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 struct fsnotify_iter_info;
84 
85 /*
86  * Each group much define these ops.  The fsnotify infrastructure will call
87  * these operations for each relevant group.
88  *
89  * handle_event - main call for a group to handle an fs event
90  * free_group_priv - called when a group refcnt hits 0 to clean up the private union
91  * freeing_mark - called when a mark is being destroyed for some reason.  The group
92  * 		MUST be holding a reference on each mark and that reference must be
93  * 		dropped in this function.  inotify uses this function to send
94  * 		userspace messages that marks have been removed.
95  */
96 struct fsnotify_ops {
97 	int (*handle_event)(struct fsnotify_group *group,
98 			    struct inode *inode,
99 			    struct fsnotify_mark *inode_mark,
100 			    struct fsnotify_mark *vfsmount_mark,
101 			    u32 mask, const void *data, int data_type,
102 			    const unsigned char *file_name, u32 cookie,
103 			    struct fsnotify_iter_info *iter_info);
104 	void (*free_group_priv)(struct fsnotify_group *group);
105 	void (*freeing_mark)(struct fsnotify_mark *mark, struct fsnotify_group *group);
106 	void (*free_event)(struct fsnotify_event *event);
107 	/* called on final put+free to free memory */
108 	void (*free_mark)(struct fsnotify_mark *mark);
109 };
110 
111 /*
112  * all of the information about the original object we want to now send to
113  * a group.  If you want to carry more info from the accessing task to the
114  * listener this structure is where you need to be adding fields.
115  */
116 struct fsnotify_event {
117 	struct list_head list;
118 	/* inode may ONLY be dereferenced during handle_event(). */
119 	struct inode *inode;	/* either the inode the event happened to or its parent */
120 	u32 mask;		/* the type of access, bitwise OR for FS_* event types */
121 };
122 
123 /*
124  * A group is a "thing" that wants to receive notification about filesystem
125  * events.  The mask holds the subset of event types this group cares about.
126  * refcnt on a group is up to the implementor and at any moment if it goes 0
127  * everything will be cleaned up.
128  */
129 struct fsnotify_group {
130 	/*
131 	 * How the refcnt is used is up to each group.  When the refcnt hits 0
132 	 * fsnotify will clean up all of the resources associated with this group.
133 	 * As an example, the dnotify group will always have a refcnt=1 and that
134 	 * will never change.  Inotify, on the other hand, has a group per
135 	 * inotify_init() and the refcnt will hit 0 only when that fd has been
136 	 * closed.
137 	 */
138 	atomic_t refcnt;		/* things with interest in this group */
139 
140 	const struct fsnotify_ops *ops;	/* how this group handles things */
141 
142 	/* needed to send notification to userspace */
143 	spinlock_t notification_lock;		/* protect the notification_list */
144 	struct list_head notification_list;	/* list of event_holder this group needs to send to userspace */
145 	wait_queue_head_t notification_waitq;	/* read() on the notification file blocks on this waitq */
146 	unsigned int q_len;			/* events on the queue */
147 	unsigned int max_events;		/* maximum events allowed on the list */
148 	/*
149 	 * Valid fsnotify group priorities.  Events are send in order from highest
150 	 * priority to lowest priority.  We default to the lowest priority.
