ */
390 
391 #endif	/* __LINUX_FSNOTIFY_BACKEND_H */
392