151 	 */
152 	#define FS_PRIO_0	0 /* normal notifiers, no permissions */
153 	#define FS_PRIO_1	1 /* fanotify content based access control */
154 	#define FS_PRIO_2	2 /* fanotify pre-content access */
155 	unsigned int priority;
156 	bool shutdown;		/* group is being shut down, don't queue more events */
157 
158 	/* stores all fastpath marks assoc with this group so they can be cleaned on unregister */
159 	struct mutex mark_mutex;	/* protect marks_list */
160 	atomic_t num_marks;		/* 1 for each mark and 1 for not being
161 					 * past the point of no return when freeing
162 					 * a group */
163 	struct list_head marks_list;	/* all inode marks for this group */
164 
165 	struct fasync_struct *fsn_fa;    /* async notification */
166 
167 	struct fsnotify_event *overflow_event;	/* Event we queue when the
168 						 * notification list is too
169 						 * full */
170 	atomic_t user_waits;		/* Number of tasks waiting for user
171 					 * response */
172 
173 	/* groups can define private fields here or use the void *private */
174 	union {
175 		void *private;
176 #ifdef CONFIG_INOTIFY_USER
177 		struct inotify_group_private_data {
178 			spinlock_t	idr_lock;
179 			struct idr      idr;
180 			struct ucounts *ucounts;
181 		} inotify_data;
182 #endif
183 #ifdef CONFIG_FANOTIFY
184 		struct fanotify_group_private_data {
185 #ifdef CONFIG_FANOTIFY_ACCESS_PERMISSIONS
186 			/* allows a group to block waiting for a userspace response */
187 			struct list_head access_list;
188 			wait_queue_head_t access_waitq;
189 #endif /* CONFIG_FANOTIFY_ACCESS_PERMISSIONS */
190 			int f_flags;
191 			unsigned int max_marks;
192 			struct user_struct *user;
193 		} fanotify_data;
194 #endif /* CONFIG_FANOTIFY */
195 	};
196 };
197 
198 /* when calling fsnotify tell it if the data is a path or inode */
199 #define FSNOTIFY_EVENT_NONE	0
200 #define FSNOTIFY_EVENT_PATH	1
201 #define FSNOTIFY_EVENT_INODE	2
202 
203 /*
204  * Inode / vfsmount point to this structure which tracks all marks attached to
205  * the inode / vfsmount. The reference to inode / vfsmount is held by this
206  * structure. We destroy this structure when there are no more marks attached
207  * to it. The structure is protected by fsnotify_mark_srcu.
208  */
209 struct fsnotify_mark_connector {
210 	spinlock_t lock;
211 #define FSNOTIFY_OBJ_TYPE_INODE		0x01
212 #define FSNOTIFY_OBJ_TYPE_VFSMOUNT	0x02
213 #define FSNOTIFY_OBJ_ALL_TYPES		(FSNOTIFY_OBJ_TYPE_INODE | \
214 					 FSNOTIFY_OBJ_TYPE_VFSMOUNT)
215 	unsigned int flags;	/* Type of object [lock] */
216 	union {	/* Object pointer [lock] */
217 		struct inode *inode;
218 		struct vfsmount *mnt;
219 	};
220 	union {
221 		struct hlist_head list;
222 		/* Used listing heads to free after srcu period expires */
223 		struct fsnotify_mark_connector *destroy_next;
224 	};
225 };
226 
227 /*
228  * A mark is simply an object attached to an in core inode which allows an
229  * fsnotify listener to indicate they are either no longer interested in events
230  * of a type matching mask or only interested in those events.
231  *
232  * These are flushed when an inode is evicted from core and may be flushed
233  * when the inode is modified (as seen by fsnotify_access).  Some fsnotify
234  * users (such as dnotify) will flush these when the open fd is closed and not
235  * at inode eviction or modification.
236  *
237  * Text in brackets is showing the lock(s) protecting modifications of a
238  * particular entry. obj_lock means either inode->i_lock or
239  * mnt->mnt_root->d_lock depending on the mark type.
240  */
241 struct fsnotify_mark {
242 	/* Mask this mark is for [mark->lock, group->mark_mutex] */
243 	__u32 mask;
244 	/* We hold one for presence in g_list. Also one ref for each 'thing'
245 	 * in kernel that found and may be using this mark. */
246 	atomic_t refcnt;
247 	/* Group this mark is for. Set on mark creation, stable until last ref
248 	 * is dropped */
249 	struct fsnotify_group *group;
250 	/* List of marks by group->i_fsnotify_marks. Also reused for queueing
251 	 * mark into destroy_list when it's waiting for the end of SRCU period
252 	 * before it can be freed. [group->mark_mutex] */
253 	struct list_head g_list;
254 	/* Protects inode / mnt pointers, flags, masks */
255 	spinlock_t lock;
256 	/* List of marks for inode / vfsmount [connector->lock, mark ref] */
257 	struct hlist_node obj_list;
258 	/* Head of list of marks for an object [mark ref] */
259 	struct fsnotify_mark_connector *connector;
260 	/* Events types to ignore [mark->lock, group->mark_mutex] */
261 	__u32 ignored_mask;
262 #define FSNOTIFY_MARK_FLAG_IGNORED_SURV_MODIFY	0x01
263 #define FSNOTIFY_MARK_FLAG_ALIVE		0x02
264 #define FSNOTIFY_MARK_FLAG_ATTACHED		0x04
265 	unsigned int flags;		/* flags [mark->lock] */
266 };
267 
268 #ifdef CONFIG_FSNOTIFY
269 
270 /* called from the vfs helpers */
271 
272 /* main fsnotify call to send events */
273 extern int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
274 		    const unsigned char *name, u32 cookie);
275 extern int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask);
276 extern void __fsnotify_inode_delete(struct inode *inode);
277 extern void __fsnotify_vfsmount_delete(struct vfsmount *mnt);
278 extern u32 fsnotify_get_cookie(void);
279 
280 static inline int fsnotify_inode_watches_children(struct inode *inode)
281 {
282 	/* FS_EVENT_ON_CHILD is set if the inode may care */
283 	if (!(inode->i_fsnotify_mask & FS_EVENT_ON_CHILD))
284 		return 0;
285 	/* this inode might care about child events, does it care about the
286 	 * specific set of events that can happen on a child? */
287 	return inode->i_fsnotify_mask & FS_EVENTS_POSS_ON_CHILD;
288 }
289 
290 /*
291  * Update the dentry with a flag indicating the interest of its parent to receive
292  * filesystem events when those events happens to this dentry->d_inode.
293  */
294 static inline void fsnotify_update_flags(struct dentry *dentry)
295 {
296 	assert_spin_locked(&dentry->d_lock);
297 
298 	/*
299 	 * Serialisation of setting PARENT_WATCHED on the dentries is provided
300 	 * by d_lock. If inotify_inode_watched changes after we have taken
301 	 * d_lock, the following __fsnotify_update_child_dentry_flags call will
302 	 * find our entry, so it will spin until we complete here, and update
303 	 * us with the new state.
304 	 */
305 	if (fsnotify_inode_watches_children(dentry->d_parent->d_inode))
306 		dentry->d_flags |= DCACHE_FSNOTIFY_PARENT_WATCHED;
307 	else
308 		dentry->d_flags &= ~DCACHE_FSNOTIFY_PARENT_WATCHED;
309 }
310 
311 /* called from fsnotify listeners, such as fanotify or dnotify */
312 
313 /* create a new group */
314 extern struct fsnotify_group *fsnotify_alloc_group(const struct fsnotify_ops *ops);
315 /* get reference to a group */
316 extern void fsnotify_get_group(struct fsnotify_group *group);
317 /* drop reference on a group from fsnotify_alloc_group */
318 extern void fsnotify_put_group(struct fsnotify_group *group);
319 /* group destruction begins, stop queuing new events */
320 extern void fsnotify_group_stop_queueing(struct fsnotify_group *group);
321 /* destroy group */
322 extern void fsnotify_destroy_group(struct fsnotify_group *group);
323 /* fasync handler function */
324 extern int fsnotify_fasync(int fd, struct file *file, int on);
325 /* Free event from memory */
326 extern void fsnotify_destroy_event(struct fsnotify_group *group,
327 				   struct fsnotify_event *event);
328 /* attach the event to the group notification queue */
329 extern int fsnotify_add_event(struct fsnotify_group *group,
330 			      struct fsnotify_event *event,
331 			      int (*merge)(struct list_head *,
332 					   struct fsnotify_event *));
333 /* true if the group notification queue is empty */
334 extern bool fsnotify_notify_queue_is_empty(struct fsnotify_group *group);
335 /* return, but do not dequeue the first event on the notification queue */
336 extern struct fsnotify_event *fsnotify_peek_first_event(struct fsnotify_group *group);
337 /* return AND dequeue the first event on the notification queue */
338 extern struct fsnotify_event *fsnotify_remove_first_event(struct fsnotify_group *group);
339 
340 /* functions used to manipulate the marks attached to inodes */
341 
342 /* Calculate mask of events for a list of marks */
343 extern void fsnotify_recalc_mask(struct fsnotify_mark_connector *conn);
344 extern void fsnotify_init_mark(struct fsnotify_mark *mark,
345 			       struct fsnotify_group *group);
346 /* Find mark belonging to given group in the list of marks */
347 extern struct fsnotify_mark *fsnotify_find_mark(
348 				struct fsnotify_mark_connector __rcu **connp,
349 				struct fsnotify_group *group);
350 /* attach the mark to the inode or vfsmount */
351 extern int fsnotify_add_mark(struct fsnotify_mark *mark, struct inode *inode,
352 			     struct vfsmount *mnt, int allow_dups);
353 extern int fsnotify_add_mark_locked(struct fsnotify_mark *mark,
354 				    struct inode *inode, struct vfsmount *mnt, int allow_dups);
355 /* given a group and a mark, flag mark to be freed when all references are dropped */
356 extern void fsnotify_destroy_mark(struct fsnotify_mark *mark,
357 				  struct fsnotify_group *group);
358 /* detach mark from inode / mount list, group list, drop inode reference */
359 extern void fsnotify_detach_mark(struct fsnotify_mark *mark);
360 /* free mark */
361 extern void fsnotify_free_mark(struct fsnotify_mark *mark);
362 /* run all the marks in a group, and clear all of the marks attached to given object type */
363 extern void fsnotify_clear_marks_by_group(struct fsnotify_group *group, unsigned int type);
364 /* run all the marks in a group, and clear all of the vfsmount marks */
365 static inline void fsnotify_clear_vfsmount_marks_by_group(struct fsnotify_group *group)
366 {
367 	fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_VFSMOUNT);
368 }
369 /* run all the marks in a group, and clear all of the inode marks */
370 static inline void fsnotify_clear_inode_marks_by_group(struct fsnotify_group *group)
371 {
372 	fsnotify_clear_marks_by_group(group, FSNOTIFY_OBJ_TYPE_INODE);
373 }
374 extern void fsnotify_get_mark(struct fsnotify_mark *mark);
375 extern void fsnotify_put_mark(struct fsnotify_mark *mark);
376 extern void fsnotify_unmount_inodes(struct super_block *sb);
377 extern void fsnotify_finish_user_wait(struct fsnotify_iter_info *iter_info);
378 extern bool fsnotify_prepare_user_wait(struct fsnotify_iter_info *iter_info);
379 
380 /* put here because inotify does some weird stuff when destroying watches */
381 extern void fsnotify_init_event(struct fsnotify_event *event,
382 				struct inode *to_tell, u32 mask);
383 
384 #else
385 
386 static inline int fsnotify(struct inode *to_tell, __u32 mask, const void *data, int data_is,
387 			   const unsigned char *name, u32 cookie)
388 {
389 	return 0;
390 }
391 
392 static inline int __fsnotify_parent(const struct path *path, struct dentry *dentry, __u32 mask)
393 {
394 	return 0;
395 }
396 
397 static inline void __fsnotify_inode_delete(struct inode *inode)
398 {}
399 
400 static inline void __fsnotify_vfsmount_delete(struct vfsmount *mnt)
401 {}
402 
403 static inline void fsnotify_update_flags(struct dentry *dentry)
404 {}
405 
406 static inline u32 fsnotify_get_cookie(void)
407 {
408 	return 0;
409 }
410 
411 static inline void fsnotify_unmount_inodes(struct super_block *sb)
412 {}
413 
414 #endif	/* CONFIG_FSNOTIFY */
415 
416 #endif	/* __KERNEL __ */
417 
418 #endif	/* __LINUX_FSNOTIFY_BACKEND_H */
419